Matlab Climate & Health Review

Systematic review · PRISMA 2020

Climate, weather and air pollution research using the Matlab HDSS, Bangladesh: a systematic review of what has been done so far

Literature review compiled 16 October 2026 · Data source under review: Matlab Health and Demographic Surveillance System (icddr,b), Chandpur District, Bangladesh

183records screened
47studies included
1981–2024publication years
8exposure–outcome themes

Abstract

Background. The Matlab HDSS has followed a rural population of about 230,000 people continuously since 1966 and is the longest-running surveillance site in the global South. Its daily records of births, deaths, migrations and hospital-confirmed infections, combined with nearby meteorological stations, make it one of the few places in a low-income country where climate–health relationships can be examined over decades. This review maps the peer-reviewed literature that has used Matlab data to study meteorological, climatic, hydrological or air-quality exposures and health.

Methods. PubMed was searched on 16 October 2026 for articles indexing “Matlab” and “Bangladesh” in the title or abstract combined with climate, temperature, weather, rainfall, flood and air-pollution terms, and Scopus and the Web of Science Core Collection were searched on the same date with a proximity string tying “Matlab” to the site (Bangladesh, Chandpur, icddr,b, HDSS or surveillance) against a twenty-term exposure block. Records were screened on title and abstract against pre-stated eligibility criteria; data were extracted on design, period, exposure, outcome and principal findings; and studies were synthesised narratively by theme. The reference lists of 40 included studies and a topic review were then screened as a backward citation search, and Google Scholar was searched once per exposure–outcome theme. Reporting follows PRISMA 2020. No protocol was registered and no meta-analysis was attempted.

Results. Of 280 records across the three databases (183 screened), 39 studies met the criteria, and backward citation searching and a themed Google Scholar pass added eight more, giving 47 studies published between 1981 and 2024. Cholera dominates (14 studies), followed by flooding, flood control and salinity (7), seasonality of mortality (6), household air pollution (6), heat and season in pregnancy with climate-linked migration (5), temperature–mortality time series (3), weather and childhood pneumonia (3) and childhood diarrhoea and rotavirus (3). Findings are largely consistent in direction: cold, not heat, drives excess mortality in adults and infants; heat and warm sea-surface conditions raise the incidence of cholera, diarrhoea and pneumonia; flood protection improved child survival but altered cholera dynamics; household solid-fuel smoke roughly doubles adult respiratory mortality. The clearest dissent is the earliest cholera analysis, which found no constant relationship between the timing of the annual epidemic and monsoon rains or river levels. No study has yet linked ambient air pollution, or combined heat and pollution exposure, to health outcomes in Matlab, and none has used climate projections — although measured personal PM2.5, black carbon and carbon monoxide have now been set against vascular function in Matlab women, with a null result. Google Scholar, searched across all eight themes, added five studies; its probes for heat against cardiovascular, renal and mental-health outcomes returned nothing, which is itself evidence that those gaps are real.

Conclusions. Fifty years of Matlab data have been used mainly to explain infectious-disease seasonality, above all cholera. Temperature–mortality work stopped in 2012. Household air pollution has now been examined six times, including once with measured personal exposure against vascular function; ambient pollution and heat–pollution interactions remain untouched as health questions. These are the clearest openings for new analyses on this dataset.

At a glance

Forty-seven studies, 1981–2024, across eight exposure–outcome themes. Every section below this panel can be opened for the detail behind it.

Climate variability and cholera14
Flooding, flood control and salinity7
Household air pollution6
Seasonality of mortality6
Heat and season in pregnancy, and climate-linked migration5
Temperature, weather and mortality3
Weather and childhood diarrhoea and rotavirus3
Weather and childhood pneumonia3

Established

  • Cold, not heat, drives excess mortality — three independent time series agree
  • Heat and warm seas raise cholera, childhood diarrhoea and childhood pneumonia
  • Cholera tracks ocean and climate indices at lags of weeks to months
  • Household biomass smoke is linked to obstructive lung disease
  • Flood protection improved child survival

Contested or null

  • Rainfall and cholera — the longest series found no stable link to the monsoon
  • Temperature and pneumonia — the three studies disagree about sign
  • Flood control and diarrhoea — a relationship its author could not explain
  • The 1988 and 2004 floods — little measurable excess once controlled
  • Measured household PM2.5 against vascular function — null

Not studied at all

  • Ambient air pollution — the row is empty against every outcome
  • Heat–pollution interaction — no joint model exists
  • Climate, water quality and diarrhoeal disease — the health outcome is missing
  • Temperature and mortality after 2009 — the series simply stop
  • Cardiovascular, renal and mental-health outcomes under heat stress

1Introduction

Bangladesh sits at the delta of the Ganges, Brahmaputra and Meghna rivers and is among the countries most exposed to climate change: monsoon flooding, cyclones, rising temperatures and salinity intrusion all bear on a dense, largely rural population. Quantifying how weather and climate affect health in such settings is hampered by the absence of reliable vital registration and long-run disease surveillance. The Matlab Health and Demographic Surveillance System, run by icddr,b in Chandpur District about 55 km south-east of Dhaka, is the principal exception. Established in 1966 after the cholera vaccine trials, it registers every birth, death, marriage and migration in 142 villages, assigns causes of death by verbal autopsy, geo-references households, and is linked to the Matlab hospital’s laboratory-confirmed diarrhoeal disease surveillance. A flood-control embankment completed in 1986 bisects the area, creating a natural experiment, and a weather station was installed on the hospital roof in 2008 to complement the Bangladesh Meteorological Department station at Chandpur.

icddr,b reports that Matlab data underpin more than 500 peer-reviewed papers; a PubMed search returns 667 records naming the site. Most concern fertility, child survival, vaccines and health-service interventions. This review asks a narrower question: what has been published, in peer-reviewed journals, that uses Matlab data to study the health effects of meteorological, climatic, hydrological or air-quality exposures, what has it found, and where are the gaps? The aim is to give researchers planning new analyses on the dataset a complete map of prior work.

2Methods

2.1 Protocol and reporting

The review follows the PRISMA 2020 statement for systematic reviews. No protocol was registered in PROSPERO or elsewhere, and the review was completed in a single day with AI-assisted searching and extraction checked by hand; both are limitations discussed in section 4.3. A condensed PRISMA checklist is given in Appendix B.

2.2 Eligibility criteria

Figure 1 states the criteria in full. A study was included only where it satisfied the inclusion rule in all four columns of panel A. Calendar season was accepted as a proxy exposure only under the two-part test in panel B, and the studies admitted on it are reported as a separate theme.

Methods at a glance: eligibility criteria, the season rule and the design grades Panel A gives the eligibility criteria as four columns — population and data source, exposure, outcome, and design and publication — each listing what was included and what was excluded. Panel B gives the two-part test deciding when calendar season counts as an exposure, with the four studies included and the three excluded on it. Panel C gives the three design-strength grades and the number of studies in each: A, 18; B, 24; C, 5. Population and data source INCLUDED Residents of the Matlab HDSS area Primary data source: Matlab HDSS, icddr,b Matlab hospital surveillance, or a nested cohort such as the MINIMat trial Multi-site studies where Matlab contributed original data EXCLUDED Multi-site work with no original Matlab data Exposure INCLUDED Meteorological: temperature, heatwave, rainfall, humidity, sunshine Climate index or remote proxy: ENSO, Indian Ocean Dipole, sea- surface temperature, chlorophyll Hydrological event or works: flood, cyclone, flood-control embankment Indoor or ambient air pollution EXCLUDED Season as an incidental remark rather than an analysed exposure — see panel B Outcome INCLUDED Any measure of human health: mortality, morbidity, infection incidence, hospital use, birth outcomes A health-related behaviour (migration) analysed as a response to environmental stress EXCLUDED Fertility or births as the only outcome Water chemistry or aquifer geology as the only outcome Design and publication INCLUDED Original quantitative analysis Peer-reviewed journal article Any year of publication English language EXCLUDED Reviews and topic overviews Site and cohort profiles Commentaries When season counts as an exposure Calendar season is an exposure only if both tests pass. 1 The authors read season as the climatic or agro-climatic cycle 2 The seasonal pattern is an object of the analysis, not an aside Both pass → included, reported as a separate theme Becker 1981 · Shaikh 1990 · Muhuri 1996 · Das 2014 Either fails → excluded Zaman 2009 · Tanaka 2007 · Bingnan 1991 — a seasonal peak noted in passing while studying disease burden, serotypes or vaccines Design grading A n = 18 Individual-level design, or a time-series model adjusting for season, trend and lagged exposure B n = 24 Ecological correlation, before–after comparison, or a model without such adjustment C n = 5 Descriptive or methods-development work Design strength only — a guide to how much weight a finding can carry, not a judgement of the paper. A B C

Figure 1. Methods at a glance. A The eligibility criteria; a study had to satisfy the inclusion rule in every column. B The two-part test that decides when calendar season counts as an analysed exposure rather than an incidental observation, with the studies included and excluded on it. C The design-strength grade carried by every row of Table 1, and the number of studies in each grade. Design grades describe design strength only.

2.3 Information sources and search

PubMed (MEDLINE), Scopus and the Web of Science Core Collection were searched on 16 October 2026, with no date or language limits. PubMed was searched as Matlab[Title/Abstract] AND Bangladesh[Title/Abstract] — 667 records — combined in turn with six exposure terms. In the other two databases Matlab retrieves a large engineering literature, so both strings tie the word to the site by proximity, pairing it within five words of Bangladesh, Chandpur, icddr, HDSS or surveillance, against a twenty-term exposure block.

Every string and its record count is given in Appendix A, including the broad Scopus string that returned 160 records dominated by the MATLAB software package and was therefore not screened. Bibliographic details and abstracts not readable from PubMed were obtained from OpenAlex, PubMed Central or the publisher’s page. No grey literature and no further database were used; for the searches beyond the databases see 2.7, and for what remains undone see 4.3.

2.4 Study selection

Titles and abstracts were screened against those criteria in a single pass, each database handled separately, the six PubMed searches having first been pooled and de-duplicated on PubMed identifier. Figure 2 carries the counts and Appendix A gives them per database. Overlap between the Scopus records and the PubMed records already excluded at screening was not tracked, so the screened total counts an unknown number of records twice; that, and seven Web of Science records the platform would not render, are stated in 4.3. Two records needed full text to settle eligibility, and both were resolved in the citation-chasing pass described below.

2.5 Data extraction

For each included study the following were extracted from the abstract: study design, data period, exposure, outcome, analytic method, principal quantitative findings, and the specific role of Matlab data. Extraction was done once and the summary table was re-read against each abstract. Effect estimates are reported as given by the authors.

2.6 Synthesis and appraisal

Heterogeneity in exposures, outcomes and designs precluded meta-analysis. Studies were grouped into eight exposure–outcome themes and synthesised narratively, then cross-tabulated by exposure class against outcome class (Figure 3), plotted against the years of Matlab data each analysed (Figure 4), and mapped as exposure–outcome links classed as established, mixed, null or absent (Figure 5). No validated risk-of-bias tool exists for this mix of designs, so each study was graded on design strength alone, on the three-level scale in panel C of Figure 1.

2.7 Searching beyond the databases

Two further passes followed, both aimed at one blind spot: all three strings require Matlab in the title, abstract or keywords, so a Matlab study titled “… in rural Bangladesh” is invisible to them unless its abstract happens to name the site.

Reference lists were sought for all 47 included studies and for the Shackleton 2024 topic review through Europe PMC: 40 could be screened and seven could not, for the reasons itemised in 4.3. Google Scholar was then searched with one query per exposure–outcome theme, pairing the exact phrase “Matlab, Bangladesh” with theme terms, since Scholar indexes full text rather than fielded metadata.

Between them the two passes added eight studies, every one titled “Bangladesh” rather than “Matlab” — the blind spot made concrete — and settled three records that abstract-level screening had left open. A final Scholar query probing the gaps claimed in 4.2, pairing the site phrase with cardiovascular, renal, mental-health and cyclone terms, returned nothing eligible. Appendix C records every candidate the two passes surfaced and why each was included or excluded.

3Results

3.1 Study selection

The six PubMed searches returned 126 records, 78 after de-duplication; 48 were excluded on title and abstract, leaving 30 studies. The Scopus search added 89 records, of which 20 were studies already included from PubMed; of the remaining 69, 61 were excluded and eight met the criteria. The Web of Science search added 65 records, 22 of them studies already included and 36 of the rest assessed, from which one further study met the criteria — Nyma et al. 2024 on biomass cooking fuel and birth size — giving 39 studies from the three databases (Figure 2). In all three databases the commonest reason for exclusion was that the paper mentions weather or climate only in passing while studying something else. Scopus contributed two kinds of record that PubMed had not returned: papers in journals MEDLINE does not index (Ecology of Food and Nutrition, PLOS Water), and papers whose climatic content sits in the author keywords rather than the abstract.

PRISMA 2020 flow of records through the review Two parallel arms. From three databases, 280 records were identified, 97 removed or not assessed before screening, 183 screened, 144 excluded, 39 sought and retrieved, and 39 included. Backward citation searching and a themed Google Scholar pass contributed a further 8. Forty-seven studies were included in total. IDENTIFICATION SCREENING INCLUDED Identification of studies via databases Via other methods PubMed searched with six strings; Scopus and Web of Science with a site-proximity string against a twenty-term exposure block. All on 16 October 2026. Records identified from databases n = 280 • PubMed 126 • Scopus 89 • Web of Science Core Collection 65 Duplicates within the PubMed searches; Scopus and Web of Science records that were already included studies; and seven Web of Science records that the platform would not render, which were therefore never assessed. Removed or not assessed before screening, n = 97 • Duplicates within PubMed 48 • Already-included studies re-found: Scopus 20, Web of Science 22 • Could not be rendered from Web of Science and never assessed 7 Reference lists were sought for all 47 included studies; 40 could be screened. Google Scholar was searched with one query per theme, using the exact phrase “Matlab, Bangladesh” against theme terms. Records identified from other methods • Backward citation searching: reference lists of 40 included studies • Google Scholar: 8 themed queries, one per exposure–outcome theme Title and abstract screening against the pre-stated eligibility criteria. Records screened n = 183 • PubMed 78, Scopus 69, Web of Science 36 Commonest reason throughout: the paper mentions weather or climate only in passing while studying something else. Records excluded n = 144 • PubMed 48 • Scopus 61 • Web of Science 35 Reasons itemised in section 3.1 Every report sought was retrieved. Two records needed full text to settle eligibility and both were obtained. Reports sought for retrieval n = 39 • Retrieved: 39 Every report sought was retrieved, including the two that needed full text to settle eligibility. Reports not retrieved n = 0 Candidate reports were assessed individually; every exclusion and its reason is itemised in Appendix C. Studies included from other methods n = 8 • Citation searching 3: Emch 2008, Koelle 2005b, Alam 2015 • Google Scholar 5: Shahriar 2021, Siddharthan 2018, Akanda 2013, Das 2014, Hasan 2018 All 39 retrieved reports met the criteria; eligibility had been settled at screening for the rest. Reports assessed for eligibility n = 39 Reports excluded n = 0 47 studies, all single reports, published 1981 to 2024. Studies included in the review n = 47 • 39 from databases · 8 from other methods • PubMed 30 · Scopus 8 · Web of Science 1 · citation searching 3 · Google Scholar 5

Figure 2. PRISMA 2020 flow of records. The left arm is the three database searches; the right arm is backward citation searching and the themed Google Scholar pass, which contributed eight studies between them. The screened total of 183 counts an unknown number of records twice, because overlap between the Scopus and Web of Science records and the PubMed records already excluded at screening was not tracked (section 2.4). Hover any box for detail.

3.2 Study characteristics

Table 1 summarises the 47 studies. All are observational. Twelve use daily or monthly time series of mortality or hospital cases; fourteen address cholera; four exploit the 1986 embankment as a natural experiment; four use cohorts nested in the HDSS population (MINIMat, the ten-year fuel-use cohort, the Pregnancy Weight Gain study and the GEO Health project); and six are seasonality analyses, four of them predating explicit climate framing. Eight studies fall between 1981 and 1999, 23 between 2004 and 2015, and 16 from 2017 onwards. The exposure data come from the Chandpur meteorological station, satellite ocean products (SST, sea-surface height, chlorophyll), phytoplankton and chlorophyll-a sampled from local ponds, global climate indices (NINO3, Dipole Mode Index), national flood-extent records, the Meghna–Dhonagoda embankment and, for air pollution, household cooking-fuel type recorded in the HDSS socio-economic census.

Table 1. All 47 studies47 rowsDesign and method, data period, exposure, outcome, principal finding and design grade, grouped by theme.
StudyDesign and methodData periodExposureOutcomePrincipal findingGrade
Household air pollution
Alam et al. 2015Cross-sectional spirometry survey; random sample from the icddr,b surveillance database; adjusted logistic regressionFeb 2011–Mar 2012Household solid (biomass) cooking fuel, among several determinantsCOPD by post-bronchodilator spirometry, GOLD and lower-limit-of-normal criteria (3,744 adults aged ≥40)COPD prevalence 13.5% (GOLD) and 10.3% (LLN), higher in rural residents and men; solid fuel use associated with COPD, OR 1.5 (95% CI 1.2–1.9) by GOLD and 1.3 (1.0–1.8) by LLNB
Shahriar et al. 2021Cross-sectional, repeated 48-hour personal exposure measurement; adjusted regressionn/s (two seasons)Measured personal PM2.5 and black carbon (MicroPEM, gravimetric) and carbon monoxide (Lascar), Matlab and AraihazarEndothelial function (reactive hyperaemia index), 199 non-smoking womenNull: no association after adjustment — PM2.5 OR 0.97 (95% CI 0.92–1.01), black carbon 0.85 (0.72–1.01), CO 0.89 (0.64–1.25)A
Siddharthan et al. 2018Pooled cross-sectional, 13 LMIC settings (Matlab one site)2011–2012 (Bangladesh)Biomass as the primary household fuel (self-reported)COPD by spirometry (12,396 adults; 1,824 in Matlab)HAP associated with COPD overall (aOR 1.41, 95% CI 1.18–1.68; women 1.70). Matlab had 98.1% biomass use and 15.3% COPD prevalence, the highest of the 13 sitesB
Nyma et al. 2024Cohort analysis nested in the Pregnancy Weight Gain study; multivariable regressionn/sHousehold biomass cooking fuel vs gasBirth weight, length, head circumference, SGA and body proportionality (909 women)721 women (79.3%) used biomass; their infants were lighter by 94.3 g (95% CI −155.9 to −32.6) and shorter, with proportionality indices also affectedA
Morgan et al. 2019Pooled cross-sectional, 12 sites in 6 LMICs (Matlab one site)n/sBiomass-fuel smoke; tobacco; educationAsthma–COPD overlap (ACO), spirometryACO 7.8% in Matlab, highest of 12 sites; biomass OR 1.48 (0.98–2.23)B
Alam et al. 2012Retrospective cohort; verbal autopsy10 yearsSolid cooking fuel vs natural gas (household air pollution)Adult cardiopulmonary mortalityRespiratory mortality IRR 2.26 (95% CI 1.02–4.99) in solid-fuel homes; combined cardiopulmonary 6.2 vs 5.3 per 1,000 person-years (not significant)B
Temperature, weather and mortality
Babalola et al. 2018Monthly time series; regression with ARIMA errors1982–2008Seasonally adjusted monthly mean/max temperatureInfant mortality (49,426 deaths)Warmer months → fewer infant deaths, strongest for neonates; no heat-related excessA
Lindeboom et al. 2012Daily time series; Poisson with cubic splines1983–2009Temperature, rainfall, cyclonesAll-cause mortality by age and sexU-shaped curve, minimum ≈29 °C; ≥60 y +5.4% per 1 °C below 23 °C; rain >100 mm/day raises 5–19 y mortalityA
Hashizume et al. 2009Daily time series; Poisson GLM, lags 0–13 d1994–2002Daily mean temperatureAll-cause and cause-specific mortality (13,270 deaths)+3.2% (0.9–5.5) mortality per 1 °C fall; no detectable heat effectA
Flooding, flood control and salinity
Akanda et al. 2013Multi-site comparative analysis of six Bengal Delta surveillance sites, Matlab among them1998–2007 (Matlab)Estuarine salinity, inland flood inundation, seasonal low and high river flow of the Ganges and BrahmaputraCholera incidence and the two annual outbreak peaksThe spring peak follows low river flow and salinity intrusion, the autumn peak follows inland flood inundation; the two peaks have distinct macro-scale driversB
Milojevic et al. 2012Controlled interrupted time series2001–2007 (2004 flood)Village flooded in 2004 vs notMortality, diarrhoea, ARI (~211,000 people)No excess mortality or diarrhoea after adjustment; ARI RR 1.25 (1.06–1.47) in the 6 months afterA
Carrel et al. 2010Zero-inflated negative binomial, household leveln/sResidence inside vs outside embankmentHousehold cholera case countsNo overall association; among affected households, more cases inside the embankmentB
Carrel et al. 2009Spatial and space–time cluster detection1983–2003Flood protection (embankment)Cholera case clusteringClusters moved from dry to rainy season after embankment; onset lagged inside protected areaB
Emch 1999Case–control with a GIS database; community controls1992–1994Flood-control embankment; water availability and use; sanitation; SESHospitalised cholera and non-cholera watery diarrhoeaFlood control related to both disease types, by a mechanism the author could not explain; water availability, use and sanitation were the strongest modifiable risks, more so for non-cholera diarrhoeaB
Myaux et al. 1997Before–after comparison, protected vs unprotected villages1983–86 vs 1989–92Flood-control embankmentChild (0–4 y) mortalityAfter embankment, unprotected villages up to 29% higher mortality at 1–4 y (P<0.001)B
Stewart et al. 1990Post-flood household survey; inside vs outside the embankment1988 flood (Oct and Dec surveys)1988 flood damage; residence inside vs outside embankmentLength, weight and MUAC, children 6–35 months (600 households)Anthropometry did not differ by degree of flood damage or by embankment side, and the usual post-harvest improvement from October to December did not appear; landlessness and lack of maternal education predicted lower anthropometryB
Climate variability and cholera
Nusrat et al. 2024Model development/validation; smartphone pilotn/sSatellite-derived hydroclimatic indicatorsCholera risk forecasts1-km monthly risk maps reproduce Matlab’s bimodal seasonal cholera patternC
Wu et al. 2018Case-crossover; conditional logistic regression1983–2009Heatwaves, rainfall, tree coverIndividual cholera casesOR 1.53 (1.07–2.19) two days after a heatwave on wet days; 1.80 (1.01–3.22) with low tree coverA
Islam et al. 2015Bi-weekly pond monitoring; culture and direct fluorescent antibody; Pearson correlation1 yearPhytoplankton, blue-green algae, chlorophyll-a and water physico-chemistry in two pondsMonthly cholera cases (Matlab hospital records)Cholera cases correlated with blue-green algae, total phytoplankton, chlorophyll-a and viable-but-non-culturable V. cholerae O1; the organism stayed viable for a year in association with Anabaena variabilisB
Emch et al. 2008Ordered probit on monthly cholera magnitude, deseasonalised by construction (Matlab and two Vietnam sites)1983–2003Ocean chlorophyll, sea-surface temperature and height (satellite); Chandpur rainfall and temperature; river discharge at Demra and BhairabMonthly cholera magnitude, MatlabRising ocean chlorophyll concentration is associated with larger cholera outbreaks in Matlab; environmental effects were consistent where cholera is endemic and differed where epidemics are rareA
Cash et al. 2014Correlation of monthly series (Dhaka and Matlab)n/sENSO, Pacific SST, monsoon floodingCholera and shigellosisBoth diseases rise after heavy monsoon floods, which follow El Niño eventsB
Hashizume et al. 2013Cross-wavelet coherency (Dhaka and Matlab)1983–2008IOD, ENSO indicesMonthly cholera casesClimate–cholera coupling non-stationary; ENSO stronger in Matlab; no IOD signal thereB
Ali et al. 2013SARIMA forecasting model1988–2001Minimum temperature, SST, rainfallMonthly cholera (4,157 cases)+6% per 1 °C minimum temperature; +25% per 1 °C SST; rainfall null; RMSE 0.108A
Hashizume et al. 2011Negative binomial time series (Dhaka and Matlab)1993–2007Indian Ocean Dipole, ENSO, Bay of Bengal SSTMonthly cholera patientsMatlab +6.9% (3.2–10.8) per 0.1 DMI rise over lags 0–3; BoB SST positive at lags 0–3A
Emch et al. 2010GEE logistic models by epidemic seasonn/sSST, sea-surface height, chlorophyll × household SESLaboratory-confirmed choleraPre-monsoon: SSH inverse, chlorophyll positive; higher SES protective in both seasonsA
Islam et al. 2009CART and principal components on monthly datan/sMonthly temperature, sunshine hoursMonthly cholera cases≈23 cases/month below 23.25 °C vs 54 above 28.66 °C with ≥4.13 h/day sunshineB
Constantin de Magny et al. 2008Lagged correlation with satellite data (Matlab and Kolkata)1998–2006Coastal chlorophyll, SST, rainfallMonthly cholera casesChlorophyll anomalies precede cholera by ≈1 monthB
Koelle et al. 2005Nonlinear transmission model fitted to monthly series + climate indices≈40 yearsMonsoon rainfall, river discharge, flooding, SST, ENSOEl Tor cholera incidenceClimate drives transmission except in post-epidemic ‘refractory’ periods of high immunityB
Glass et al. 1982Descriptive analysis of 15 years of hospital surveillance records1966–1980Monsoon rains, river water levels, seasonCulture-confirmed cholera (7,141 patients, surveillance population 240,000)No constant relationship between the onset or peak of the annual epidemic and the onset or peak of the monsoon rains or river levels; children 2–9 y and adult women most often hospitalisedB
Koelle et al. 2005bAdaptive-dynamics model with a qualitative application to Matlab cholera1966–2002Seasonal forcing, interpreted through monsoon rainfall (Chandpur station)Cholera incidence and El Tor strain replacementPathogens evolve their sensitivity to seasonal forcing; the model links the observed strain replacement to a shift in monsoon rainfall. Mainly theoretical, with the Matlab series used for monthly reproductive ratesC
Weather and childhood diarrhoea and rotavirus
Asare et al. 2022Transmission model fitted to Dhaka, validated on Matlab1990–2012; validation 2013–2019Diurnal temperature range, surface water, wetness; birth rateRotavirus hospitalisations+19% transmission per 1 °C diurnal range; falling birth rate drives seasonal shiftB
Das et al. 2014Quasi-Poisson generalised additive models of monthly counts, adjusted for secular trend; four icddr,b sites2010–2012Season and calendar month, framed by the tropical monsoon cycle; no weather variable testedRotavirus, cholera, shigellosis and ETEC diarrhoea (21,138 patients; 4,240 at Matlab)Seasonality differs by pathogen and by site: Matlab cholera peaks in March–April and September–October, with a rotavirus trough in July. The authors note meteorological variables were not testedB
Wu et al. 2014Logistic regression at bari (household-cluster) level2000–2006Hot days, heavy-rain days, mean temperature, rainfallChildhood (<5 y) diarrhoea+0.8–3.8% cases per extra hot day; +1–6.2% per extra heavy-rain dayA
Weather and childhood pneumonia
Tao et al. 2022Interrupted time series + case-crossover2012–2016Daily temperature, heatwaves; PCV10 rolloutChildhood pneumonia hospitalisationHeat lag-0 OR 1.28 (1.02–1.60); cold lag-28 OR 1.53 (1.06–2.22); cold effect fell and heat effect rose after PCV10 (RR 0.59)A
Hossain et al. 2020bEcological analysis of admissions; generalised linear model, Poisson link2012–2016Daily mean temperature; temperature variation during the stay; humidity variationLength of stay for under-5 pneumonia, and its costLOS +1.0% per 1 °C mean temperature (RR 1.010, 95% CI 1.001–1.018); +9.3% per 1 °C of temperature variation during the stay (RR 1.093, 1.051–1.138); +2.2% per 1% humidity variation (RR 1.022, 1.004–1.039)B
Hossain et al. 2020aPoisson GLM with PCV adjustment2012–2016Temperature, relative humidityChildhood pneumonia (Matlab Hospital)RR 0.985 (0.974–0.997) per 1 °C; PCV RR 0.489 (0.435–0.551); humidity nullA
Heat and season in pregnancy, and climate-linked migration
Ahmed et al. 2020Multivariable linear regression on linked HDSS and hospital birth recordsBirths Jul 2011–Jun 2015Weeks of exposure to each agro-climatic season during the 2nd and 3rd trimestersBirth weight (3,831 term singletons)Highest birth weight in the post-aman harvest (March, 2,930.5 g) and lowest in the monsoon (July, 2,830.6 g); +6.3 g (95% CI 1.6–10.9) per week of post-aman exposure in the third trimester, +88.2 g across a full 14 weeksA
Rahman et al. 2019Population-based pregnancy cohort; time-series decomposition for trend and seasonality1990–2014Season of birth (seasonal component of the series)Preterm birth (63,063 live births)Preterm birth fell from 29% (1990–94) to 11% (2010–14), 27% of the decline attributable to rising maternal education and falling parity; marked seasonal variation persisted throughout, which the authors read as environmental stress on pregnant womenB
Hasan et al. 2018Hospital-based cohort of pregnant women; multivariable logistic regression on determinantsBirths 2012–2014Season of conception (monsoon or dry season vs summer), among several determinantsInadequate third-trimester weight gain, ≤4 kg (1,883 women; 1,026 affected, 54%)Conceiving in the monsoon or dry season raised the odds of inadequate third-trimester weight gain by 40% against summer conception (OR 1.4, 95% CI 1.1–1.8)B
Call et al. 2017Discrete-time event historyn/sTemperature, precipitation, flooding anomaliesTemporary out-migrationFloods briefly suppress migration; high temperatures raise it for 1–2 years; prolonged heavy rain suppresses itA
Rashid et al. 2017Linear regression, MINIMat birth cohortBirths 2001–2003Gestational temperature (2-week means), seasonBirth weight and length (3,267 births)Winter-born shortest (47.5 vs 47.9 cm, P<0.001); warm last month → longer; warm mid-gestation → heavierA
Seasonality of mortality (season as climatic proxy)
Iqbal et al. 2007Descriptive analysis of surveillance deaths, with seasonal breakdown1985–2000Season (monsoon months); type of water bodyChildhood drowning deaths (989 deaths)Most drownings fell in April–October, that is the monsoon; 80.5% were aged 1–4 and 45% drowned in ponds; drowning rates stayed flat while all-cause 1–4 y mortality fell from 20.7 to 5.2 per 1,000C
Moore et al. 2004Hazard models, birth cohorts1974–2000 (172,228 births)Season of birth (hungry vs harvest)Survival in infancy and after age 15Hungry-season birth raises infant mortality; adult HR 1.12 (0.87–1.45), not significantA
Becker & Weng 1998Trigonometric regression on monthly deaths1982–1990Calendar month (climatic cycle)Deaths by age and cause (20,328)Seasonality persists despite mortality decline; all-cause and neonatal peaks in winterB
Muhuri 1996Hazard models with season termsn/sSeason (hot-dry, monsoon, hungry, post-harvest)Child mortalityPeaks April and November; hungry-season excess confined to children of unschooled mothersB
Shaikh et al. 1990Descriptive analysis of deaths by age, sex, season and year1966–1987Season (calendar month as the climatic cycle)Diarrhoeal deathsNovember was the peak month; the diarrhoeal death rate ran at 2.0–4.0 per 1,000 but doubled in the 1971 war and 1974–75 famine years; over 20% of all deaths were diarrhoea-related, a share that did not fall over the periodC
Becker 1981Descriptive seasonal analysis1972–1974Calendar month (climatic cycle)Deaths by age and cause (pop. ≈260,000)Post-neonatal peak April (dysentery); drownings peak July; older adults in cool seasonC

Table 1. Characteristics of included studies, newest first within each theme. n/s = not stated in abstract. Grade: A adjusted time-series or individual-level design; B ecological or unadjusted comparison; C descriptive or methods development. The eight studies added by the Scopus search are Emch 1999, Stewart 1990, Islam 2015, Glass 1982, Ahmed 2020, Rahman 2019, Iqbal 2007 and Shaikh 1990; Nyma 2024 was added by Web of Science.

3.3 Evidence map

Table 1 lists the studies by theme; Figure 3 lays the same 47 studies out by what was measured against what was counted, each study placed once at its primary exposure and its primary outcome. The themes of section 3.4 do not map one-to-one onto these cells: the two Carrel studies sit under flood control as a theme but under cholera as an outcome, which is why the cholera column holds 17 studies while the cholera theme holds 14.

Evidence map of 47 Matlab studies by exposure and outcome A matrix of eight exposure classes against six outcome classes. Cholera accounts for 17 of the 47 study assignments. The ambient air pollution row is empty across every outcome. Mortality Cholera Diarrhoea & rotavirus Pneumonia & ARI Birth & child growth Other health Studies in row Temperature, heat and heatwaves 3 studies · Temperature, heat and heatwaves × Mortality3 2 studies · Temperature, heat and heatwaves × Cholera2 2 studies · Temperature, heat and heatwaves × Diarrhoea and rotavirus2 3 studies · Temperature, heat and heatwaves × Pneumonia and ARI3 1 study · Temperature, heat and heatwaves × Birth and child growth1 1 study · Temperature, heat and heatwaves × Other health1 12 Rainfall, humidity and river level no study · Rainfall, humidity and river level × Mortality· 2 studies · Rainfall, humidity and river level × Cholera2 no study · Rainfall, humidity and river level × Diarrhoea and rotavirus· no study · Rainfall, humidity and river level × Pneumonia and ARI· no study · Rainfall, humidity and river level × Birth and child growth· no study · Rainfall, humidity and river level × Other health· 2 Flood, river flow and salinity 2 studies · Flood, river flow and salinity × Mortality2 3 studies · Flood, river flow and salinity × Cholera3 1 study · Flood, river flow and salinity × Diarrhoea and rotavirus1 no study · Flood, river flow and salinity × Pneumonia and ARI· 1 study · Flood, river flow and salinity × Birth and child growth1 no study · Flood, river flow and salinity × Other health· 7 Ocean and climate indices no study · Ocean and climate indices × Mortality· 9 studies · Ocean and climate indices × Cholera9 no study · Ocean and climate indices × Diarrhoea and rotavirus· no study · Ocean and climate indices × Pneumonia and ARI· no study · Ocean and climate indices × Birth and child growth· no study · Ocean and climate indices × Other health· 9 Local water and plankton ecology no study · Local water and plankton ecology × Mortality· 1 study · Local water and plankton ecology × Cholera1 no study · Local water and plankton ecology × Diarrhoea and rotavirus· no study · Local water and plankton ecology × Pneumonia and ARI· no study · Local water and plankton ecology × Birth and child growth· no study · Local water and plankton ecology × Other health· 1 Season / agro-climatic cycle 5 studies · Season / agro-climatic cycle × Mortality5 no study · Season / agro-climatic cycle × Cholera· 1 study · Season / agro-climatic cycle × Diarrhoea and rotavirus1 no study · Season / agro-climatic cycle × Pneumonia and ARI· 3 studies · Season / agro-climatic cycle × Birth and child growth3 1 study · Season / agro-climatic cycle × Other health1 10 Household air pollution 1 study · Household air pollution × Mortality1 no study · Household air pollution × Cholera· no study · Household air pollution × Diarrhoea and rotavirus· no study · Household air pollution × Pneumonia and ARI· 1 study · Household air pollution × Birth and child growth1 4 studies · Household air pollution × Other health4 6 Ambient air pollution no study · Ambient air pollution × Mortality· no study · Ambient air pollution × Cholera· no study · Ambient air pollution × Diarrhoea and rotavirus· no study · Ambient air pollution × Pneumonia and ARI· no study · Ambient air pollution × Birth and child growth· no study · Ambient air pollution × Other health· 0 Studies in column 11 17 4 3 6 6 47 Studies per cell 1 2–3 4–5 6+ none

Figure 3. Evidence map of the 47 included studies, by exposure class and outcome class. Each study appears once, at its primary exposure and primary outcome. Shading is the count in the cell; a dotted cell is one no study has occupied.

Three features of the map carry the review. The cholera column (17 studies) is half again the size of the next largest, and nine of those arrive on ocean and climate indices rather than on anything measured in Matlab itself. The ambient air pollution row is empty across all six outcomes — not thin, empty. And the rainfall row holds only two studies, and the one that tested rainfall against cholera incidence directly, Glass 1982, found nothing; every other rainfall result in this review is a secondary term inside a temperature or flood analysis rather than a study of rainfall as such.

3.4 Coverage of the Matlab record

The HDSS has registered births, deaths and migrations without a break since 1966. Figure 4 shows how much of that record each study actually analysed. Thirty-five of the 47 studies state a data period; the other twelve do not, including four of the six household air pollution studies.

Years of Matlab data analysed, by study Bars showing the span of Matlab data each study analysed, 1966 to 2026, grouped by theme. Mortality and cholera coverage ends in 2009; no included study analyses data after 2019. no data analysed after 2019 1970 1980 1990 2000 2010 2020 Matlab HDSS record The Matlab HDSS has registered births, deaths and migrations continuously since 1966 continuous since 1966 Temperature, weather and mortality Babalola 2018 Babalola 2018 · 1982–2008 · grade A A Lindeboom 2012 Lindeboom 2012 · 1983–2009 · grade A A Hashizume 2009 Hashizume 2009 · 1994–2002 · grade A A Flooding, flood control and salinity Myaux 1997 Myaux 1997 · 1983–1992 · grade B B Carrel 2009 Carrel 2009 · 1983–2003 · grade B B Stewart 1990 Stewart 1990 · 1988–1988 · grade B B Emch 1999 Emch 1999 · 1992–1994 · grade B B Akanda 2013 Akanda 2013 · 1998–2007 · grade B B Milojevic 2012 Milojevic 2012 · 2001–2007 · grade A A Climate variability and cholera Glass 1982 Glass 1982 · 1966–1980 · grade B B Koelle 2005a Koelle 2005a · 1966–2004 · grade B B* Koelle 2005b Koelle 2005b · 1966–2002 · grade C C Emch 2008 Emch 2008 · 1983–2003 · grade A A Hashizume 2013 Hashizume 2013 · 1983–2008 · grade B B Wu 2018 Wu 2018 · 1983–2009 · grade A A Ali 2013 Ali 2013 · 1988–2001 · grade A A Hashizume 2011 Hashizume 2011 · 1993–2007 · grade A A de Magny 2008 de Magny 2008 · 1998–2006 · grade B B Childhood diarrhoea and rotavirus Wu 2014 Wu 2014 · 2000–2006 · grade A A Das 2014 Das 2014 · 2010–2012 · grade B B Asare 2022 Asare 2022 · 2013–2019 · grade B B Childhood pneumonia Hossain 2020a Hossain 2020a · 2012–2016 · grade A A Hossain 2020b Hossain 2020b · 2012–2016 · grade B B Tao 2022 Tao 2022 · 2012–2016 · grade A A Heat and season in pregnancy Rahman 2019 Rahman 2019 · 1990–2014 · grade B B Rashid 2017 Rashid 2017 · 2001–2003 · grade A A Hasan 2018 Hasan 2018 · 2012–2014 · grade B B Ahmed 2020 Ahmed 2020 · 2011–2015 · grade A A Seasonality of mortality Shaikh 1990 Shaikh 1990 · 1966–1987 · grade C C Becker 1981 Becker 1981 · 1972–1974 · grade C C Moore 2004 Moore 2004 · 1974–2000 · grade A A Becker & Weng 1998 Becker and Weng 1998 · 1982–1990 · grade B B Iqbal 2007 Iqbal 2007 · 1985–2000 · grade C C Household air pollution Siddharthan 2018 Siddharthan 2018 · 2011–2012 · grade B B Alam 2015 Alam 2015 · 2011–2012 · grade B B 4 further household air pollution studies state no data period

Figure 4. Years of Matlab data analysed, by study and theme, for the 35 studies that state a period. All bars share one colour; the letter at the bar end is the design grade, so grade is carried by the label rather than by shade. * Koelle 2005 reports roughly four decades without giving exact years. The twelve studies that state no period are Morgan 2019, Alam 2012, Nyma 2024, Shahriar 2021, Carrel 2010, Nusrat 2024, Islam 2015, Cash 2014, Emch 2010, Islam 2009, Call 2017 and Muhuri 1996.

The gap this review exists to identify is visible as white space on the right. Mortality analysis stops at 2009 and cholera at the same year; ten studies reach past 2010 — the three pneumonia analyses (2012–2016), Rahman’s preterm series (to 2014), Hasan’s weight-gain cohort (to 2014), Ahmed’s birth-weight study (to 2015), Das’s pathogen seasonality (to 2012), the two COPD surveys (to 2012) and Asare’s rotavirus validation (to 2019). Nothing in this literature analyses a Matlab observation made after 2019, although the HDSS has gone on recording throughout — roughly seven years of continuous mortality, hospital and weather data that no published climate–health analysis has touched. That is the same window in which the local weather station, satellite PM2.5 products and the 2026 personal-exposure measurements all became available.

Figures 3 and 4 count studies and years. Figure 5 asks the question the gaps section turns on: between which exposures and which outcomes has a relationship actually been established, where is the evidence mixed, where did somebody look and find nothing, and where has nobody looked at all. The four states are worth keeping apart, because a null result and an absence of research are not the same claim and the review should not let them blur.

Exposure to outcome links in the Matlab literature, and the links that are absent Exposures on the left join health outcomes on the right by ribbons whose width is the number of studies. Solid marks an established association, hatched a contested one, outlined a relationship searched for and not found, dashed a link no study has examined. EXPOSURE HEALTH OUTCOME 47 studies 47 study assignments Ocean, climate indices and aquatic ecology → Cholera · 10 studies · warm sea-surface, ENSO, IOD and chlorophyll precede cholera Rainfall and river level → Cholera · 2 studies · Glass 1982 found no relation to monsoon rains; Koelle 2005 ties strain replacement to monsoon shifts Flood, river flow and salinity → Cholera · 3 studies · flood protection concentrated cases; estuarine salinity and flood inundation drive the two annual peaks Flood, river flow and salinity → Diarrhoea and rotavirus · 1 study · flood control related to diarrhoea by an unexplained mechanism Flood, river flow and salinity → Mortality, all-cause and cause-specific · 2 studies · embankment improved child survival; the 2004 flood showed no excess once controlled Flood, river flow and salinity → Birth and child growth · 1 study · no anthropometric difference by flood damage or embankment side Temperature, heat and heatwaves → Cholera · 2 studies · heatwaves raise individual cholera risk two days later on wet days Temperature, heat and heatwaves → Diarrhoea and rotavirus · 2 studies · each extra hot day raises childhood diarrhoea 0.8-3.8% Temperature, heat and heatwaves → Pneumonia and ARI · 3 studies · heat raises same-day admissions, cold acts at 4 weeks, variability lengthens stay Temperature, heat and heatwaves → Mortality, all-cause and cause-specific · 3 studies · cold harmful; heat searched for and not found Temperature, heat and heatwaves → Birth and child growth · 1 study · cold late gestation shortens infants Temperature, heat and heatwaves → Migration · 1 study · hot periods raise temporary out-migration for 1-2 years Temperature, heat and heatwaves → Vascular and other non-communicable disease · no study · no study has examined heat against cardiovascular, renal or mental-health outcomes Season / agro-climatic cycle → Diarrhoea and rotavirus · 1 study · pathogen-specific seasonal peaks differ between Matlab and the other sites Season / agro-climatic cycle → Mortality, all-cause and cause-specific · 5 studies · mortality peaks in April and November Season / agro-climatic cycle → Birth and child growth · 3 studies · monsoon births lightest; preterm seasonality persists over 25 years; monsoon or dry-season conception raises inadequate third-trimester weight gain Season / agro-climatic cycle → Drowning · 1 study · drownings concentrate April-October and the rate has not fallen Household air pollution → Mortality, all-cause and cause-specific · 1 study · solid-fuel homes: respiratory mortality IRR 2.26 Household air pollution → Birth and child growth · 1 study · biomass users' infants 94.3 g lighter Household air pollution → Chronic respiratory disease · 3 studies · biomass raises asthma-COPD overlap and COPD; solid fuel OR 1.5 for COPD in Matlab and Kamlapur Household air pollution → Vascular and other non-communicable disease · 1 study · measured personal PM2.5, black carbon and CO showed no association with endothelial function Ambient air pollution → Pneumonia and ARI · no study · no study has linked ambient pollution to Matlab hospital admissions Ambient air pollution → Mortality, all-cause and cause-specific · no study · no study has linked ambient PM2.5 or ozone to Matlab mortality Heat and pollution combined → Mortality, all-cause and cause-specific · no study · no joint model of heat and pollution exists for this population Ocean, climate indices and aquatic ecology 10 studies Rainfall and river level 2 studies Flood, river flow and salinity 7 studies Temperature, heat and heatwaves 12 studies Season / agro-climatic cycle 10 studies Household air pollution 6 studies Ambient air pollution none Heat and pollution combined none Cholera 17 studies Diarrhoea and rotavirus 4 studies Pneumonia and ARI 3 studies Mortality, all-cause and cause-specific 11 studies Birth and child growth 6 studies Drowning 1 study Migration 1 study Chronic respiratory disease 3 studies Vascular and other non-communicable disease 1 study Ribbon width = studies association established mixed or contested searched for, not found no study exists

Figure 5. Exposure–outcome links across the 47 studies. Ribbon width is the number of studies, counted as in Figure 3 with each study at its primary exposure and outcome. Solid = association established; hatched = mixed or contested; outline only = searched for and not found; dashed = no study exists. Hover any ribbon for the finding it carries.

Read left to right, the literature narrows sharply. Twenty-eight of the 47 studies land on two outcomes, cholera (17) and mortality (11), and ten of the seventeen cholera studies come in on one exposure family — nine on ocean and climate indices measured far from Matlab, and a tenth on the ecology of two ponds. Four ribbons are hatched rather than solid: rainfall against cholera, flooding against mortality, flood control against diarrhoea, and temperature against pneumonia, where the studies disagree about sign. Two are outlines, and they matter more than their width suggests: Stewart 1990 looked for a flood effect on child anthropometry and found none, and Shahriar 2021 measured personal PM2.5, black carbon and carbon monoxide against vascular function and found none. Glass 1982’s null on the monsoon rains now sits inside the hatched rainfall ribbon, beside Koelle’s reading of the same seasonality. The heat arm of the temperature–mortality ribbon is a third such null, folded into a solid ribbon because the same three studies establish the cold arm — cold kills in Matlab, and heat, searched for at every lag out to 13 days, does not.

The dashed ribbons are the review’s argument in one shape. Ambient air pollution reaches nothing: not mortality, not hospital admissions, nothing, in a population lying downwind of Dhaka inside the Indo-Gangetic winter haze belt. Heat and pollution have never been modelled together, although the daily mortality series starts in 1983 and PM2.5 has now been measured on the ground. And the only exposure reaching non-communicable disease is household smoke, in Shahriar’s single null result, in a population whose epidemiological transition is documented in the same surveillance system. Those absences, plus the temporal cliff in Figure 4, are what a new analysis on this dataset would be filling.

3.6 Synthesis by theme

Temperature, weather and mortality3 studiesCold, not heat: 3.2% more deaths per 1 °C fall, with no heat effect detectable anywhere in the record.

Three time-series analyses, together spanning 1982–2009, give a coherent picture: in Matlab it is cold rather than heat that kills. Hashizume et al. 2009 found that each 1 °C fall in daily mean temperature raised all-cause mortality by 3.2% (95% CI 0.9–5.5) over 1994–2002, with the steepest gradients for cardiovascular, respiratory and perinatal deaths, and could detect no heat effect at any lag up to 13 days. Lindeboom et al. 2012 extended the series to 1983–2009 and described a U-shaped curve with its minimum near 29 °C; the cold arm was concentrated in people aged 60 and over (about 5.4% per 1 °C below 23 °C), while heavy rain above 100 mm/day raised mortality among 5–19-year-olds, consistent with drowning. At the monthly scale Babalola et al. 2018 found that warmer months carried fewer infant deaths, with the effect strongest in the neonatal period, again with no sign of heat-related excess. All three are grade A designs. None, however, extends past 2009, so the effect of the warming recorded locally since then is unmeasured.

Flooding, flood control and salinity7 studiesThe 1986 embankment improved child survival but shifted cholera into the rainy season.

The 1986 Meghna–Dhonagoda embankment and the 2004 flood provide two natural experiments. Myaux et al. 1997 compared child mortality inside and outside the embankment before (1983–86) and after (1989–92) its completion and reported that unprotected villages had up to 29% higher mortality at ages 1–4, interpreting the embankment as beneficial for child survival. The cholera work points the other way: Carrel et al. 2009 showed that space–time clusters of cholera shifted from the dry to the rainy season after flood control and began later inside the protected area, and Carrel et al. 2010 found that although protected and unprotected households were equally likely to have any cholera, affected households inside the embankment had more cases, suggesting that impounded water sustains transmission. For acute flood events, the controlled interrupted time series of Milojevic et al. 2012 around the 2004 flood is the most rigorous study in this review: once pre-existing village differences and seasonality were controlled, neither mortality nor diarrhoea showed a convincing excess, and only acute respiratory infection rose (RR 1.25, 95% CI 1.06–1.47 in the following six months). The authors note that earlier flood studies with weaker comparison groups had reported much larger effects.

Two older studies, both added by the Scopus search, bracket this work. Stewart et al. 1990 surveyed 600 households after the 1988 flood and found that anthropometry in children aged 6–35 months differed neither by the degree of flood damage nor by which side of the embankment a household stood on; what did show was the absence of the usual post-harvest improvement between October and December, and a clear gradient by landlessness and maternal education. The null result is consistent with Milojevic’s later and better-controlled finding. Emch 1999 ran a case–control analysis of 1992–1994 hospitalisations against a GIS database and found flood control related to both cholera and non-cholera watery diarrhoea, while writing plainly that he could not explain why; water availability, water use and sanitation were the strongest modifiable risks, and more important for non-cholera diarrhoea than for cholera. Taken together these studies say that the embankment matters for child survival and for cholera, but that acute floods themselves leave less of a mark on mortality and diarrhoea than the earlier literature implied. Akanda et al. 2013 adds the only salinity result in the review: comparing six Bengal Delta surveillance sites, Matlab among them over 1998–2007, it attributes the spring cholera peak to low river flow and estuarine salinity intrusion and the autumn peak to inland flood inundation, so that the two annual peaks have different macro-scale drivers. That is the one study to treat salinity as a health exposure anywhere in this literature.

Climate variability and cholera14 studiesTracks ENSO, the Indian Ocean Dipole, Bay of Bengal temperature and chlorophyll at lags of weeks to months — all of it read against an early null on the monsoon rains.

Cholera is by far the best-studied climate-sensitive outcome, reflecting the site’s origins. The studies progress from regional climate forcing to local weather and finally to operational forecasting — but they do not begin from agreement. The earliest, Glass et al. 1982, analysed 7,141 culture-confirmed cases over 1966–1980 and reported no constant relationship between the onset or peak of the annual epidemic and the onset or peak of the monsoon rains or river water levels, concluding that little progress had been made in understanding transmission. Every later quantitative finding in this theme has to be read against that null. Koelle et al. 2005 fitted a nonlinear transmission model to four decades of monthly cases and showed that monsoon rainfall, river discharge, flood extent, Bay of Bengal SST and ENSO all modulate transmission, but that the climate signal is muted in ‘refractory’ periods after large epidemics when population immunity is high. Constantin de Magny et al. 2008 linked satellite chlorophyll anomalies to cholera about one month later; Emch et al. 2010 found pre-monsoon associations with sea-surface height (inverse) and chlorophyll (positive) that were absent after the monsoon, with household socio-economic status protective in both seasons. Hashizume et al. 2011 quantified the Indian Ocean Dipole effect (+6.9%, 95% CI 3.2–10.8, cholera per 0.1 rise in the Dipole Mode Index in Matlab), but Hashizume et al. 2013 then showed with wavelet analysis that these couplings are non-stationary, with ENSO dominant in Matlab and no consistent IOD signal, and Cash et al. 2014 found that shigellosis, which lacks an aquatic reservoir, responds to El Niño-driven flooding in the same way as cholera, implying that climate acts largely through human exposure rather than vibrio ecology.

At the local scale Islam et al. 2009 partitioned months by temperature and sunshine (about 23 cases a month below 23.25 °C against 54 in the hottest, sunniest months), and Ali et al. 2013 estimated roughly +6% incidence per 1 °C of minimum temperature and +25% per 1 °C of sea-surface temperature in a SARIMA model with rainfall contributing nothing. The only individual-level analysis, Wu et al. 2018, used a case-crossover design on 1983–2009 cases and found cholera risk elevated two days after a heatwave on wet days (OR 1.53, 95% CI 1.07–2.19), more so around households with little tree cover (OR 1.80). Most recently Nusrat et al. 2024 turned these relationships into CholeraMap, a satellite-driven 1-km monthly risk map delivered to Matlab households by smartphone, which reproduced the bimodal seasonal pattern. On mechanism, Islam et al. 2015 moved the chlorophyll argument from satellite to pond: monitoring two Matlab ponds fortnightly for a year, they found monthly cholera cases correlated with blue-green algae, total phytoplankton and chlorophyll-a measured locally, and showed V. cholerae O1 persisting for a year in a viable but non-culturable state in association with Anabaena variabilis. Emch et al. 2008 set Matlab beside two Vietnamese sites over 1983–2003 and found rising ocean chlorophyll associated with larger outbreaks, with the environmental signal consistent where cholera is endemic and different where epidemics are rare; Koelle et al. 2005b, a companion to the Nature paper, argues from an adaptive-dynamics model that the pathogen evolves its own sensitivity to seasonal forcing, tying the El Tor strain replacement to a shift in monsoon rainfall. Taken together, thirteen of the fourteen studies point the same way — warm sea-surface and ambient temperature, El Niño conditions and heavy monsoon flooding precede cholera increases, with lags of weeks to months — while Glass et al. found no stable link to rains or river levels at all. What remains contested is the mechanism and the stability of the relationships over time, and the 1982 null is a reminder that the later signal is partly a product of better exposure data and more flexible models.

Weather and childhood diarrhoea and rotavirus3 studiesExtreme-weather counts beat averages — each additional hot day raised cases by 0.8–3.8%.

Wu et al. 2014 analysed under-five diarrhoea in 2000–2006 at the level of the bari and found that extreme-weather counts, not averages, mattered: each additional hot day raised daily cases by 0.8–3.8% and each heavy-rain day by 1–6.2%. Asare et al. 2022 fitted rotavirus transmission models to 23 years of Dhaka surveillance and validated them on Matlab data for 2013–2019; a 1 °C wider diurnal temperature range increased transmission by 19%, although the long-run decline in the birth rate explained more of the shift in winter seasonality than weather did. Das et al. 2014 set Matlab beside Mirzapur and two Dhaka sites for 2010–2012 and found that seasonality is pathogen-specific and site-specific rather than general: Matlab cholera peaked in March–April and September–October with a rotavirus trough in July. The authors are explicit that they tested no meteorological variable, so the study establishes the pattern without explaining it — which is the same limitation the older seasonality studies carry.

Weather and childhood pneumonia3 studiesHeat raises admissions on the same day and cold at a four-week lag; the studies disagree on the sign.

A cluster of recent work uses Matlab Hospital admissions for 2012–2016 to study temperature and the pneumococcal conjugate vaccine jointly. Hossain et al. 2020a found higher temperature mildly protective (RR 0.985 per 1 °C, 95% CI 0.974–0.997), humidity unrelated, and PCV introduction associated with a halving of incidence (RR 0.489). Tao et al. 2022 separated the two directions of temperature with a case-crossover design: heat raised admissions on the same day (OR 1.28, 95% CI 1.02–1.60) while cold acted with a four-week lag (OR 1.53, 95% CI 1.06–2.22); after PCV10 the cold-related risk fell but heat and heatwave effects grew, with boys, infants and poorer households most heat-sensitive. Hossain et al. 2020b examined length of stay rather than incidence, over the same 2012–2016 window, and found that it is variability rather than level that matters: each 1 °C of mean temperature lengthened a stay by 1.0% (RR 1.010, 95% CI 1.001–1.018), but each 1 °C of temperature variation during the stay lengthened it by 9.3% (RR 1.093, 1.051–1.138), with humidity variation adding 2.2% per 1% (RR 1.022, 1.004–1.039); the authors attach a cost of roughly 1.8 USD per patient-day to each degree of that variation. Read beside Hossain 2020a, where higher temperature slightly reduced incidence, the pair suggest heat does little to put children in hospital here but a good deal to keep them there.

Household air pollution6 studiesBiomass smoke is linked to obstructive lung disease in all three studies that tested it; the one study to measure concentrations found nothing.

Six studies touch air quality, all of them on household smoke rather than outdoor air, and one of them is the only study in this review to measure a pollutant concentration rather than infer it from fuel type. Shahriar et al. 2021 carried 48-hour personal monitors on 199 non-smoking women in Matlab and Araihazar, twice a year, measuring PM2.5 gravimetrically, black carbon by reflectometry and carbon monoxide by logger, and tested them against endothelial function by peripheral arterial tonometry. Nothing came of it: PM2.5 OR 0.97 (95% CI 0.92–1.01), black carbon 0.85 (0.72–1.01), carbon monoxide 0.89 (0.64–1.25). That is a null result from the best exposure measurement anyone has yet brought to this population, and it is the only vascular or non-communicable outcome in the entire review. Siddharthan et al. 2018, the companion to Morgan from the same consortium, pooled 12,396 adults across 13 settings and found biomass associated with COPD (aOR 1.41, 95% CI 1.18–1.68, and 1.70 in women); Matlab contributed 1,824 participants, 98.1% of them cooking on biomass, and recorded the highest COPD prevalence of all 13 sites at 15.3%. Alam et al. 2015 is the Matlab-anchored version of the same question: 3,744 adults aged 40 and over, sampled at random from the icddr,b surveillance database across rural Matlab and semi-urban Kamlapur and given post-bronchodilator spirometry between February 2011 and March 2012. COPD prevalence was 13.5% by GOLD criteria and 10.3% by the lower limit of normal, higher among rural residents than urban ones, and solid fuel use carried an adjusted odds ratio of 1.5 (95% CI 1.2–1.9). Three of the six studies in this theme therefore put biomass against obstructive lung disease and all three find it, which makes this the one firmly established exposure–outcome link in the air-pollution literature on Matlab — and it rests entirely on fuel type as a proxy, since the only study to measure concentrations found nothing. The newest, Nyma et al. 2024, is the only one in this review to carry a household air pollution exposure through to a birth outcome: among 909 women in the Matlab Pregnancy Weight Gain study, 721 (79.3%) cooked with biomass, and their infants were born 94.3 g lighter (95% CI −155.9 to −32.6) and shorter than those of gas users, with body-proportionality indices affected as well. It sits alongside the agro-climatic birth-weight findings of Ahmed et al. 2020 and Rashid et al. 2017 without either study having adjusted for the other’s exposure, which is a confounding problem waiting to be addressed rather than a settled result. Alam et al. 2012 followed adults for ten years by cooking-fuel type recorded in the HDSS census and found respiratory mortality more than doubled in solid-fuel households (IRR 2.26, 95% CI 1.02–4.99), although combined cardiopulmonary mortality (6.2 vs 5.3 per 1,000 person-years) did not reach significance. Morgan et al. 2019 pooled spirometry from 12 LMIC sites; Matlab had the highest prevalence of asthma–COPD overlap (7.8%), and biomass exposure was associated with higher odds of the condition (OR 1.48, 95% CI 0.98–2.23). No study has linked ambient particulate matter, ozone or other outdoor pollutants to a health outcome in Matlab, and none has examined pollution and temperature together. The exposure data, however, no longer need to be assumed. Akhtar et al. 2026, which falls outside this review because it reports no health outcome, measured 24-hour personal PM2.5 and black carbon in 400 women across rural Matlab and peri-urban Araihazar, with 72–120-hour PM2.5 in 200 kitchens and outdoors; personal and indoor measurements tracked each other closely (R² = 0.722), daily personal and kitchen PM2.5 were higher in rural Matlab than in the peri-urban site, and personal PM2.5 rose as the distance to factories fell below about 300 m. That is the first measured ambient and personal particulate exposure in the HDSS area.

Heat and season in pregnancy, and climate-linked migration5 studiesWinter-born infants are shortest, and unusually hot periods raise out-migration for one to two years.

Rashid et al. 2017 used 3,267 MINIMat births (2001–2003) and Chandpur weather to show that winter-born infants were shortest (47.5 vs 47.9 cm in the monsoon, P<0.001), that warmth in the last month of pregnancy predicted greater length, and that warmth in mid-gestation predicted higher birth weight; the authors read this as a cold-exposure risk for undernourished women rather than a heat risk. Call et al. 2017 applied event-history models to HDSS migration records and found that floods briefly suppressed temporary out-migration, unusually hot periods increased it for one to two years, and prolonged heavy rain reduced it, arguing that environmental shocks disrupt livelihood strategies rather than displace people.

Two studies added by the Scopus search enlarge the pregnancy side and, between them, carry more births than Rashid’s cohort. Ahmed et al. 2020 linked 3,831 term singleton births from the HDSS and Matlab hospital (July 2011 to June 2015) to the agro-climatic calendar and measured exposure in weeks rather than by season of birth: weight peaked in the post-aman harvest (March, 2,930.5 g) and bottomed in the monsoon (July, 2,830.6 g), and each week of post-aman exposure in the third trimester added 6.3 g (95% CI 1.6–10.9), or 88.2 g across a full 14 weeks. The direction agrees with Rashid: the harvest-and-cool part of the cycle is favourable and the monsoon is not, which the authors attribute to seasonal food insecurity and energy stress rather than to temperature as such. Rahman et al. 2019 then showed, across 63,063 live births over 25 years, that preterm birth fell from 29% to 11% — about 27% of the fall attributable to rising maternal education and falling parity — while a marked seasonal variation in the preterm proportion persisted unchanged throughout, which they read as a standing environmental stressor on pregnant women that development has not removed. Hasan et al. 2018 reaches back a step, to the mother rather than the infant: among 1,883 women delivering at Matlab hospital in 2012–2014, 54% gained 4 kg or less in the third trimester, and conceiving in the monsoon or dry season rather than summer raised those odds by 40% (OR 1.4, 95% CI 1.1–1.8). Five studies now point at the agro-climatic cycle as a determinant of pregnancy and birth outcomes in Matlab, and not one of them separates seasonal food availability from seasonal temperature — the two move together in this calendar, and no study in this review has tried to break them apart.

Seasonality of mortality6 studiesFour of the six predate explicit climate framing and read the calendar as the agro-climatic cycle.

These earlier analyses treat the calendar as a stand-in for the agro-climatic cycle and are the root of the later weather work. Becker 1981 described 1972–1974 deaths peaking in April for post-neonates (dysentery), July for drownings and the cool season for older adults; Becker & Weng 1998 showed with trigonometric regression that the cycles persisted through 1982–1990 despite falling mortality. Muhuri 1996 located child-mortality peaks in April and November, with a hungry-season excess confined to children of unschooled mothers, and Moore et al. 2004 found that hungry-season birth raised infant but not adult mortality (adult HR 1.12, 95% CI 0.87–1.45), in contrast to Gambian findings. Two further analyses come from the Scopus search. Shaikh et al. 1990 traced diarrhoeal deaths over 1966–1987 and put the peak in November, with rates of 2.0–4.0 per 1,000 that doubled in the war year of 1971 and the 1974–75 famine, and with diarrhoea accounting for more than a fifth of all deaths throughout — a share that did not fall even as mortality did. Iqbal et al. 2007 analysed 989 drowning deaths over 1985–2000 and found them concentrated in April to October, that is the monsoon, overwhelmingly at ages 1–4 and mostly in ponds; while all-cause mortality at 1–4 years fell from 20.7 to 5.2 per 1,000 over those years, the drowning rate barely moved. That last finding is the mechanism behind two results elsewhere in this review — Becker’s July peak in accidental deaths and Lindeboom’s excess mortality among 5–19-year-olds on days with more than 100 mm of rain — and it is the one seasonal hazard in Matlab that half a century of health-service improvement has not reduced.

3.7 Methodological quality

Eighteen studies were graded A, twenty-four B and five C; every included study is graded. The eight studies added by the Scopus search are weaker on average than the PubMed set — one A, five B and two C — because they include most of the earliest descriptive work; the single Web of Science addition is graded A. Strengths common to the stronger studies are long, complete denominators from the HDSS, laboratory confirmation of cholera, and explicit control for season and trend with lagged exposures. Recurrent weaknesses are reliance on a single meteorological station roughly 10 km from the area, absence of individual exposure measurement, ecological designs in most of the cholera literature, inconsistent handling of the embankment as a confounder, and, before 2012, no account of changing vaccination coverage. Two of the included studies use Matlab only as one site among many or as a validation set, which limits what they say about Matlab itself. A further caution applies to the four studies added from before 2000: they report peak months and rate ratios without confidence intervals or adjustment for trend, so they establish that a seasonal pattern existed rather than how large it was.

4Discussion

4.1 Principal findings

Cold, not heat, is the mortality hazard so far recorded. Three independent time-series analyses agree, and the pregnancy and infant-mortality work points the same way — now including Ahmed et al. 2020, where birth weight is highest in the post-aman harvest and lowest in the monsoon. No study has yet looked for a heat effect on mortality in the warmer years after 2009.

Heat and warm seas drive infectious morbidity. Cholera, childhood diarrhoea and (on the same day) childhood pneumonia all rise with heat, and cholera additionally tracks ENSO, the Indian Ocean Dipole, Bay of Bengal temperature and chlorophyll with lags of weeks to months — though the earliest and longest cholera series, Glass et al. 1982, found no stable link to monsoon rains or river levels.

Flood protection has mixed effects. It improved child survival but moved cholera into the rainy season and concentrated cases behind the embankment, and Emch 1999 found it related to diarrhoea by a mechanism he could not explain; the 1988 and 2004 floods themselves produced little measurable excess in nutrition or mortality once properly controlled.

Air pollution research is thin, and entirely indoors. Six studies, all on household smoke: a mortality cohort, two pooled spirometry studies, a single-site spirometry survey, a birth-size analysis and one measurement study. Five infer exposure from cooking fuel; only Shahriar et al. 2021 measured concentrations, and its vascular outcome was null. Personal, kitchen and outdoor PM2.5 and black carbon were measured in 400 Matlab and Araihazar women in 2026, and daily rainfall and temperature have been shown to predict faecal contamination of Matlab tubewells — neither study reports a health outcome, so neither is included here, but together they remove the usual excuse for the gap.

4.2 Gaps and opportunities

  • Ambient air pollution. Matlab lies downwind of Dhaka and within the Indo-Gangetic winter haze belt, yet no study has linked PM2.5 or ozone to its mortality or hospital records. The exposure side is no longer the obstacle: Akhtar et al. 2026 measured 24-hour personal PM2.5 and black carbon in 400 women in rural Matlab and peri-urban Araihazar, with 72–120-hour PM2.5 in 200 kitchens and in the outdoor air beside them, and mapped personal exposure against distance to factories and main roads. That study, which reports no health outcome and is therefore not included in this review, gives a ground-truth layer against which satellite or reanalysis PM2.5 at 1-km resolution could be validated before being carried into the time-series designs already used here for temperature.
  • Heat–pollution interaction. With daily mortality from 1983, daily cholera and pneumonia admissions, a local weather station since 2008 and now measured particulate exposure, the dataset can support joint models of heatwaves and pollution; none exists. The 2026 measurements suggest the interaction is worth specifying rather than assuming: rural Matlab women carried higher personal and kitchen PM2.5 than peri-urban women, so a heat–pollution model here would be testing household smoke against heat, not traffic against heat.
  • Climate, water quality and diarrhoeal disease. This is now a decade-old line of work in Matlab with the health outcome still missing from it. Wu et al. 2016 sampled 125 shallow tubewells across six Matlab villages and Araihazar roughly monthly for 17 months and related E. coli to counts of hot and heavy-rain days over the preceding 3 to 30 days; Wu et al. 2025 repeated the exercise with machine learning on 1,495 Matlab samples and found the same rainfall and temperature windows among the strongest predictors of E. coli presence and concentration, and Broyles et al. 2023 found household water insecurity in 2,014 Matlab households significantly lower after the monsoon than before it. Neither reports a health outcome, so neither is included, but the two of them supply the missing middle term between weather and the diarrhoeal disease that this review’s strongest studies measure.
  • Updated temperature–mortality series. The mortality analyses end in 2009; extending them to the present, with a distributed-lag non-linear model and cause-specific verbal-autopsy outcomes, would show whether a heat arm has emerged.
  • Non-communicable disease. Cardiovascular, renal and mental-health outcomes under heat stress are unexamined, despite the epidemiological transition documented in the same population. The single NCD result in this review, Shahriar et al. 2021 on household pollutants and endothelial function, is null and says nothing about temperature.
  • Projections and attribution. No study combines Matlab exposure–response functions with climate scenarios to estimate future burden, and none attributes observed trends to anthropogenic warming.
  • Cyclones and groundwater. Cyclone mortality appears only as a covariate in one study, and shallow-aquifer flooding is not studied as a health exposure. Salinity is no longer wholly absent — Akanda et al. 2013 treats estuarine salinity intrusion as a driver of the spring cholera peak across six Bengal Delta sites including Matlab — but it remains a single multi-site ecological result, with no Matlab-specific salinity exposure–response function and nothing on salinity and blood pressure, pregnancy or renal outcomes, which is where the coastal Bangladesh literature has gone.

4.3 Limitations of this review

Limitations of this review673 wordsScreening, extraction and grading done once rather than in duplicate; no registered protocol and no meta-analysis; AI-assisted searching checked by hand; seven Web of Science records that never rendered and seven reference lists Europe PMC cannot serve.

Three databases were searched, and each pass changed the result. Scopus added eight studies to the original 30, a 27% increase, including the earliest cholera analysis and the largest season-and-birth-weight study; Web of Science then added one more, the only study linking household air pollution to a birth outcome. No single database retrieved more than two-thirds of the final 47: PubMed returned 30, Scopus 28 and Web of Science 23, and eight studies were retrieved by none of them. That is a warning about single-database searching on this site rather than a comment on any one index. The backward citation search is now complete as far as Europe PMC allows: reference lists were sought for all 47 included studies, 40 were screened, five hold no reference list, one holds references without titles and one is not indexed. Some studies may still remain — the Reviews on Environmental Health state-of-the-science review of environmental cholera transmission in India and Bangladesh (Shackleton et al. 2024), which screened 56 texts and is excluded here as a review, is the obvious starting point for a backward search.

Two incompletenesses should be stated plainly. Seven of the 65 Web of Science records could not be rendered from that platform’s results list and were never assessed, so the Web of Science screening is 58 of 65 rather than complete. The Google Scholar pass has now covered all eight themes and added five studies, every one of which names Bangladesh rather than Matlab in its title — precisely the class of study a field-tagged database string cannot see. One of them, Akanda et al. 2013, falsified a gap this review had asserted, that salinity intrusion was not studied as a health exposure. Against that, the final query probed the remaining 4.2 gaps directly, pairing the site phrase with cardiovascular, renal, mental-health and cyclone terms, and found nothing: those absences survive a search built to break them. The honest reading is that 46 is close to the ceiling that these methods can reach for this site. The backward search is finished and added nothing beyond what is already here; the Google Scholar pass is finished and its gap probes came back empty. What is left is narrow and nameable: seven Web of Science records that never rendered, and seven reference lists that Europe PMC cannot serve in a screenable form. The one record left unresolved at the previous revision, Alam et al. 2015, has since been resolved from a PubMed Central copy and included. Any claimed absence in 4.2 is worth only as much as the searching behind it, and that searching is now documented well enough for a reader to judge it.

The Scopus search traded recall for precision. Because Matlab collides with the MATLAB software package, the screened string required the site name within five words of Bangladesh, Chandpur, icddr,b, HDSS or surveillance; that string returned only 20 of the 30 studies already included, missing ten, mostly papers presenting Dhaka and Matlab together. A complete search would pool the broad and proximity strings and screen all 160 records. The two databases were also screened in sequence rather than pooled and de-duplicated as PRISMA intends, so the screened total of 147 double-counts an unknown number of records that appear in both.

Screening, extraction and grading were performed once rather than in duplicate, by an AI system with manual checking, and no protocol was registered. Eligibility was decided on abstracts except where a full text was needed to settle a case, which happened twice. Bibliographic identifiers for the eight studies added from Scopus were taken from the Scopus record and their DOIs were not separately verified. Effect estimates are reported as abstracted and have not been recalculated. The design grade is a crude proxy for risk of bias. The review should therefore be read as a mapped, PRISMA-structured scoping of the literature rather than as a fully independent systematic review, and users planning formal evidence synthesis should repeat the search in Web of Science and Google Scholar with dual screening.

5Conclusion

Across half a century, Matlab data have been used in 47 peer-reviewed studies of climate, weather and air pollution. The literature is rich on cholera and informative on cold-season mortality, flooding, the agro-climatic cycle in pregnancy and recent childhood pneumonia, but thin on everything else and still silent on ambient air pollution and on heat–pollution interactions as health questions. It is also, as Figure 4 shows, a literature that stops: nothing in it analyses a Matlab observation made after 2019. What changed between the two versions of this review is that the exposure data for those gaps now exist on the ground: measured personal, kitchen and outdoor PM2.5 in 400 Matlab and Araihazar women, and weather-driven models of faecal contamination in 1,495 Matlab tubewells, both published in the last two years and both lacking only a health outcome. Given the completeness of the mortality and hospital records, these gaps are tractable, and the existing Matlab time-series designs provide ready templates for filling them.

References

  1. Ai Milojevic, Ben Armstrong, Masahiro Hashizume, Katherine McAllister, Abu Syed Golam Faruque, Mohammad Yunus, Peter Kim Streatfield, Kazuhiko Moji, Paul Wilkinson. Health effects of flooding in rural Bangladesh. Epidemiology 2012;23(1):107-115. doi:10.1097/EDE.0b013e31823ac606
  2. Ali M, Kim DR, Yunus M, Emch M. Time series analysis of cholera in Matlab, Bangladesh, during 1988–2001. Journal of Health, Population and Nutrition 2013;31(1):11–19. doi:10.3329/jhpn.v31i1.14744
  3. Becker S, Weng S. Seasonal patterns of deaths in Matlab, Bangladesh. International Journal of Epidemiology 1998;27(5):814-823. doi:10.1093/ije/27.5.814
  4. Brooks W. Morgan, Matthew R. Grigsby, Trishul Siddharthan, Muhammad A. H. Chowdhury, Adolfo Rubinstein, Laura Gutierrez, Vilma Irazola, J. Jaime Miranda, Antonio Bernabe-Ortiz, Dewan Shamsul Alam, Robert A. Wise, William Checkley. Epidemiology and risk factors of asthma-chronic obstructive pulmonary disease overlap in low- and middle-income countries. Journal of Allergy and Clinical Immunology 2019;143(4):1598-1606. doi:10.1016/j.jaci.2018.06.052
  5. Cash BA, Rodó X, Emch M, Yunus M, Faruque ASG, Pascual M. Cholera and shigellosis: different epidemiology but similar responses to climate variability. PLoS ONE 2014;9(9):e107223. doi:10.1371/journal.pone.0107223
  6. Constantin de Magny G, Murtugudde R, Sapiano MRP, Nizam A, Brown CW, Busalacchi AJ, Yunus M, Nair GB, Gil AI, Lanata CF, Calkins J, Manna B, Rajendran K, Bhattacharya MK, Huq A, Sack RB, Colwell RR. Environmental signatures associated with cholera epidemics. Proceedings of the National Academy of Sciences of the United States of America 2008;105(46):17676-17681. doi:10.1073/pnas.0809654105
  7. Dewan Shamsul Alam, Muhammad Ashique Haider Chowdhury, Ali Tanweer Siddiquee, Shyfuddin Ahmed, Mohammad Didar Hossain, Sonia Pervin, Kim Streatfield, Alejandro Cravioto, Louis Wilhelmus Niessen. Adult Cardiopulmonary Mortality and Indoor Air Pollution: A 10-Year Retrospective Cohort Study in a Low-Income Rural Setting. Global Heart 2012;7(3):215. doi:10.1016/j.gheart.2012.06.008
  8. Emch M, Yunus M, Escamilla V, Feldacker C, Ali M. Local population and regional environmental drivers of cholera in Bangladesh. Environmental Health 2010;9:2. doi:10.1186/1476-069X-9-2
  9. Ernest O. Asare, Mohammad A. Al-Mamun, Monira Sarmin, A. S. G. Faruque, Tahmeed Ahmed, Virginia E. Pitzer. The influence of demographic and meteorological factors on temporal patterns of rotavirus infection in Dhaka, Bangladesh. Proceedings of the Royal Society B: Biological Sciences 2022;289(1976):20212727. doi:10.1098/rspb.2021.2727
  10. Harunor Rashid, Miki Kagami, Farzana Ferdous, Enbo Ma, Toru Terao, Taiichi Hayashi, Yukiko Wagatsuma. Temperature during pregnancy influences the fetal growth and birth size. Tropical Medicine and Health 2017;45:1. doi:10.1186/s41182-016-0041-6
  11. Hashizume M, Faruque AS, Terao T, Yunus M, Streatfield K, Yamamoto T, Moji K. The Indian Ocean Dipole and cholera incidence in Bangladesh: a time-series analysis. Environmental Health Perspectives 2011;119(2):239-244. doi:10.1289/ehp.1002302
  12. Hashizume M, Chaves LF, Faruque SM, Yunus M, Streatfield K, Moji K. A differential effect of Indian Ocean Dipole and El Niño on cholera dynamics in Bangladesh. PLoS ONE 2013;8(3):e60001. doi:10.1371/journal.pone.0060001
  13. Islam MS, Sharker MA, Rheman S, Hossain S, Mahmud ZH, Islam MS, Uddin AM, Yunus M, Osman MS, Ernst R, Rector I, Larson CP, Luby SP, Endtz HP, Cravioto A. Effects of local climate variability on transmission dynamics of cholera in Matlab, Bangladesh. Transactions of the Royal Society of Tropical Medicine and Hygiene 2009;103(11):1165-1170. doi:10.1016/j.trstmh.2009.04.016
  14. J. A. Myaux, M. Ali, J. Chakraborty, A. de Francisco. Flood control embankments contribute to the improvement of the health status of children in rural Bangladesh. Bulletin of the World Health Organization 1997;75(6):533-539. PubMed 9509625
  15. Junwen Tao, Mohammad Zahid Hossain, Zhiwei Xu, Hung Chak Ho, Md Alfazal Khan, Cunrui Huang, Hao Zheng, Jing Ni, Yinguan Fan, Daniel Bogale, Hong Su, Jian Cheng. Protective effect of pneumococcal conjugate vaccination on the short-term association between low temperatures and childhood pneumonia hospitalizations: Interrupted time-series and case-crossover analyses in Matlab, Bangladesh. Environmental Research 2022;212(Pt A):113156. doi:10.1016/j.envres.2022.113156
  16. Katia Koelle, Xavier Rodo, Mercedes Pascual, Md. Yunus, Golam Mostafa. Refractory periods and climate forcing in cholera dynamics. Nature 2005;436(7051):696-700. doi:10.1038/nature03820
  17. Maia A. Call, Clark L. Gray, Mohammad Yunus, Michael E. Emch. Disruption, not displacement: Environmental variability and temporary migration in Bangladesh. Global Environmental Change 2017;46:157-165. doi:10.1016/j.gloenvcha.2017.08.008
  18. Margaret Carrel, Michael Emch, Peter Kim Streatfield, Mohammad Yunus. Spatio-temporal clustering of cholera: the impact of flood control in Matlab, Bangladesh, 1983-2003. Health & Place 2009;15(3):771-782. doi:10.1016/j.healthplace.2008.12.008
  19. Margaret Carrel, Paul R. Voss, Peter Kim Streatfield, Mohammad Yunus, Michael Emch. Protection from annual flooding is correlated with increased cholera prevalence in Bangladesh: a zero-inflated regression analysis. Environmental Health 2010;9:13. doi:10.1186/1476-069X-9-13
  20. Masahiro Hashizume, Yukiko Wagatsuma, Taiichi Hayashi, Sajal Kumar Saha, Kim Streatfield, Mohammad Yunus. The effect of temperature on mortality in rural Bangladesh--a population-based time-series study. International Journal of Epidemiology 2009;38(6):1689-1697. doi:10.1093/ije/dyn376
  21. Mohammad Zahid Hossain, Shilu Tong, Md Alfazal Khan, Wenbiao Hu. Impact of climate variability on length of stay in hospital for childhood pneumonia in rural Bangladesh. Public Health 2020b;183:69-75. doi:10.1016/j.puhe.2020.03.014
  22. Mohammad Zahid Hossain, Shilu Tong, Hilary Bambrick, Al Fazal Khan, Samar Kumar Hore, Wenbiao Hu. Weather factors, PCV intervention and childhood pneumonia in rural Bangladesh. International Journal of Biometeorology 2020a;64(4):561-569. doi:10.1007/s00484-019-01842-7
  23. Nusrat F, Akanda AS, Islam A, Aziz S, Pakhtigian EL, Boyle K, Hanifi SMA. Satellite-derived, smartphone-delivered geospatial cholera risk information for vulnerable populations. GeoHealth 2024;8(11):e2024GH001039. doi:10.1029/2024GH001039
  24. Olufemi Babalola, Abdur Razzaque, David Bishai. Temperature extremes and infant mortality in Bangladesh: Hotter months, lower mortality. PLoS ONE 2018;13(1):e0189252. doi:10.1371/journal.pone.0189252. Correction: PLoS ONE 2019;14(5):e0216570 — Table 1 means and all six supplementary tables were wrong in the original and are restated in the correction. doi:10.1371/journal.pone.0216570
  25. Pradip K. Muhuri. Estimating seasonality effects on child mortality in Matlab, Bangladesh. Demography 1996;33(1):98-110. doi:10.2307/2061716
  26. Sophie E. Moore, Anthony J. C. Fulford, P. Kim Streatfield, Lars Ake Persson, Andrew M. Prentice. Comparative analysis of patterns of survival by season of birth in rural Bangladeshi and Gambian populations. International Journal of Epidemiology 2004;33(1):137-143. doi:10.1093/ije/dyh007
  27. Stan Becker. Seasonality of deaths in Matlab, Bangladesh. International Journal of Epidemiology 1981;10(3):271-280. doi:10.1093/ije/10.3.271
  28. Wietze Lindeboom, Nurul Alam, Dilruba Begum, Peter Kim Streatfield. The association of meteorological factors and mortality in rural Bangladesh, 1983-2009. Global Health Action 2012;5:19063. doi:10.3402/gha.v5i0.19063
  29. Wu J, Yunus M, Streatfield PK, Emch M. Association of climate variability and childhood diarrhoeal disease in rural Bangladesh, 2000–2006. Epidemiology and Infection 2014;142(9):1859–1868. doi:10.1017/S095026881300277X
  30. Wu J, Yunus M, Ali M, Escamilla V, Emch M. Influences of heatwave, rainfall, and tree cover on cholera in Bangladesh. Environment International 2018;120:304–311. doi:10.1016/j.envint.2018.08.012
  31. Ahmed SI, Tafsir Hasan SM, Khan MA, Ahmed T. Effect of Maternal Exposure to Seasons during the Second and Third Trimesters of Pregnancy on Infant Birth Weight in Rural Bangladesh. Current Developments in Nutrition 2020;4(3):nzaa016.
  32. Nyma Z, Hasan SMT, Khan MTF, Saqeeb KN, Ahmed T, et al. Effects of maternal exposure to biomass cooking fuel on birth size and body proportionality in full-term infants born by vaginal delivery. Scientific Reports 2024;14(1).
  33. Ruiz-Moreno D, Pascual M, Yunus M, et al. Spatial clustering in the spatio-temporal dynamics of endemic cholera. BMC Infectious Diseases 2010;10:51. doi:10.1186/1471-2334-10-51 Excluded on full text: no climatic exposure analysed; see Appendix C.
  34. Alam DS, Chowdhury MAH, Siddiquee AT, Ahmed S, Clemens JD. Prevalence and determinants of chronic obstructive pulmonary disease (COPD) in Bangladesh. COPD: Journal of Chronic Obstructive Pulmonary Disease 2015;12(6):658–667. doi:10.3109/15412555.2015.1041101
  35. Akhtar E, Haq MA, Hossain S, et al., Yunus M, Raqib R. Air Pollution Exposures of Bangladeshi Women from Rural and Peri-Urban Areas: Baseline Assessment for Behavior Change Communication Intervention as a Sustainable Approach. Sustainability 2026;18(7):3507. Cited in sections 3.3, 4.1, 4.2 and 5; not an included study (no health outcome).
  36. Broyles LMT, Pakhtigian EL, Aziz S, Akanda AS, Mejia A. Seasonal variation in household water insecurity in rural Bangladesh: A longitudinal analysis. PLOS Water 2023;2(7):e0000157. Cited in section 4.2; not an included study (outcome is water insecurity, not health).
  37. Emch M. Diarrheal disease risk in Matlab, Bangladesh. Social Science & Medicine 1999;49(4):519–530.
  38. Glass RI, Becker S, Huq MI, et al., Lee JV, Black RE. Endemic cholera in rural Bangladesh, 1966–1980. American Journal of Epidemiology 1982;116(6):959–970.
  39. Iqbal A, Shirin T, Ahmed T, et al., Sobhan A, Siddique AK. Childhood mortality due to drowning in rural Matlab of Bangladesh: magnitude of the problem and proposed solutions. Journal of Health, Population and Nutrition 2007;25(3):370–376.
  40. Islam MS, Islam MS, Mahmuda ZH, et al., Clemens JD, Collins AE. Role of phytoplankton in maintaining endemicity and seasonality of cholera in Bangladesh. Transactions of the Royal Society of Tropical Medicine and Hygiene 2015;109(9):572–578.
  41. Rahman A, Rahman M, Pervin J, et al., El Arifeen S, Persson LÅ. Time trends and sociodemographic determinants of preterm births in pregnancy cohorts in Matlab, Bangladesh, 1990–2014. BMJ Global Health 2019;4(4):e001462.
  42. Shaikh K, Wojtyniak B, Mostafa G, Khan MU. Pattern of diarrhoeal deaths during 1966–1987 in a demographic surveillance area in rural Bangladesh. Journal of Diarrhoeal Diseases Research 1990;8(4):147–154.
  43. Stewart MK, Fauveau V, Chakraborty J, et al., Yunus M, Sarder AM. Post-flood nutritional anthropometry of children in Matlab, Bangladesh. Ecology of Food and Nutrition 1990;24(2):121–131.
  44. Hasan SMT, Rahman S, Locks LM, Khan MA, Ahmed T. Magnitude and determinants of inadequate third-trimester weight gain in rural Bangladesh. PLoS ONE 2018;13(4):e0196190. doi:10.1371/journal.pone.0196190
  45. Akanda AS, Jutla AS, Gute DM, et al. Population vulnerability to biannual cholera outbreaks and associated macro-scale drivers in the Bengal Delta. American Journal of Tropical Medicine and Hygiene 2013;89(5):950–959. PubMed 24019441
  46. Das SK, Begum D, Ahmed S, Ferdous F, et al. Geographical diversity in seasonality of major diarrhoeal pathogens in Bangladesh observed between 2010 and 2012. Epidemiology and Infection 2014;142(12):2530–2541. PubMed 24534384
  47. Emch M, Feldacker C, Yunus M, Streatfield PK, Dinh Thiem V, Canh DG, Ali M. Local environmental predictors of cholera in Bangladesh and Vietnam. American Journal of Tropical Medicine and Hygiene 2008;78(5):823–832. PubMed 18458320
  48. Koelle K, Pascual M, Yunus M. Pathogen adaptation to seasonal forcing and climate change. Proceedings of the Royal Society B 2005;272(1566):971–977. PubMed 16024354
  49. Shahriar MH, Chowdhury MAH, Ahmed S, et al. Exposure to household air pollutants and endothelial dysfunction in rural Bangladesh: a cross-sectional study. Environmental Epidemiology 2021;5(2):e132. PubMed 33870008
  50. Siddharthan T, Grigsby MR, Goodman D, et al. Association between household air pollution exposure and chronic obstructive pulmonary disease outcomes in 13 low- and middle-income country settings. American Journal of Respiratory and Critical Care Medicine 2018;197(5):611–620. PubMed 29323928
  51. Wu J, Yunus M, Islam MS, Emch M. Influence of climate extremes and land use on fecal contamination of shallow tubewells in Bangladesh. Environmental Science & Technology 2016;50(5):2669–2676. doi:10.1021/acs.est.5b05193 Cited in section 4.2; not an included study (no human health outcome).
  52. Wu J, Cao Y, Islam MS, Emch M. Application of Machine Learning to Identify Influential Factors for Fecal Contamination of Shallow Groundwater. Water 2025;17(2):160. Cited in sections 4.1, 4.2 and 5; not an included study (no human health outcome).

Entries added in the Scopus update are listed with the bibliographic details as Scopus records them; their DOIs were not separately verified (section 4.3).

Appendix A. Search strategy

Show Appendix A279 wordsEvery search string and its record count, for all three databases.

PubMed, searched 16 October 2026, no date or language limits, sorted by date. Record counts as returned.

#Search stringRecords
0Matlab[Title/Abstract] AND Bangladesh[Title/Abstract]667
1#0 AND climate (automatic term mapping to the Climate/Weather MeSH tree)60
2#0 AND temperature24
3#0 AND flood17
4#0 AND weather15
5#0 AND rainfall8
6#0 AND air pollution2
Total / unique after de-duplication126 / 78

Scopus, searched 16 October 2026, no date or language limits. Search 8 is the string screened; search 7 is reported for transparency and was not screened.

#Search stringRecords
Eclimat* OR temperature OR weather OR rainfall OR precipitation OR flood* OR cyclone* OR heat OR heatwave OR "air pollution" OR "particulate matter" OR "pm2.5" OR salinity OR season* OR monsoon OR "sea surface" OR enso OR "el nino" OR drought OR humidity (the exposure block, used inside TITLE-ABS-KEY)—
7TITLE-ABS-KEY(matlab AND bangladesh) AND TITLE-ABS-KEY(#E) — not screened; dominated by MATLAB-software papers in energy and engineering160
8TITLE-ABS-KEY((matlab W/5 bangladesh) OR (matlab W/5 chandpur) OR (matlab W/5 icddr) OR (matlab W/5 hdss) OR (matlab W/5 surveillance)) AND TITLE-ABS-KEY(#E) — the string screened89
Of which already included from PubMed / newly screened20 / 69

Web of Science Core Collection, all editions, searched 16 October 2026, no date or language limits.

#Search stringRecords
9TS=(((Matlab NEAR/5 Bangladesh) OR (Matlab NEAR/5 Chandpur) OR (Matlab NEAR/5 icddr) OR (Matlab NEAR/5 HDSS) OR (Matlab NEAR/5 surveillance)) AND (#E))65
Of which already included / assessed / not rendered from the results list22 / 36 / 7

Appendix B. PRISMA 2020 checklist (condensed)

Show Appendix B193 wordsEach PRISMA 2020 item and where this review addresses it.
ItemWhere addressed
1 TitleIdentified as a systematic review in the title
2 AbstractStructured abstract (background, methods, results, conclusions)
3–4 Rationale and objectivesSection 1
5 Eligibility criteriaSection 2.2
6–7 Information sources and search strategySection 2.3; Appendix A
8 Selection processSection 2.4 (single reviewer; stated as a limitation)
9–10 Data collection and itemsSection 2.5; Table 1
11 Risk of biasSection 2.6 design grading (no validated tool available for this mix of designs)
12–13 Effect measures and synthesis methodsSection 2.6; narrative synthesis, no meta-analysis
14–15 Reporting bias and certainty assessmentNot performed; stated in Section 4.3
16 Study selection resultsSection 3.1; Figure 2
17–18 Study characteristics and risk of biasTable 1; Section 3.4
19–20 Results of individual studies and synthesesTable 1; Section 3.3
21–22 Reporting biases and certaintyNot assessed; Section 4.3
23 DiscussionSection 4
24 Registration and protocolNone; Section 2.1
25–26 Support and competing interestsNo funding; no competing interests declared
27 Availability of dataExtracted data in Table 1 and Appendix D; search strings in Appendix A

Appendix C. Records resolved in the extended search

Show Appendix C1,847 wordsEvery candidate the citation and Google Scholar passes surfaced, and why each was included or excluded.

The backward citation search and the Google Scholar pass each surfaced candidates needing individual assessment. This appendix records them and the reason each was included or excluded, so the judgements can be audited rather than taken on trust. It is the long form of section 2.7.

C.1 Backward citation search

After the three database searches, the reference lists of seven sources were screened for Matlab studies the database strings had missed: the Reviews on Environmental Health state-of-the-science review of environmental cholera transmission in India and Bangladesh (Shackleton et al. 2024, 56 texts), and six included studies chosen to span the themes and the date range — Wu 2018, Lindeboom 2012, Tao 2022, Rashid 2017, Alam 2012 and Nusrat 2024. Reference lists were read through Europe PMC. Two records that the abstract screening had left unresolved were settled from full text.

The pass was later extended to every included study whose reference list Europe PMC serves — 40 lists in all — and was accompanied by the themed Google Scholar search described in C.2; between them the two passes produced the eight studies shown on the right-hand arm of Figure 2. It also produced three things worth recording about the searching itself. First, it confirmed a blind spot in the database strings rather than in the literature: all three searches required Matlab in the title, abstract or keywords, so a Matlab study titled “… in rural Bangladesh” is only caught if the site is named in its abstract. The two candidates this surfaced — Hashizume et al. 2008 on rainfall and cholera and Hashizume et al. 2010 on the climatic components of cholera seasonality, both in Epidemiology — turned out to use Dhaka hospital data, not Matlab, and are correctly excluded; but they are the shape of study the strings would have missed had they been Matlab. Second, it added Wu et al. 2016 to the climate-and-water-quality line discussed in 4.2. Third, it resolved the two open records: Ruiz-Moreno et al. 2010 and Hossain et al. 2020b.

Ruiz-Moreno et al. 2010 is a Matlab study of 7,241 cholera cases and 8,340 baris over 1983–2004, and it is excluded. The full text confirms the analysis is Ripley’s K and L functions with Monte Carlo envelopes on case and water-source locations; rivers and canals were explicitly left out, and the only seasonal element is a rescaling of each epidemic onto a common phase, not a climatic covariate. Temperature, monsoon and ENSO appear in the introduction and discussion but in no model. It fails the exposure criterion on the authors’ own description.

Hossain et al. 2020b was retained in the first version of this review on its title and companion papers because its abstract could not be retrieved. It has now been read and is reported in full in Table 1, with a design grade. No included study is now ungraded.

Completing the remaining lists added no further study. Six candidates surfaced and all six failed on site or on exposure, and they are listed here because each is the kind of paper a reader might expect to find in this review. Brown, Black and Becker 1982 tracks seasonal change in child nutritional status over 15 months but in “two rural villages of Bangladesh” that it does not name. Siddique et al. 1991 reports illness and causes of death across 72 flood-affected upazilas after the 1988 flood, a national survey rather than a Matlab one. Stine et al. 2008 carries “seasonal cholera” in its title but analyses genetic relatedness of isolates from two unnamed rural locales, with no climatic variable at all. Alam et al. 2011 works in Dhaka and coastal villages and mentions Matlab only as the source of a comparison strain lineage. Sack et al. 2003 covers four unnamed locations and states that its environmental correlations were planned rather than reported. Begum et al. 1993 on the geographical and seasonal distribution of diarrhoeal disease could not be retrieved through Europe PMC and is a national distribution study by title.

One record from this pass was initially excluded pending full text and has since been resolved, which is worth setting out because it shows how thin the margin can be. Alam et al. 2015 (COPD 12(6):658–667) reports solid-fuel use as a determinant of COPD in 3,744 adults; its abstract places it only in “rural and urban areas in Bangladesh”, and the publisher’s site is subscription-only. A PubMed Central copy was eventually located, and its methods are unambiguous: the sample was drawn “in rural Matlab (Chandpur district) and semi-urban Kamlapur (Dhaka district)”, with the icddr,b Matlab demographic surveillance system as the sampling frame and fieldwork between February 2011 and March 2012. It is included, and takes the household air pollution theme to six. Had the search stopped at the abstract, as the first version of this review did for Hossain 2020b, it would have been lost. Because it joined the review after the citation pass had run, its own reference list was screened last, the fortieth: 27 references, all titled, of which the only Matlab study is Alam et al. 2012, already included. It adds nothing, which closes the pass at 47.

C.2 Themed Google Scholar pass

Because Google Scholar indexes full text rather than title, abstract and keywords, it reaches exactly the studies the three database strings were built to miss: those that name Matlab only in their methods. The pass therefore searched the exact phrase "Matlab, Bangladesh" against theme-specific terms, one query per exposure–outcome theme, rather than repeating the field-tagged strings of 2.3.

All eight themes were searched: household air pollution; temperature and mortality; flooding, embankments, cyclones and salinity; cholera and climate; diarrhoea, rotavirus, pneumonia and respiratory infection in one combined query; heat and season in pregnancy with climate-linked migration; and seasonality of mortality, with the last query doubling as a probe for the gaps claimed in 4.2 by pairing the site phrase with cardiovascular, blood-pressure, mental-health, depression, renal, kidney and cyclone terms. Google Scholar interrupted the pass with a CAPTCHA after the sixth query, which was not answered; the remaining two were run after the challenge was cleared.

The gap probe is worth reporting as a result rather than as a procedure. Searching the full text of the Matlab literature for heat or season against cardiovascular, renal, mental-health or cyclone outcomes returned nothing that met the criteria — only the temperature–mortality studies already included, which report cardiovascular deaths as a cause stratum rather than as an exposure–response in their own right, two national reviews, and the icddr,b surveillance reports. The absence claimed in 4.2 therefore survives a search designed to break it, which is a stronger statement than the earlier version of this review could make.

The pass added four studies, all of them papers whose titles say “Bangladesh” and not “Matlab”. Shahriar et al. 2021 and Siddharthan et al. 2018 are described in C.3. Akanda et al. 2013 uses Matlab cholera records for 1998–2007 as one of six Bengal Delta series and is the only study in this review to analyse estuarine salinity as a health exposure; the gap statement in 4.2 has been corrected accordingly. Das et al. 2014 models monthly counts of four diarrhoeal pathogens across four icddr,b sites including Matlab, with seasonality as the object of the analysis, and is admitted on the same basis as the older descriptive seasonality studies — its authors state explicitly that no meteorological variable was tested. Hasan et al. 2018 is the fifth, and the closest call in this review: season of conception is one determinant among several in a model of third-trimester weight gain rather than the object of the analysis, but it is entered as an explicit climatic category, monsoon or dry season against summer, and carries an effect estimate, which is the basis on which it is admitted.

Four candidates from the pass were resolved as exclusions on their stated data source. Alam et al. 2012, the sister paper to Lindeboom in the same journal and year, analyses weather and mortality in Abhoynagar, not Matlab. Burkart et al. 2011 and 2014 draw on the Bangladesh Bureau of Statistics Sample Vital Registration System rather than any HDSS, so their heat and seasonality results, which bear directly on this review’s central claim, cannot be read as Matlab findings. Two spatial-epidemiology papers by Ali and Emch were excluded on exposure, as in C.3.

C.3 What the two passes added

The backward search was subsequently completed. Reference lists were sought for all 47 included studies through Europe PMC, plus the Shackleton review. Thirty-nine yielded screenable titles and were screened. Five hold no reference list at all, all of them pre-2005 (Glass 1982, Becker 1981, Shaikh 1990, Becker and Weng 1998, Moore 2004). One, Emch 2008, holds references without titles, so it could not be screened there. One, Stewart 1990, is not indexed in Europe PMC at all. The screened lists returned, overwhelmingly, studies already included. Two did not. Emch et al. 2008, cited by Cash 2014 and by Emch 2010, carries a title naming only “Bangladesh and Vietnam”; its full text states that “cholera incidence time series and contemporaneous environmental variables are compiled for Matlab, Bangladesh… from 1983 to 2003”, with ocean chlorophyll, sea-surface temperature and height, Chandpur rainfall and temperature, and river discharge at Demra and Bhairab. Koelle, Pascual and Yunus 2005, the theoretical companion to the Nature paper, draws its monthly reproductive rates from the Matlab series for 1966–2002 and its rainfall from Chandpur. Both are now included.

Three further candidates from the same cluster were resolved as exclusions on full text or on their stated data source. Pascual et al. 2000 (Science) and Rodó et al. 2002 (PNAS) use the icddr,b Dhaka hospital series, the latter confirmed from its full text; Koelle and Pascual 2004 (American Naturalist) applies its method to historical Bengal mortality for 1891–1940. Lobitz et al. 2000 names only the icddr,b hospital surveillance programme, with Matlab nowhere in the paper. A separate group of Matlab studies — Ali et al. 2001 and 2002, Ruiz-Moreno et al. 2010 — are excluded not on site but on exposure: their environmental variables are static geography, proximity to surface water or pond location, rather than weather or a hydrological event.

The Google Scholar pass was aimed at the blind spot the reference lists had exposed, searching the phrase “Matlab, Bangladesh” against air-pollution terms rather than relying on the databases’ field-tagged strings. It returned two included studies on its first query, both in the thinnest theme in this review. Shahriar et al. 2021 measured 48-hour personal PM2.5, black carbon and carbon monoxide in 199 Matlab and Araihazar women and tested them against endothelial function, finding nothing. Siddharthan et al. 2018 pooled 13 settings including Matlab, which had 98.1% biomass use and the highest COPD prevalence of the thirteen. That a single Scholar query yielded two eligible studies is the strongest evidence in this review that the search is not yet exhausted, and it is reported as such in 4.3.

Appendix D. Included studies with abstract summaries

Show Appendix D11,448 wordsPlain-language summary of each of the 47 included studies.

Summaries are rewritten in plain words from the published abstracts with the quantitative findings kept; follow the DOI or PubMed link for the authors’ own text. Newest first within each theme.

Household air pollution 6

2015COPD: Journal of Chronic Obstructive Pulmonary Disease

Prevalence and Determinants of Chronic Obstructive Pulmonary Disease (COPD) in Bangladesh

Dewan S. Alam, Muhammad A. H. Chowdhury, Ali T. Siddiquee, Shyfuddin Ahmed, John D. Clemens. COPD 2015;12(6):658–667.

resolved from full text after being flagged unresolved; see Appendix C

Population-based data on COPD in Bangladesh were scarce, and this study set out to measure its prevalence and determinants directly. Between February 2011 and March 2012 the authors drew a random sample of 3,744 adults aged 40 and over from the icddr,b surveillance database, across rural Matlab in Chandpur district and semi-urban Kamlapur in Dhaka district, and measured lung function by handheld spirometry with bronchodilator testing, classifying COPD by both GOLD criteria and the lower limit of normal. Prevalence was 13.5% by GOLD and 10.3% by LLN, higher in rural residents than urban and higher in men; more than half of cases were stage II, and the condition was largely undiagnosed. In the adjusted model, household solid fuel use was associated with COPD at an odds ratio of 1.5 (95% CI 1.2–1.9) under GOLD criteria and 1.3 (1.0–1.8) under LLN, alongside older age, male sex, illiteracy, being underweight, current or former smoking and a history of asthma. The authors name illiteracy, smoking and biomass fuel burning as the modifiable determinants.

Matlab data The icddr,b Matlab demographic surveillance system, running since 1966, provided the sampling frame for the rural half of the sample; the spirometry and interview data were collected for this study.

PubMed 26263031 · DOI 10.3109/15412555.2015.1041101

2021Environmental Epidemiology

Exposure to household air pollutants and endothelial dysfunction in rural Bangladesh: a cross-sectional study

Mohammad Hasan Shahriar, Muhammad Ashique Haider Chowdhury, Shyfuddin Ahmed and colleagues. Environmental Epidemiology 2021;5(2):e132.

added by the themed Google Scholar pass; see Appendix C

This is the only study in the review to measure pollutant concentrations rather than infer exposure from fuel type, and the only one with a vascular outcome. Implemented in the Matlab and Araihazar subdistricts as part of the Bangladesh GEO Health project, it enrolled 199 healthy non-smoking women who cooked with biomass. Personal exposure was measured over 48 hours, twice a year at least six months apart, once in the dry season and once in the wet: PM2.5 and black carbon on RTI MicroPEM monitors, PM2.5 determined gravimetrically from pre- and post-weighed filters and black carbon from the same filters by smoke-stain reflectometer, with carbon monoxide on Lascar loggers. Endothelial function was measured as the reactive hyperaemia index by peripheral arterial tonometry. After adjustment no association was found with any pollutant: PM2.5 odds ratio 0.97 (95% CI 0.92–1.01), black carbon 0.85 (0.72–1.01), carbon monoxide 0.89 (0.64–1.25). The authors conclude that household air pollution was not associated with endothelial dysfunction among non-smoking women in this population.

Matlab data Matlab was one of two subdistricts in which the study collected its own data; the icddr,b Matlab field site, covering about 225,826 people in 142 villages, is the setting, although HDSS records are not named as the sampling frame.

PubMed 33870008

2018American Journal of Respiratory and Critical Care Medicine

Association between household air pollution exposure and chronic obstructive pulmonary disease outcomes in 13 low- and middle-income country settings

Trishul Siddharthan, Matthew R. Grigsby, Dina Goodman and colleagues. American Journal of Respiratory and Critical Care Medicine 2018;197(5):611–620.

added by the themed Google Scholar pass; see Appendix C

The companion to Morgan and others from the same consortium, this pooled analysis covered 12,396 adults aged 35 to 95 across 13 population-based settings, Matlab among them, with household air pollution defined as self-reported use of biomass as the primary household fuel. Overall COPD prevalence was 8.8%, and exposure carried an adjusted odds ratio of 1.41 (95% CI 1.18–1.68), stronger in women (1.70) than men (1.21); the authors attribute 13.5% of COPD to household air pollution, slightly more than the 12.4% they attribute to daily cigarette smoking. The Matlab site is the striking one in the table: 1,824 participants, 98.1% of them using biomass as the primary fuel, and a COPD prevalence of 15.3%, the highest of any of the 13 settings. The authors note they could not measure biomass exposure beyond questionnaires.

Matlab data Matlab contributed 1,824 participants and recorded the highest COPD prevalence of the 13 sites; the Bangladesh component drew on icddr,b data collected in 2011–2012.

PubMed 29323928

2024Scientific Reports

Effects of maternal exposure to biomass cooking fuel on birth size and body proportionality in full-term infants born by vaginal delivery

Zannatun Nyma, S. M. Tafsir Hasan, Md Tareq Ferdous Khan, Kazi Nazmus Saqeeb, Tahmeed Ahmed and colleagues. Scientific Reports 2024;14(1).

added by the Web of Science search; bibliographic details from the Web of Science record, DOI not separately verified

It has remained unclear whether and how maternal exposure to biomass fuel influences infant anthropometry or body proportionality at birth, both of which bear on survival, physical growth and neurodevelopment. The authors drew 909 women from the Pregnancy Weight Gain study, conducted in Matlab, and examined household-level exposure to biomass cooking fuel against infant weight, length and head circumference, small-for-gestational-age by weight, length and head circumference, and the ponderal and cephalization indices at birth. Of the women, 721 (79.3%) depended on biomass fuel. Compared with infants born to mothers who cooked with gas, infants of biomass users were lighter (beta −94.3 g, 95% CI −155.9 to −32.6) and shorter, with body proportionality also affected. This is the only study in the review to carry an air-pollution exposure through to a birth outcome in Matlab, and the only household air pollution study of the three to use an individual-level adjusted design.

Matlab data The Pregnancy Weight Gain study cohort, conducted in Matlab, supplied the 909 mother–infant pairs, their household cooking-fuel type and the birth anthropometry.

Web of Science record

2019Journal of Allergy and Clinical Immunology

Epidemiology and risk factors of asthma-chronic obstructive pulmonary disease overlap in low- and middle-income countries

Brooks W. Morgan, Matthew R. Grigsby, Trishul Siddharthan, Muhammad A. H. Chowdhury, Adolfo Rubinstein, Laura Gutierrez, Vilma Irazola, J. Jaime Miranda, Antonio Bernabe-Ortiz, Dewan Shamsul Alam, Robert A. Wise, William Checkley. Journal of Allergy and Clinical Immunology 2019;143(4):1598-1606.

multi-country pooled study; Matlab is one of 12 sites

The authors pooled cross-sectional spirometry and questionnaire data on 11,923 adults aged 35-92 from four population-based studies spanning 12 sites in six low- and middle-income countries, one of which was Matlab, Bangladesh. The environmental exposure examined was household air pollution from burning biomass fuel, alongside tobacco smoking and low educational attainment. The outcome was asthma-COPD overlap (ACO), defined as post-bronchodilator FEV1/FVC below the lower limit of normal together with wheeze, asthma medication use, or a physician diagnosis of asthma. Prevalence was estimated by site and multivariable regression was used to compute odds ratios and compare lung-function z-scores across disease groups. Overall ACO prevalence was 3.8%, ranging from 0% in rural Puno, Peru, to 7.8% in Matlab; biomass smoke exposure (OR 1.48, 95% CI 0.98-2.23), each 10 pack-years of smoking (OR 1.28, 95% CI 1.22-1.34), and primary-or-less schooling (OR 1.35, 95% CI 1.07-1.70) were linked to higher odds of ACO. People with ACO had more severe obstruction (FEV1 below 50% predicted in 31.6% versus 10.9% of COPD-only cases) and larger FEV1 deficits (-1.61 z versus asthma alone; -0.94 z versus COPD alone), leading the authors to flag biomass smoke as an under-recognized contributor to ACO in these settings.

Matlab data Matlab, Bangladesh (icddr,b HDSS site) contributed one of the 12 population-based cohorts in the pooled analysis and recorded the highest ACO prevalence (7.8%).

PubMed 30291842 · DOI 10.1016/j.jaci.2018.06.052

2012Global Heart

Adult Cardiopulmonary Mortality and Indoor Air Pollution: A 10-Year Retrospective Cohort Study in a Low-Income Rural Setting

Dewan Shamsul Alam, Muhammad Ashique Haider Chowdhury, Ali Tanweer Siddiquee, Shyfuddin Ahmed, Mohammad Didar Hossain, Sonia Pervin, Kim Streatfield, Alejandro Cravioto, Louis Wilhelmus Niessen. Global Heart 2012;7(3):215.

This retrospective cohort study followed adults in the Matlab surveillance area of rural Bangladesh over a ten-year window, with deaths and their causes established through verbal autopsy. The exposure of interest was indoor air pollution, approximated by the type of fuel a household used for cooking: solid biomass fuels versus piped natural gas. The outcome was death from cardiovascular or respiratory causes among adults. The authors computed cause-specific mortality rates per 1,000 person-years in each fuel group and compared them using incidence rate ratios. Combined cardiopulmonary mortality was 6.2 per 1,000 person-years in solid-fuel households against 5.3 in gas-using households, a gap that did not reach statistical significance, whereas respiratory deaths alone were more than twice as frequent in solid-fuel homes (IRR 2.26, 95% CI 1.02-4.99). The authors infer that cutting household smoke exposure would probably lengthen adult survival in such settings.

Matlab data Adult cohort, person-time, and verbal-autopsy cause-of-death data drawn from the icddr,b Matlab HDSS over a 10-year period, with household cooking-fuel type from the surveillance census as the exposure.

PubMed 25691484 · DOI 10.1016/j.gheart.2012.06.008

Temperature, weather and mortality 3

2018PLoS ONE

Temperature extremes and infant mortality in Bangladesh: Hotter months, lower mortality

Olufemi Babalola, Abdur Razzaque, David Bishai. PLoS ONE 2018;13(1):e0189252.

This paper examines whether unusually hot or cold months influence infant deaths in rural Matlab, Bangladesh, using monthly surveillance data from January 1982 through December 2008 that encompass 49,426 infant deaths. The exposures were seasonally adjusted monthly mean and maximum temperatures, tested at same-month and one-month lags. The outcome was the infant mortality rate, further split by sex and by neonatal versus post-neonatal age. The authors fitted regression models with ARIMA error terms to handle autocorrelation in the monthly series. Contrary to the expectation that heat harms infants, warmer months were associated with fewer deaths: a 1 degree C increase in mean monthly temperature corresponded to 3.672 fewer infant deaths per unit of the mortality rate (SE 1.544, p<0.05), with a stronger effect among neonates than older infants. The study finds no sign of heat-related excess infant mortality in this setting and suggests that susceptibility to cold (hypothermia) may dominate in the first weeks of life.

Matlab data Monthly infant deaths and births (1982-2008) from the icddr,b Matlab HDSS, analysed against monthly temperature series for the area.

PubMed 29304145 · DOI 10.1371/journal.pone.0189252

2012Global Health Action

The association of meteorological factors and mortality in rural Bangladesh, 1983-2009

Wietze Lindeboom, Nurul Alam, Dilruba Begum, Peter Kim Streatfield. Global Health Action 2012;5:19063.

Using 27 years of daily death records (1983-2009) from the Matlab health and demographic surveillance site, the authors investigated how short-term weather conditions relate to mortality in rural Bangladesh. Exposures were daily temperature and rainfall from a nearby government meteorological station, plus the occurrence of cyclone episodes. All-cause daily mortality, stratified by age and sex, was the outcome. The team fitted Poisson regression models with cubic splines to capture non-linear and lagged temperature effects while adjusting for season and secular trend, and ran a separate multivariate analysis for cyclone periods. Temperature showed a U-shaped relationship with death counts, with the minimum around 29 degrees C; the cold arm was most pronounced in older adults, whose mortality climbed roughly 5.4% for each 1 degree C fall below 23 degrees C. Rainfall effects were slight overall but among 5-19-year-olds mortality rose 0.6% per extra millimetre of daily rain, becoming significant beyond 100 mm/day, and women aged 20-59 faced elevated risk during cyclones. The authors conclude that vulnerability to weather varies by age and sex and should guide targeted interventions.

Matlab data Daily mortality by age and sex from the Matlab HDSS, 1983-2009, linked to temperature and rainfall from a nearby Bangladesh Meteorological Department station and to cyclone dates.

PubMed 23195514 · DOI 10.3402/gha.v5i0.19063

2009International Journal of Epidemiology

The effect of temperature on mortality in rural Bangladesh--a population-based time-series study

Masahiro Hashizume, Yukiko Wagatsuma, Taiichi Hayashi, Sajal Kumar Saha, Kim Streatfield, Mohammad Yunus. International Journal of Epidemiology 2009;38(6):1689-1697.

This study drew on 13,270 non-external-cause deaths recorded by the Matlab demographic surveillance system between January 1994 and December 2002 to ask whether temperature affects mortality in a rural, low-income population, a question previously studied almost exclusively in cities. Daily mean temperature was the exposure, and deaths from all natural causes as well as cardiovascular, respiratory, and perinatal subgroups were the outcomes. The analysis used generalized linear Poisson regression on the daily time series, adjusting for season, long-term trend, day of week, and public holidays, and examining lags of up to 13 days. Cold was clearly harmful: every 1 degree C drop in mean temperature raised all-cause mortality by 3.2% (95% CI 0.9-5.5), with the steepest increases for cardiovascular, respiratory, and perinatal deaths. In contrast, the authors found no discernible excess mortality attributable to heat at any lag. They recommend cold-season protective measures, particularly for newborns.

Matlab data Daily all-cause and cause-specific mortality counts from the Matlab HDSS (1994-2002) analysed against local daily temperature records.

PubMed 19181749 · DOI 10.1093/ije/dyn376

Flooding, flood control and salinity 7

2013American Journal of Tropical Medicine and Hygiene

Population vulnerability to biannual cholera outbreaks and associated macro-scale drivers in the Bengal Delta

Ali Shafqat Akanda, Antarpreet S. Jutla, David M. Gute and colleagues. American Journal of Tropical Medicine and Hygiene 2013;89(5):950–959.

added by the themed Google Scholar pass; see Appendix C

Cholera in the Bengal Delta comes in two annual peaks, and this study asks whether the two have different macro-scale causes. The authors assembled cholera incidence from six surveillance sites — Dhaka, Matlab (1998–2007), Bakerganj, Chhatak, Mathbaria and Kolkata — and set them against estuarine salinity, inland flood inundation, and the seasonal low and high flow of the Ganges and Brahmaputra. The spring peak tracks the low-flow season, when reduced river discharge allows saline water to intrude up the estuary; the autumn peak tracks the opposite condition, inland flood inundation after the monsoon. The authors argue that vulnerability therefore differs by district and by season, and that a single national cholera forecasting rule will not serve. For this review it is the only study to treat salinity intrusion as a health exposure, and one of the few to use river flow directly rather than rainfall as a proxy for it.

Matlab data Matlab cholera incidence records for 1998–2007 were one of six site series; the authors note that much of the earlier climate–cholera literature relied on Matlab alone.

PubMed 24019441

2012Epidemiology

Health effects of flooding in rural Bangladesh

Ai Milojevic, Ben Armstrong, Masahiro Hashizume, Katherine McAllister, Abu Syed Golam Faruque, Mohammad Yunus, Peter Kim Streatfield, Kazuhiko Moji, Paul Wilkinson. Epidemiology 2012;23(1):107-115.

Because evidence on the non-injury health consequences of floods in poor countries is scarce, the authors studied the roughly 211,000 residents of the Matlab surveillance area over 2001-2007, a span that includes the major 2004 flood. Villages were classified as flooded or not flooded in 2004, and the health outcomes tracked were deaths, diarrhoea episodes, and acute respiratory infections. The design was a controlled interrupted time-series comparison, which adjusted for pre-existing differences between the two groups of villages and for seasonal cycles. Once those adjustments were made, neither mortality nor diarrhoea showed convincing excesses during or after the flood. Acute respiratory infection did rise, both during the flood itself and in the following six months (RR 1.25, 95% CI 1.06-1.47), with a further elevation over the subsequent 18 months. The authors argue that their largely null findings, which contrast with earlier studies using weaker control for confounding, underscore the need for rigorous comparison groups in flood-health research.

Matlab data Mortality, diarrhoea (icddr,b Matlab hospital surveillance), and ARI data for ~211,000 Matlab HDSS residents, 2001-2007, with villages classified by exposure to the 2004 flood.

PubMed 22082995 · DOI 10.1097/EDE.0b013e31823ac606

2010Environmental Health

Protection from annual flooding is correlated with increased cholera prevalence in Bangladesh: a zero-inflated regression analysis

Margaret Carrel, Paul R. Voss, Peter Kim Streatfield, Mohammad Yunus, Michael Emch. Environmental Health 2010;9:13.

Motivated by the poorly understood health consequences of altering natural water flows in regions where waterborne infections are endemic, the authors examined cholera in Matlab, Bangladesh, where part of the closely monitored surveillance population lives behind a flood-protection embankment. The exposure was residence inside versus outside the flood-protected zone, and the outcome was the count of cholera cases per household drawn from the long-running hospital-based surveillance. Because most households never report a case, the analysis used a zero-inflated negative binomial model that distinguishes households with no cholera from those with low or high case counts, while accounting for household-level clustering. Across all households, living in the protected area was not significantly related to cholera occurrence. However, among households that did experience cholera, those inside the embankment had significantly more cases. The authors caution that flood-control and water-impoundment schemes can carry unanticipated consequences for waterborne disease and urge water-management actors to weigh these health effects.

Matlab data Household-level cholera case counts from icddr,b Matlab surveillance linked to Matlab HDSS household locations, classified as inside or outside the flood-control embankment.

PubMed 20307294 · DOI 10.1186/1476-069X-9-13

2009Health & Place

Spatio-temporal clustering of cholera: the impact of flood control in Matlab, Bangladesh, 1983-2003

Margaret Carrel, Michael Emch, Peter Kim Streatfield, Mohammad Yunus. Health & Place 2009;15(3):771-782.

This study asks how building flood-control infrastructure in an area where cholera is endemic reshapes when and where the disease occurs. It draws on 21 years (1983-2003) of cholera cases recorded in Matlab, Bangladesh, where an embankment protects part of the surveillance population from seasonal river flooding, so the environmental exposure is the presence or absence of flood protection against the monsoon. The outcome is laboratory-confirmed cholera incidence, geolocated to households through the surveillance system. The authors applied spatial and space-time cluster detection methods to case data before and after the embankment was constructed and compared the resulting cluster patterns. Temporal clustering changed substantially: clusters that had concentrated in the dry season shifted into the rainy season after protection, and cluster onset inside the embanked area lagged behind the outside area. Spatial shifts in cluster locations were less distinct. The authors conclude that altered seasonality of cholera clustering has implications for how local prevention and response are timed.

Matlab data Cholera cases from icddr,b Matlab hospital surveillance (1983-2003) geo-referenced to Matlab HDSS households and classified by location inside or outside the flood-control embankment.

PubMed 19217821 · DOI 10.1016/j.healthplace.2008.12.008

1999Social Science & Medicine

Diarrheal disease risk in Matlab, Bangladesh

Michael Emch. Social Science & Medicine 1999;49(4):519–530.

added by the Scopus search; bibliographic details from the Scopus record, DOI not separately verified

This study set out to weigh biological, socioeconomic, cultural and environmental risk factors for diarrhoeal disease against one another in rural Bangladesh, over space and time. Cases were people hospitalised at the icddr,b Matlab facility between 1 January 1992 and 31 December 1994, split by laboratory result into cholera and non-cholera watery diarrhoea; age-matched controls were drawn at random from the surveillance community. Independent variables came from questionnaires, the demographic surveillance records, community health worker books and a geographic information system. Water availability and use, and sanitation, emerged as the most important modifiable risks for both disease categories, and mattered more for non-cholera watery diarrhoea than for cholera; socioeconomic status acted as an indirect cause, since poverty underlies most of the other variables. Flood control was related to both types of diarrhoea, and the author states plainly that he does not understand why, calling for the relationship to be investigated elsewhere in the country given that the Bangladesh Flood Action Plan will go on building embankments.

Matlab data Cases and controls, household and demographic variables all came from the icddr,b Matlab hospital surveillance and the Matlab demographic surveillance system, geo-referenced in a purpose-built GIS.

Scopus record

1997Bulletin of the World Health Organization

Flood control embankments contribute to the improvement of the health status of children in rural Bangladesh

J. A. Myaux, M. Ali, J. Chakraborty, A. de Francisco. Bulletin of the World Health Organization 1997;75(6):533-539.

Bangladesh builds earthen embankments and drainage works to protect low-lying communities from the annual floods that cover around a third of the country, but critics stress their ecological costs, so this study asked whether such protection translates into child survival gains. The authors used the ICDDR,B Demographic Surveillance System in Matlab to compare children aged 0-4 living inside versus outside a flood-control embankment, further subdivided by whether the villages received intensive health services. Adjusted mortality rates were compared across four adjacent child populations between the periods 1983-86 (before protection) and 1989-92 (after). Over that interval, crude under-five mortality in the whole area fell 37%, from 185.9 to 117.9 per 1,000 live births. After the embankment was completed, death rates in unprotected villages were up to 29% higher among 1-4-year-olds and 9% higher for the whole 0-4 group than in protected villages (P<0.001), while health interventions in the same area were associated with a 40% drop in under-five deaths from all causes (P<0.001). The paper also considers how migration and distance to hospital could shape these contrasts.

Matlab data Child (0-4 years) mortality and population data from the icddr,b Demographic Surveillance System in Matlab, 1983-86 vs 1989-92, with villages classified as inside or outside the Meghna-Dhonagoda flood-control embankment and by health-service area.

PubMed 9509625

1990Ecology of Food and Nutrition

Post-flood nutritional anthropometry of children in Matlab, Bangladesh

M. K. Stewart, V. Fauveau, J. Chakraborty, Md. Yunus, A. M. Sarder. Ecology of Food and Nutrition 1990;24(2):121–131.

added by the Scopus search; bibliographic details from the Scopus record, DOI not separately verified

The Matlab area was badly hit by the 1988 floods, and icddr,b ran post-flood surveys of 600 households to see what this meant for young children. Length, weight and mid-upper-arm circumference were measured in children aged 6–35 months, alongside household economic status and flood damage. Baseline economic status was higher inside the embankment, where the average material damage was in fact worse. Anthropometry did not differ significantly by degree of flood damage, nor between children inside and outside the embankment, and there was no difference between the October and December rounds, although a post-harvest improvement would normally be expected by December. What did predict lower anthropometry was landlessness and lack of maternal education; girls and children aged 12–36 months also had lower mean percent-of-median indices. The study is one of the few to measure a nutritional outcome after a named flood in this population, and its null result for flood exposure anticipates the better-controlled null that Milojevic et al. reported for the 2004 flood.

Matlab data Household survey and child anthropometry conducted by icddr,b within the Matlab surveillance area after the 1988 floods, with households classified by flood damage and by position relative to the Meghna–Dhonagoda embankment.

Scopus record

Climate variability and cholera 14

2024GeoHealth

Satellite-derived, smartphone-delivered geospatial cholera risk information for vulnerable populations

Nusrat F, Akanda AS, Islam A, Aziz S, Pakhtigian EL, Boyle K, Hanifi SMA. GeoHealth 2024;8(11):e2024GH001039.

Bangladesh has no early-warning system that tells households when cholera risk is high, and existing risk communication reaches officials rather than rural residents. The authors developed and validated CholeraMap, a system that forecasts cholera risk from high-resolution satellite and remote-sensing data (avoiding the need for station-based hydroclimatic measurements) and delivers monthly, colour-coded risk maps at 1 km × 1 km resolution to people in Matlab through a smartphone app. The local-scale model reproduced the two annual cholera peaks (pre-monsoon spring and post-monsoon autumn) in the Matlab region and showed in fine detail how at-risk areas expand during outbreak months. The paper presents this as a practical route to household-level cholera early warning in riverine rural communities.

Matlab data Matlab was the study area: the model was validated against the Matlab cholera seasonal pattern and the app was piloted with Matlab households, in collaboration with icddr,b.

PubMed 39524318 · DOI 10.1029/2024GH001039

2018Environment International

Influences of heatwave, rainfall, and tree cover on cholera in Bangladesh

Wu J, Yunus M, Ali M, Escamilla V, Emch M. Environment International 2018;120:304–311.

The authors asked whether heatwaves raise an individual's risk of cholera and whether rainfall and nearby vegetation modify that risk. They used a case-crossover design, in which each person serves as their own control on other days, applied to hospital-confirmed cholera cases from Matlab recorded between January 1983 and April 2009, with conditional logistic regression. Two days after a heatwave, cholera risk was elevated on wet days (rainfall above zero), with an odds ratio of 1.53 (95% CI 1.07–2.19). The effect was stronger for households with less medium-density tree cover (OR 1.80, 95% CI 1.01–3.22) and was not statistically significant where tree cover was greater. The paper concludes that heatwaves may promote cholera, with rainfall and tree cover acting as modifiers, and calls for work on the underlying mechanisms.

Matlab data Individual cholera cases from icddr,b Matlab hospital surveillance (1983–2009) linked to Matlab HDSS household locations and local weather.

PubMed 30107291 · DOI 10.1016/j.envint.2018.08.012

2015Transactions of the Royal Society of Tropical Medicine and Hygiene

Role of phytoplankton in maintaining endemicity and seasonality of cholera in Bangladesh

M. Sirajul Islam, M. Shafiqul Islam, Z. H. Mahmud, J. D. Clemens, A. E. Collins. Transactions of the Royal Society of Tropical Medicine and Hygiene 2015;109(9):572–578.

added by the Scopus search; bibliographic details from the Scopus record, DOI not separately verified

Cholera in Bangladesh is endemic and keeps a regular seasonal rhythm, and this study asked what role phytoplankton plays in sustaining it. Phytoplankton and water samples were taken fortnightly from two ponds in Matlab over a year, and the association of Vibrio cholerae O1 with phytoplankton was examined by culture and by direct fluorescent antibody staining, alongside the bio-physicochemical parameters of the water. Cholera cases were taken from the records of Matlab hospital, and the series were compared with Pearson correlation coefficients. V. cholerae O1 survived 48 days in a culturable state in association with Anabaena variabilis, but persisted for a year in a viable but non-culturable state; it lasted 12 days in control water without algae and 32 days in water with algae. Changing levels of cholera cases in the community correlated significantly with blue-green algae and with total phytoplankton in the aquatic environment, and also with chlorophyll-a and with viable but non-culturable V. cholerae O1. The paper supplies locally measured chlorophyll-a as the counterpart to the satellite chlorophyll signals reported by Constantin de Magny et al. and Emch et al.

Matlab data Cholera case counts came from the icddr,b Matlab hospital records; the environmental sampling was done in two ponds within the Matlab area.

Scopus record

2008American Journal of Tropical Medicine and Hygiene

Local environmental predictors of cholera in Bangladesh and Vietnam

Michael Emch, Caryl Feldacker, Mohammad Yunus, Peter Kim Streatfield, Vu Dinh Thiem, Do Gia Canh, Mohammad Ali. American Journal of Tropical Medicine and Hygiene 2008;78(5):823–832.

added by backward citation searching; see Appendix C

The authors compiled cholera incidence time series and contemporaneous environmental variables for Matlab, Bangladesh from 1983 to 2003, and for Hue and Nha Trang in Vietnam from 1985 to 2003, to ask whether local environmental information can predict cholera. Satellite-derived exposures were monthly sea-surface temperature, sea-surface height and ocean chlorophyll concentration; in-situ exposures were ambient temperature and rainfall from Chandpur, just outside the study area, and river discharge from the two nearest stations, Demra and Bhairab. Because Bangladesh always has cholera cases and seasonal fluctuation is expected, the Bangladesh analysis used ordered probit models asking what makes a given month’s outbreak unusually large compared with the same month in other years, which deseasonalises the comparison by construction. Increases in ocean chlorophyll concentration were associated with a greater magnitude of cholera in Matlab, while sea-surface temperature mattered most in Hue and river height in Nha Trang. The authors conclude that local environmental parameters have consistent effects where cholera is endemic but differ where epidemics are rare.

Matlab data Monthly cholera cases for the Matlab research site were collected through the icddr,b surveillance system from 1983 to 2003, with stool samples taken from all patients admitted to the Matlab hospital.

PubMed 18458320

2014PLoS ONE

Cholera and shigellosis: different epidemiology but similar responses to climate variability

Cash BA, Rodó X, Emch M, Yunus M, Faruque ASG, Pascual M. PLoS ONE 2014;9(9):e107223.

Cholera and shigellosis spread by different routes, so the authors asked whether they respond differently to climate. Using long-run monthly surveillance from icddr,b in urban Dhaka and rural Matlab, they related case counts for both diseases to ENSO, Pacific sea-surface temperatures and the extent of monsoon flooding in Bangladesh. The two diseases turned out to share almost the same climate signature at both sites: heavier monsoon flooding was followed by more cases in the post-monsoon season, and those flood years were themselves preceded by El Niño events. Because the pattern holds for a pathogen without the aquatic ecology of Vibrio cholerae, the authors argue that climate acts mainly by increasing human exposure and eroding population health during floods, rather than through pathogen biology alone, and they discuss the use of ENSO and flood signals as early-warning indicators.

Matlab data Monthly cholera and shigellosis counts from icddr,b Matlab surveillance provided the rural series alongside Dhaka.

PubMed 25229494 · DOI 10.1371/journal.pone.0107223

2013PLoS ONE

A differential effect of Indian Ocean Dipole and El Niño on cholera dynamics in Bangladesh

Hashizume M, Chaves LF, Faruque SM, Yunus M, Streatfield K, Moji K. PLoS ONE 2013;8(3):e60001.

Most earlier work treated the links between El Niño–Southern Oscillation (ENSO), the Indian Ocean Dipole (IOD) and cholera in Bangladesh as fixed over time. This paper tests that assumption using monthly cholera case counts from the icddr,b hospitals in Dhaka and Matlab for 1983–2008, analysed with cross-wavelet coherency against the IOD and ENSO indices. The strength of the climate–cholera coupling changed across timescales and periods: in Dhaka a coherent four-year cycle between ENSO and cholera appeared only during 1988–1997, while in Matlab ENSO exerted the stronger influence and no IOD signal on hospitalisations was detected. The authors conclude that cholera early-warning systems need to allow for non-stationary, possibly non-linear climate effects rather than constant relationships.

Matlab data Monthly cholera cases from icddr,b Matlab hospital surveillance (1983–2008) formed one of the two case series.

PubMed 23555861 · DOI 10.1371/journal.pone.0060001

2013Journal of Health, Population and Nutrition

Time series analysis of cholera in Matlab, Bangladesh, during 1988–2001

Ali M, Kim DR, Yunus M, Emch M. Journal of Health, Population and Nutrition 2013;31(1):11–19.

This paper relates local weather and remotely sensed sea-surface temperature to monthly cholera incidence in Matlab over 168 months (1988–2001), and builds a forecasting model. A case was a Matlab resident with Vibrio cholerae O1 isolated from stool at the treatment centres serving the area; 4,157 cases were recorded, about 1.4 per 1,000 people. A seasonal ARIMA model was fitted to log-transformed monthly counts, which showed autocorrelation at one- and twelve-month lags. A 1 °C rise in minimum ambient temperature was associated with roughly a 6% increase in same-month incidence, and a 1 °C rise in sea-surface temperature with about a 25% increase in the same month and 18% two months later; rainfall showed no clear effect. The model tracked monthly fluctuations well (RMSE 0.108), and the authors propose it for forecasting cholera outbreaks in Matlab.

Matlab data icddr,b Matlab diarrhoea surveillance (1988–2001) supplied the monthly cholera series; the denominator came from the Matlab HDSS population.

PubMed 23617200 · DOI 10.3329/jhpn.v31i1.14744

2011Environmental Health Perspectives

The Indian Ocean Dipole and cholera incidence in Bangladesh: a time-series analysis

Hashizume M, Faruque AS, Terao T, Yunus M, Streatfield K, Yamamoto T, Moji K. Environmental Health Perspectives 2011;119(2):239-244.

The authors analysed monthly counts of cholera patients seen at icddr,b hospitals in Dhaka and Matlab from 1993 to 2007 to test whether the Indian Ocean Dipole (IOD), measured by the Dipole Mode Index (DMI), influences cholera beyond the known effect of ENSO. Negative binomial time-series regression models adjusted for NINO3 sea surface temperature, seasonality and long-term trend, and Bay of Bengal sea surface temperature and height were also examined. A 0.1-unit rise in DMI over the current and preceding three months was associated with a 2.6% increase in cholera in Dhaka (95% CI 0.0 to 5.2%, p = 0.05) and a 6.9% increase in Matlab (95% CI 3.2 to 10.8%, p < 0.01). In Dhaka, a 0.1-unit fall in DMI 4 to 7 months earlier was linked to a 2.4% rise in cases (95% CI 0.0 to 5.0%, p = 0.06). Cholera at both sites was also positively related to northern Bay of Bengal sea surface temperature at lags of 0 to 3 months after adjusting for sea surface height (p < 0.01). The work suggests both phases of the IOD raise cholera risk, acting over different lag windows.

Matlab data Matlab icddr,b hospital surveillance supplied one of the two monthly cholera case series (the other being Dhaka).

PubMed 20980219 · DOI 10.1289/ehp.1002302

2010Environmental Health

Local population and regional environmental drivers of cholera in Bangladesh

Emch M, Yunus M, Escamilla V, Feldacker C, Ali M. Environmental Health 2010;9:2.

This paper asked whether household socio-economic status modifies the influence of large-scale environmental forcing on cholera in the Matlab HDSS area of Bangladesh, a surveillance population of roughly 200,000 people (the abstract does not give the exact study years). Exposures were satellite-measured sea surface temperature, sea surface height and ocean chlorophyll concentration, together with geo-referenced household SES and demographic variables; the outcome was laboratory-confirmed cholera. Generalized estimating equations with a logit link were fitted separately for the pre-monsoon and post-monsoon epidemic seasons. Higher SES was consistently protective in both seasons. In the pre-monsoon season, sea surface height was inversely related to cholera and chlorophyll concentration was positively related, whereas neither showed an association after the monsoon, and sea surface temperature was unrelated in either period. The authors conclude that local conditions such as water and sanitation, for which SES is a proxy, interact with regional ocean signals, chiefly before the monsoon. No effect sizes or confidence intervals are reported in the abstract.

Matlab data Matlab HDSS provided the geo-referenced population, household SES/demographic data and lab-confirmed cholera cases for the whole surveillance population.

PubMed 20074356 · DOI 10.1186/1476-069X-9-2

2009Transactions of the Royal Society of Tropical Medicine and Hygiene

Effects of local climate variability on transmission dynamics of cholera in Matlab, Bangladesh

Islam MS, Sharker MA, Rheman S, Hossain S, Mahmud ZH, Islam MS, Uddin AM, Yunus M, Osman MS, Ernst R, Rector I, Larson CP, Luby SP, Endtz HP, Cravioto A. Transactions of the Royal Society of Tropical Medicine and Hygiene 2009;103(11):1165-1170.

Using monthly totals of cholera cases recorded by icddr,b surveillance in Matlab together with local meteorological records, the authors examined how temperature and hours of sunshine shape the twin summer and winter cholera peaks seen in Bangladesh (the abstract does not state the exact years covered). The exposures were monthly average temperature and daily sunshine duration; the outcome was the monthly count of cholera cases. Classification and regression tree (CART) modelling and principal component analysis were used to partition months by climate conditions. Months with mean temperature under 23.25 degrees C had the fewest cases (about 23 per month); warmer months with less than 4.13 h/day of sunshine averaged 39 cases; months between 23.25 and 28.66 degrees C with at least 4.13 h/day of sunshine averaged 44 cases; and the hottest, sunniest months (above 28.66 degrees C, 4.13 h/day or more) had the highest burden at 54 cases per month. The authors interpret this as a joint effect of warmth and sunshine on cholera transmission, with the two seasons complementing each other to sustain high incidence.

Matlab data Matlab icddr,b hospital/surveillance cholera case counts were the outcome series; Matlab-area climate records were the exposures.

PubMed 19477477 · DOI 10.1016/j.trstmh.2009.04.016

2008Proceedings of the National Academy of Sciences of the United States of America

Environmental signatures associated with cholera epidemics

Constantin de Magny G, Murtugudde R, Sapiano MRP, Nizam A, Brown CW, Busalacchi AJ, Yunus M, Nair GB, Gil AI, Lanata CF, Calkins J, Manna B, Rajendran K, Bhattacharya MK, Huq A, Sack RB, Colwell RR. Proceedings of the National Academy of Sciences of the United States of America 2008;105(46):17676-17681.

This study paired satellite-derived ocean observations (chlorophyll concentration and sea surface temperature) and rainfall records with cholera case counts from two endemic sites, Kolkata (India) and Matlab (Bangladesh), for the years 1998 through 2006. The exposures of interest were coastal and regional environmental signals that might drive Vibrio cholerae abundance, and the outcome was the monthly number of cholera cases at each site. The authors analysed lagged relationships between the environmental time series and case counts. For Bangladesh, anomalies in coastal chlorophyll were significantly related to cholera case numbers with a lead time of roughly one month, and comparable but site-specific environmental patterns were seen for Kolkata. The authors argue that such remotely sensed ocean and climate signals could underpin an early-warning system for cholera outbreaks.

Matlab data Matlab (icddr,b surveillance) supplied the Bangladesh monthly cholera case series that was correlated with satellite ocean and rainfall data.

PubMed 19001267 · DOI 10.1073/pnas.0809654105

2005Nature

Refractory periods and climate forcing in cholera dynamics

Katia Koelle, Xavier Rodo, Mercedes Pascual, Md. Yunus, Golam Mostafa. Nature 2005;436(7051):696-700.

This paper seeks to explain why cholera outbreaks fluctuate on multi-year timescales, drawing on roughly four decades of cholera case records from Matlab, Bangladesh. The climate exposures considered operate at two scales: slow variation in monsoon rainfall and Brahmaputra river discharge, and faster variation in the extent of flooding in Bangladesh, Bay of Bengal sea-surface temperature, and the El Nino-Southern Oscillation. The health outcome is the incidence of El Tor cholera over time. The authors fitted a nonlinear population-dynamic (susceptible-infected-type) model to the time series to back out the time-varying transmission rate and then related that reconstructed rate to the climate indices while incorporating temporary immunity acquired after infection. Reconstructed transmission tracked the climate signals at both long and short periods. Crucially, the climate influence weakened during what the authors call refractory periods, intervals when recent epidemics had left most of the population immune and therefore insensitive to favourable environmental conditions, showing that host immunity and climate jointly govern outbreak timing.

Matlab data Approximately 40 years of monthly cholera case data from icddr,b Matlab hospital surveillance used as the time series for the nonlinear transmission model.

PubMed 16079845 · DOI 10.1038/nature03820

1982American Journal of Epidemiology

Endemic cholera in rural Bangladesh, 1966–1980

R. I. Glass, S. Becker, M. I. Huq, J. V. Lee, R. E. Black and colleagues. American Journal of Epidemiology 1982;116(6):959–970.

added by the Scopus search; bibliographic details from the Scopus record, DOI not separately verified

This is the earliest study in the review and the longest single run of cholera surveillance in it. Since 1963 icddr,b, then the Cholera Research Laboratory, had maintained a field station at Matlab treating patients from a surveillance population of 240,000. The authors analysed hospital records for 7,141 surveillance-area patients culture-positive for V. cholerae O1 from 1966 onward, relating seasonality, age and sex distribution and geographic trend to hypotheses about transmission, immunity and risk groups. Children aged 2–9 years and adult women were most often hospitalised; V. cholerae O1 emerged simultaneously across the surveillance area, with early cases of differing phage types; and only three patients were hospitalised twice against 29 expected on life-table analysis (p < 0.01), implying durable immunity to severe disease after infection. On climate the finding was negative and explicit: no constant relationship was found between the times of onset or peak of the yearly cholera epidemic and the times of onset or peak of the monsoon rains or river water levels. An outbreak of multiply antibiotic-resistant V. cholerae O1 in 1979 raised separate questions about resistance plasmids and drug policy. The authors concluded that little progress had been made in understanding transmission, while fluid therapy had cut case fatality sharply.

Matlab data Fifteen years of culture-confirmed cholera hospitalisations from the Matlab field station, with the surveillance population of 240,000 as the denominator.

Scopus record

2005Proceedings of the Royal Society B

Pathogen adaptation to seasonal forcing and climate change

Katia Koelle, Mercedes Pascual, Md. Yunus. Proceedings of the Royal Society B 2005;272(1566):971–977.

added by backward citation searching; see Appendix C

This is the theoretical companion to the same group’s Nature paper. It asks whether a pathogen evolves its own sensitivity to seasonal environmental forcing, and what that implies when the forcing itself shifts with climate. The core results are simulations of an adaptive-dynamics model in which a generic seasonal driver acts on transmission; the authors describe the cholera component as a largely qualitative application. The empirical side draws on monthly symptomatic cholera counts from Matlab obtained through icddr,b surveillance, covering 1966 to 2002, from which monthly reproductive rates are derived, together with monthly rainfall from the Chandpur station. The argument is that the replacement of the classical biotype by El Tor in Bangladesh can be read as adaptation to a change in the seasonal pattern of monsoon rainfall. River discharge and water salinity are discussed as related drivers, with the authors noting that isolating any specific one is difficult.

Matlab data Monthly cholera case counts from the icddr,b Matlab surveillance programme, 1966–2002, supplied the monthly reproductive rates and the strain-replacement pattern; the main model results are simulations rather than fits to those data.

PubMed 16024354

Weather and childhood diarrhoea and rotavirus 3

2022Proceedings of the Royal Society B: Biological Sciences

The influence of demographic and meteorological factors on temporal patterns of rotavirus infection in Dhaka, Bangladesh

Ernest O. Asare, Mohammad A. Al-Mamun, Monira Sarmin, A. S. G. Faruque, Tahmeed Ahmed, Virginia E. Pitzer. Proceedings of the Royal Society B: Biological Sciences 2022;289(1976):20212727.

fitted to Dhaka data; Matlab used for external validation

To understand what shapes rotavirus dynamics in Bangladesh before vaccine introduction, the authors fitted transmission models to 23 years of weekly rotavirus surveillance from icddr,b's Dhaka hospital (1990-2012), comparing versions that did or did not incorporate long-term and seasonal changes in the birth rate and meteorological covariates. Model performance was then checked against independent 2013-2019 surveillance from both Dhaka and Matlab. The fitted models reproduced the age profile of cases and captured a transition in winter rotavirus hospitalizations from two peaks per year to a single annual peak. The long-run fall in Bangladesh's birth rate emerged as the main driver of that seasonal shift, while weather also mattered: each 1 degree C rise in diurnal temperature range was associated with a 19% increase in the transmission rate, each 1% increase in surface water presence with a 3.9% increase, and each 1 mm increase in wetness with a 0.6% increase. The models performed reasonably well in both Dhaka and Matlab, and the authors propose them as a tool for assessing rotavirus vaccination impact as transmission patterns continue to evolve.

Matlab data icddr,b Matlab surveillance data (2013-2019) served as an external validation site for models fitted to Dhaka data.

PubMed 35673869 · DOI 10.1098/rspb.2021.2727

2014Epidemiology and Infection

Geographical diversity in seasonality of major diarrhoeal pathogens in Bangladesh observed between 2010 and 2012

Sumon Kumar Das, Dilruba Begum, Shahnawaz Ahmed, Farzana Ferdous and colleagues. Epidemiology and Infection 2014;142(12):2530–2541.

added by the themed Google Scholar pass; see Appendix C

Seasonality in diarrhoeal disease is usually described for a country as a whole. This study asks whether it actually holds across sites, using icddr,b surveillance from two urban hospitals in Dhaka and Mirpur and two rural ones, the Matlab Hospital and Kumudini Hospital at Mirzapur, over January 2010 to December 2012 and 21,138 patients. Monthly counts of rotavirus, cholera, shigellosis and ETEC diarrhoea were modelled with quasi-Poisson generalised additive models adjusted for secular trend. The seasonal pattern turned out to be specific to both pathogen and place rather than general: at Matlab, cholera peaked in March–April and again in September–October, while rotavirus fell to a trough in July. The authors discuss rainfall and ambient temperature as possible explanations but state plainly that meteorological variables were not tested, so the study establishes the pattern without accounting for it — the same limitation the older Matlab seasonality analyses carry.

Matlab data Matlab Hospital, which has treated diarrhoea patients from the surveillance area since 1963, was one of the four sites and contributed 4,240 patients.

PubMed 24534384

2014Epidemiology and Infection

Association of climate variability and childhood diarrhoeal disease in rural Bangladesh, 2000–2006

Wu J, Yunus M, Streatfield PK, Emch M. Epidemiology and Infection 2014;142(9):1859–1868.

This study examined how ordinary weather and extreme weather events relate to diarrhoea in young children in Matlab between 2000 and 2006. Daily diarrhoea cases among under-fives from icddr,b surveillance were analysed at the level of the bari (a cluster of related households) with logistic regression, using mean temperature, rainfall, the number of hot days and the number of heavy-rainfall days as exposures. Each additional hot day was associated with a 0.8–3.8% rise in daily childhood diarrhoea cases and each additional heavy-rain day with a 1–6.2% rise; in multivariable stepwise models the extreme-weather indicators stayed positively associated, whereas average temperature and rainfall could become negatively associated once other factors were controlled. The authors conclude that both the frequency and intensity of extreme weather matter for childhood diarrhoea in rural Bangladesh.

Matlab data Childhood diarrhoea cases from icddr,b Matlab surveillance (2000–2006) geo-linked to Matlab HDSS baris.

PubMed 24168897 · DOI 10.1017/S095026881300277X

Weather and childhood pneumonia 3

2022Environmental Research

Protective effect of pneumococcal conjugate vaccination on the short-term association between low temperatures and childhood pneumonia hospitalizations: Interrupted time-series and case-crossover analyses in Matlab, Bangladesh

Junwen Tao, Mohammad Zahid Hossain, Zhiwei Xu, Hung Chak Ho, Md Alfazal Khan, Cunrui Huang, Hao Zheng, Jing Ni, Yinguan Fan, Daniel Bogale, Hong Su, Jian Cheng. Environmental Research 2022;212(Pt A):113156.

Using daily ambient temperature and childhood pneumonia hospitalization data from Matlab, Bangladesh for 2012 through 2016, this study asked whether introducing the 10-valent pneumococcal conjugate vaccine (PCV10) changed children's vulnerability to temperature extremes. The authors combined an interrupted time-series design to quantify the overall vaccine effect with a time-stratified case-crossover analysis using conditional logistic regression, comparing temperature-pneumonia associations before and after PCV10 and stratifying by season, age, sex, and household socioeconomic level. Hospitalizations fell markedly once PCV10 was in use (RR 0.59, 95% CI 0.43-0.83). Heat raised pneumonia risk on the same day (lag 0 OR 1.28, 95% CI 1.02-1.60), whereas cold acted with a roughly four-week delay (lag 28 OR 1.53, 95% CI 1.06-2.22). After vaccination the cold-related risk shrank significantly (p < 0.05) but the heat and heatwave effects grew (p < 0.05), with boys, infants, and children from middle- or low-income households most sensitive to heat. The authors conclude that PCV10 appears to blunt cold-driven but not heat-driven childhood pneumonia in this setting.

Matlab data All exposure and outcome data (childhood pneumonia hospitalizations) came from the Matlab HDSS area / icddr,b Matlab Hospital; Matlab is the named study site.

PubMed 35331698 · DOI 10.1016/j.envres.2022.113156

2020Public Health

Impact of climate variability on length of stay in hospital for childhood pneumonia in rural Bangladesh

Mohammad Zahid Hossain, Shilu Tong, Md Alfazal Khan, Wenbiao Hu. Public Health 2020;183:69-75.

abstract retrieved in the citation-chasing pass; see Appendix C

From the same group as the other Matlab pneumonia–weather studies, this paper asks whether climate variability affects how long children under five stay in hospital with pneumonia, rather than how many are admitted. Admission records from icddr,b’s Matlab Hospital for 1 January 2012 to 31 December 2016 were paired with Bangladesh Meteorological Department data, and a generalised linear model with a Poisson link related length of stay to daily mean temperature, to the temperature variation experienced during the stay, and to humidity variation, adjusting for the child’s age, sex and weight, parental education and socio-economic status. Level mattered little and variability mattered a great deal: a 1 °C rise in mean temperature lengthened a stay by 1.0% (RR 1.010, 95% CI 1.001–1.018), while each 1 °C of temperature variation during the stay lengthened it by 9.3% (RR 1.093, 95% CI 1.051–1.138) and each 1% of humidity variation by 2.2% (RR 1.022, 95% CI 1.004–1.039). The authors cost each degree of in-stay temperature variation at about 0.81 USD per patient-day in direct costs and 1.8 USD per patient-day in total, and describe the design as ecological, so the estimates are associations.

Matlab data Pediatric pneumonia admission records from icddr,b's Matlab Hospital (Matlab HDSS area) provided the length-of-stay outcome data.

PubMed 32438214 · DOI 10.1016/j.puhe.2020.03.014

2020International Journal of Biometeorology

Weather factors, PCV intervention and childhood pneumonia in rural Bangladesh

Mohammad Zahid Hossain, Shilu Tong, Hilary Bambrick, Al Fazal Khan, Samar Kumar Hore, Wenbiao Hu. International Journal of Biometeorology 2020;64(4):561-569.

This study drew on five years of childhood pneumonia case records (1 January 2012 to 31 December 2016) from icddr,b's Matlab Hospital, paired with daily temperature and relative humidity readings supplied by the Bangladesh Meteorological Department. The authors asked whether weather variability still shaped pneumonia incidence in young children once the pneumococcal conjugate vaccine (PCV) had been rolled out in the area. Cross-correlation functions were first used to identify appropriate lags for each weather variable, and a Poisson generalized linear model was then fitted with an adjustment term for the PCV rollout. Annual pneumonia incidence dropped from roughly 5,691 to about 2,000 per 100,000 children after the vaccine was introduced. Higher temperature was linked to lower pneumonia risk (RR 0.985, 95% CI 0.974-0.997) and the PCV period itself was strongly protective (RR 0.489, 95% CI 0.435-0.551), while relative humidity showed no meaningful association. The authors argue that weather-informed early warning combined with vaccination could support pneumonia control in lower-middle-income settings such as Bangladesh.

Matlab data Childhood pneumonia case data came from icddr,b's Matlab Hospital in the Matlab HDSS area; Matlab was the sole study site.

PubMed 31848699 · DOI 10.1007/s00484-019-01842-7

Heat and season in pregnancy, and climate-linked migration 5

2020Current Developments in Nutrition

Effect of Maternal Exposure to Seasons during the Second and Third Trimesters of Pregnancy on Infant Birth Weight in Rural Bangladesh

Sharif Imam Ahmed, S. M. Tafsir Hasan, Md. Alfazal Khan, Tahmeed Ahmed. Current Developments in Nutrition 2020;4(3):nzaa016.

added by the Scopus search; bibliographic details from the Scopus record, DOI not separately verified

Pregnant women in agricultural communities face seasonal food insecurity and energy stress, and this study asked how exposure to each season during the second and third trimesters affects birth weight. Information on 3,831 singleton live births was taken from the electronic databases of the Matlab Health and Demographic Surveillance System and the icddr,b Matlab hospital, covering all term births from July 2011 to June 2015, with congenital anomalies and records with missing data excluded. Each year was divided into three agro-climatic seasons: the post-aman harvest period (January–April), the height of the monsoon (May–August) and the post-aus harvest period (September–December). Seasonal exposure was measured in weeks rather than by season of birth, and multivariable linear regression was used to isolate the effect of each week of exposure in each trimester. Birth weight peaked in the post-aman harvest season, highest among infants born in March (mean 2,930.5 g, SD 462.1), and was lowest in July during the monsoon (2,830.6 g, SD 385.4). Exposure to the post-aman harvest season in the third trimester, and to the post-aus harvest in the second, both raised birth weight; in the final adjusted model each week of post-aman exposure in the third trimester was worth 6.3 g (95% CI 1.6–10.9, P = 0.008), so that infants exposed for the full 14 weeks were 88.2 g heavier. The authors call for work on the interplay between seasonal energy stress and maternal and fetal nutrition.

Matlab data Birth records from the icddr,b Matlab HDSS and Matlab hospital databases, July 2011 to June 2015, with the agro-climatic calendar as the exposure.

Scopus record

2019BMJ Global Health

Time trends and sociodemographic determinants of preterm births in pregnancy cohorts in Matlab, Bangladesh, 1990–2014

Anisur Rahman, Monjur Rahman, Jesmin Pervin, Shams El Arifeen, Lars Åke Persson and colleagues. BMJ Global Health 2019;4(4):e001462.

added by the Scopus search; bibliographic details from the Scopus record, DOI not separately verified

Preterm birth is the leading cause of under-five mortality and population-based data on it are scarce in low-income countries. Using Health and Demographic Surveillance System data from Matlab for 1990 to 2014, with gestational age based on reported last menstrual period and verified by ultrasound, the authors assessed social, reproductive and environmental determinants of preterm birth across 63,063 live births, running a time-series analysis with decomposition for trend and seasonality. The preterm proportion fell from 29% (95% CI 28.6–30.1) in 1990–94 to 11% (95% CI 10.5–11.6) in 2010–14. Low education, older maternal age and multiparity were associated with higher proportions throughout, and roughly 27% of the overall reduction could be attributed to the rapid rise in women’s education and the fall in parity. Preterm births showed a marked seasonal variation which persisted across the whole 25-year period, and which the authors interpret as reflecting environmental stressors acting on pregnant women that the sociodemographic transition did not remove. It is the longest birth series in this review and the only one to show a seasonal signal holding steady while the underlying rate falls by two-thirds.

Matlab data The full Matlab HDSS pregnancy cohort, 1990–2014, with ultrasound-verified gestational age.

Scopus record

2018PLoS ONE

Magnitude and determinants of inadequate third-trimester weight gain in rural Bangladesh

S. M. Tafsir Hasan, Sabuktagin Rahman, Lindsey M. Locks, Md. Alfazal Khan, Tahmeed Ahmed. PLoS ONE 2018;13(4):e0196190.

added by the themed Google Scholar pass; see Appendix C

Weight gain in pregnancy is a determinant of birth size, and this study asked how often it falls short and why, among rural women in Matlab, Bangladesh. The cohort was 1,883 women admitted for childbirth at the icddr,b Matlab hospital during 2012–2014, with inadequate third-trimester gain defined as 4 kg or less; 1,026 of them, or 54%, met that definition. Determinants were examined in a multivariable logistic model covering nutritional, socio-economic, obstetric and seasonal factors. Among them, the season in which the pregnancy was conceived mattered: conceiving during the monsoon or the dry season rather than the summer raised the odds of inadequate third-trimester weight gain (OR 1.4, 95% CI 1.1–1.8). Season is one determinant among several here rather than the object of the analysis, but it is entered as an explicit climatic category with an effect estimate, which is the basis on which the study is included.

Matlab data Pregnant women admitted for childbirth at the icddr,b Matlab hospital, 2012–2014, with anthropometry collected in the hospital record.

PubMed 29698483

2017Global Environmental Change

Disruption, not displacement: Environmental variability and temporary migration in Bangladesh

Maia A. Call, Clark L. Gray, Mohammad Yunus, Michael E. Emch. Global Environmental Change 2017;46:157-165.

This study tested whether temperature, rainfall, and flooding predict short-term out-migration from Matlab, Bangladesh, challenging the simple 'environmental refugee' narrative in which climate stress uniformly pushes people to leave. It drew on high-frequency individual-level records from the Matlab demographic surveillance system, linked to environmental time series, and applied a discrete-time event history model with controls for household sociodemographic characteristics and local context. Temporary migration fell sharply in the period immediately after floods but rebounded quickly to baseline levels. By contrast, unusually high temperatures were followed by increased temporary migration that persisted for one to two years, whereas extended spells of extreme precipitation suppressed movement. The authors interpret these mixed results as evidence that environmental shocks disrupt established household livelihood strategies that blend farming with seasonal labor migration, rather than causing permanent displacement. The abstract reports directions and durations of effects rather than numeric coefficients.

Matlab data Longitudinal migration and household records from the icddr,b Matlab HDSS were the core dataset for the event history analysis.

PubMed 29375196 · DOI 10.1016/j.gloenvcha.2017.08.008

2017Tropical Medicine and Health

Temperature during pregnancy influences the fetal growth and birth size

Harunor Rashid, Miki Kagami, Farzana Ferdous, Enbo Ma, Toru Terao, Taiichi Hayashi, Yukiko Wagatsuma. Tropical Medicine and Health 2017;45:1.

The authors used 3,267 singleton births from the MINIMat maternal food and micronutrient supplementation trial, which enrolled pregnant women in Matlab, Bangladesh between November 2001 and October 2003, to test which weather variable and which gestational window matter most for newborn size. Temperature, rainfall, and humidity records came from the Chandpur weather station about 10 km away, converted into two-week moving averages at several gestational time points, and births were also grouped into winter (October-February), hot-dry (March-May), and monsoon (June-September) seasons. Linear regression models adjusted for gestational age, season, infant sex, parity, early-pregnancy BMI, maternal education, and precipitation. Winter-born infants were shortest, with a mean birth length of 47.5 cm (SD 2.2) versus 47.8 cm (SD 2.1) in the hot-dry season and 47.9 cm (SD 2.1) in the monsoon (P < 0.001). Warmer temperatures in the final month of pregnancy predicted greater birth length (P < 0.01), and warmer mid-gestation temperatures predicted higher birth weight (P < 0.05). The authors conclude that gestational temperature exposure affects both weight and length at birth and suggest that undernourished women (BMI below 18.5) in particular should limit cold exposure late in pregnancy.

Matlab data Birth outcome and maternal data came from the MINIMat randomized trial conducted within the icddr,b Matlab HDSS population.

PubMed 28077924 · DOI 10.1186/s41182-016-0041-6

Seasonality of mortality (season as climatic proxy) 6

2007Journal of Health, Population and Nutrition

Childhood mortality due to drowning in rural Matlab of Bangladesh: magnitude of the problem and proposed solutions

A. Iqbal, T. Shirin, T. Ahmed, A. Sobhan, A. K. Siddique and colleagues. Journal of Health, Population and Nutrition 2007;25(3):370–376.

added by the Scopus search; bibliographic details from the Scopus record, DOI not separately verified

Drowning is a major cause of child death in rural Bangladesh and had attracted little preventive effort, so this study described the problem using the Matlab demographic surveillance records for 1985 to 2000. Of 989 drowning deaths reported over those years, 796 (80.5%) were children aged 1–4, 48 (4.8%) were infants and 145 (14.7%) were aged 5–19. Across the same period all-cause mortality at ages 1–4 fell steeply, from 20.7 to 5.2 per 1,000, while the drowning death rate did not: the proportion of child deaths attributable to drowning therefore rose. Forty-five percent of deaths occurred in ponds, 16.8% in ditches, 8.1% in canals and 4.4% in rivers, with the site unrecorded in over a quarter. Analysis of seasonal variation showed that most drownings fell in April to October, that is largely the monsoon months. Two-thirds of the mothers of victims had no formal education, and boys drowned about 20% more often than girls (OR 1.2, 95% CI 1.04–1.38, p < 0.012). The authors recommend awareness among mothers and close relatives, door-fencing, and filling unused ditches and water holes.

Matlab data Drowning deaths and child population denominators from the icddr,b Matlab demographic surveillance system, 1985–2000.

Scopus record

2004International Journal of Epidemiology

Comparative analysis of patterns of survival by season of birth in rural Bangladeshi and Gambian populations

Sophie E. Moore, Anthony J. C. Fulford, P. Kim Streatfield, Lars Ake Persson, Andrew M. Prentice. International Journal of Epidemiology 2004;33(1):137-143.

Prompted by earlier Gambian evidence that being born in the annual hungry season raised the risk of dying from infections in young adulthood, this study asked whether the same pattern appeared in rural Bangladesh, where pregnancy also faces strong seasonal nutritional stress. The authors used the Matlab demographic surveillance system, in operation since 1966, analyzing 172,228 births and 24,697 deaths recorded between 1974 and 2000. Based on monthly patterns of conception and neonatal mortality, they classified births into a harvest season (January-June) and a hungry season (July-December) and compared survival with hazard models. Hungry-season birth was associated with higher mortality during infancy, but there was no excess among deaths after age 15: the hazard ratio for July-December versus January-June births was 1.12 (95% CI 0.87-1.45). The authors note that adult mortality was far lower in Matlab (about 0.1%) than in The Gambia (about 3%), which may make subtle immune differences undetectable as deaths, and raise the possibility that the Gambian result is a local phenomenon that does not generalize.

Matlab data The icddr,b Matlab HDSS (1974-2000 births and deaths) was the Bangladeshi cohort compared against the Gambian population.

PubMed 15075159 · DOI 10.1093/ije/dyh007

1998International Journal of Epidemiology

Seasonal patterns of deaths in Matlab, Bangladesh

Becker S, Weng S. International Journal of Epidemiology 1998;27(5):814-823.

This paper revisited the seasonality of mortality in Matlab, Bangladesh roughly 15 years after an earlier analysis of 1972-1974 vital registration data had documented sinusoidal seasonal cycles, asking whether the sharp mortality decline of the intervening decades had dampened or shifted those cycles. The authors fitted trigonometric regression models to monthly death counts by age group and cause from the Matlab vital registration system for 1982-1990, covering 20,328 deaths. Significant sinusoidal seasonality persisted in every age and cause group except one. All-cause and neonatal deaths peaked in winter, whereas post-neonatal deaths peaked in April and child deaths in July; among causes, injury deaths (mostly drownings) showed the largest seasonal amplitude. Compared with 1972-1974, the timing of the peak had moved only for neonates, and the amplitude of the seasonal swing had fallen significantly only for neonatal deaths and for injury deaths. The authors conclude that pronounced seasonal mortality patterns endure in Matlab even as overall death rates have fallen.

Matlab data Monthly death records (1982-1990) from the icddr,b Matlab vital registration / HDSS were the entire dataset.

PubMed 9839738 · DOI 10.1093/ije/27.5.814

1996Demography

Estimating seasonality effects on child mortality in Matlab, Bangladesh

Pradip K. Muhuri. Demography 1996;33(1):98-110.

This demographic analysis sought to isolate the net influence of calendar season on the risk of death among young children in Matlab, Bangladesh, using the longitudinal surveillance records maintained there. Seasonality here is a proxy for the agricultural and climatic cycle: the hot dry pre-monsoon period, the monsoon, the lean 'hungry' months before the aman rice harvest, and the post-harvest months. Child mortality was found to run well above the monthly average in April, a hot and dry month, and again in November when the first aman harvest comes in, while it was markedly low in the post-harvest months of February and March and also in August. During the hungry months of September and October, children of mothers with no schooling faced a substantially elevated risk of dying, especially from diarrheal disease, whereas children of schooled mothers did not show this excess. The protective effect of the Matlab maternal and child health intervention against diarrheal deaths was also larger during these hungry months than in the rest of the year. The abstract reports the pattern of relative risks rather than specific point estimates.

Matlab data Child mortality and household data came from the icddr,b Matlab HDSS; the analysis also compared the Matlab MCH-FP intervention area with the comparison area.

PubMed 8690143 · DOI 10.2307/2061716

1990Journal of Diarrhoeal Diseases Research

Pattern of diarrhoeal deaths during 1966–1987 in a demographic surveillance area in rural Bangladesh

K. Shaikh, B. Wojtyniak, G. Mostafa, M. U. Khan. Journal of Diarrhoeal Diseases Research 1990;8(4):147–154.

added by the Scopus search; bibliographic details from the Scopus record, DOI not separately verified

This paper analyses the pattern of death in relation to age, sex, seasonality and year-to-year variation over 22 years in Matlab upazila, Chandpur district. The diarrhoeal death rate moved between 2.0 and 4.0 per 1,000 population for most of the period, but ran at twice the preceding five-year average during the war year of 1971 and the famine of 1974–75. On average more than 20% of all deaths appeared to be diarrhoea-related, and the relative importance of diarrhoea as a cause of death did not diminish over time; persistent diarrhoea killed more people than acute diarrhoea. Diarrhoeal mortality was consistently highest in children aged 1–4, and particularly in girls, and women under 20 died of diarrhoea more often than men of the same age. November was the peak month for diarrhoeal deaths in this rural area. The introduction of the Maternity, Child Health and Family Planning services had a significant effect in reducing diarrhoeal deaths as well as deaths from other causes. The November peak aligns with the second mortality peak that Muhuri later located at the start of the aman harvest.

Matlab data Twenty-two years of death registrations from the icddr,b Matlab demographic surveillance area, analysed by cause, age, sex and calendar month.

Scopus record

1981International Journal of Epidemiology

Seasonality of deaths in Matlab, Bangladesh

Stan Becker. International Journal of Epidemiology 1981;10(3):271-280.

This early study examined the timing of deaths registered between 1972 and 1974 in the Matlab surveillance area, a rural population of roughly 260,000 people, breaking the data down by age and cause to see how mortality tracked the annual climatic cycle. Neonatal deaths appeared to peak in October, but once the seasonal pattern of births was taken into account the underlying neonatal risk actually peaked about two months earlier. Post-neonatal deaths peaked in April, with dysentery accounting for the excess in that month, and deaths among people over 45 clustered in the cool season. Mortality from dysentery and chronic diarrhoea was highest in December, matching the seasonal peak of shigella infection, while accidental deaths, chiefly drownings, peaked in July at the height of the monsoon. The author observed that the age and cause groups contributing most to total deaths were also the ones with the clearest seasonal cycles. The analysis is descriptive, reporting peak months and cause attribution rather than regression-based effect estimates.

Matlab data Death registrations (1972-1974) from the Cholera Research Laboratory / icddr,b Matlab surveillance system were the sole data source.

PubMed 7287288 · DOI 10.1093/ije/10.3.271

The same review as a deck: eleven slides, each with the speaker notes underneath. Best read on a wide screen — the slides keep their 16:9 shape and scale down to fit.

01 Climate, weather and air pollution research using the Matlab HDSS

Systematic review · PRISMA 2020

Climate, weather and air pollution research using the Matlab HDSS

What 47 studies have established, what they still dispute, and what nobody has looked at — in the longest-running surveillance population in the global South.

47

studies included

1981–2024

publication years

8

exposure–outcome themes

0

on ambient air pollution

Speaker notesslide 1

This is a PRISMA 2020 systematic review of every peer-reviewed study that has used the Matlab Health and Demographic Surveillance System to look at weather, climate, flooding or air quality against health.

Forty-seven studies, published between 1981 and 2024.

The number I want to hold onto from the start is the last one. Zero studies have linked ambient air pollution to health in Matlab, despite the site sitting downwind of Dhaka and inside the Indo-Gangetic winter haze belt. That absence is the main finding of the review.

02 Why this site can answer climate questions at all

Why this site can answer climate questions at all

Matlab has followed about 230,000 people continuously since 1966 — the longest-running health and demographic surveillance site in the global South.

Daily records of births, deaths, migrations and hospital-confirmed infections, next to meteorological stations. Very few places in a low-income country let you examine climate and health across decades of individual-level data.

Daily mortality from 1983

Cause of death by verbal autopsy, individually linked

Culture-confirmed cholera since 1966

The record that made Matlab the site it is

A local weather station since 2008

And measured personal PM2.5 in 400 women from 2026

Speaker notesslide 2

The reason this review is worth doing is the dataset, not the literature.

Matlab is run by icddr,b in Chandpur district. It has followed roughly 230,000 people continuously since 1966, which makes it the longest-running surveillance site in the global South.

What matters for climate epidemiology is that the records are daily and individually linked — births, deaths, migrations, hospital-confirmed infections — and that there are meteorological stations nearby. That combination is rare in a low-income setting.

So the question is not whether the data can support climate–health work. It is what has actually been done with it.

03 Where Matlab sits, who it follows, and what it breathes

Where Matlab sits, who it follows, and what it breathes

Chandpur district, 55 km south-east of Dhaka — rural by population, but not by the air it breathes

DhakaMATLABChandpur districtAmbient PM2.53580 µg/m³

Mean ambient PM2.5, µg/m³ (2022), by district · 3,149 survey clusters

246,893

people under continuous surveillance (2020)
129,230 icddr,b service area · 117,663 government area

1981–2020, from the icddr,b Matlab DSS annual reports

Rural to urban, in four steps

degree of urbanisation, 3,149 clusters

 1 most rural234 city core
People per km²8301,4005,77024,390
Ambient PM2.5, µg/m³55.560.662.672.9
Cooking on polluting fuel93%87%46%8%

Ambient pollution rises; household smoke falls far faster. They cross between steps 2 and 3.

Matlab is rural by population — 94% of Chandpur’s clusters are steps 1–2, at 1,620 people per km² — but it breathes 61 µg/m³, 12× the WHO guideline, the air of a town. Not one of the 47 studies has linked that to health.

Speaker notesslide 3

Three things about the site, because people always ask.

Left: where it is. Chandpur district, about 55 kilometres south-east of Dhaka, on the Meghna. Each dot is a district, shaded by mean ambient PM2.5 in 2022, built from 3,149 survey clusters. Dhaka is the dark cluster to the north-west. Matlab is the ringed one.

Top right: who it follows. The surveillance population has grown from about 182,977 in 1981 to about 246,893 in 2020, split almost evenly between the icddr,b service area and the government comparison area. Those numbers are from the Matlab annual reports, not from me.

Bottom right is the part worth arguing about. Take the four-step degree of urbanisation rather than a crude rural-urban binary, which lumps small towns in with Dhaka and hides everything. Population density runs from about 830 people per square kilometre at step one to about 24,390 in the city core. Ambient PM2.5 rises across that, 55.5 to 72.9. But cooking on polluting fuel collapses, 93 per cent down to 8 per cent. The two lines cross between steps two and three.

Now put Matlab on it. 94 per cent of Chandpur's clusters are steps one and two, at about 1,620 people per square kilometre. Rural. But its ambient PM2.5 is 61 micrograms, which is step-three air, around 12 times the WHO guideline. It sits in the Dhaka plume.

So Matlab has rural density and household smoke, and town-level outdoor pollution on top. Every air-pollution study here has looked only at the smoke.

04 How the 47 were found

How the 47 were found

280

records from PubMed, Scopus and Web of Science

183

screened on title and abstract

39

met the criteria from the databases

47

after citation chasing and Google Scholar

All three database strings need “Matlab” in the title, abstract or keywords. A Matlab study titled “… in rural Bangladesh” is invisible to them.

Every one of the 8 extra studies found by citation chasing and Scholar was titled that way. The blind spot is real, and it is in the search, not the literature.

Design graded A (18 studies) · B (24) · C (5) on design strength only — not a judgement of the paper

Speaker notesslide 4

Briefly on method, because one point here matters.

Three databases, searched the same day: 280 records, 183 screened after removing duplicates and already-included studies, 39 meeting the criteria.

Then backward citation searching of 40 reference lists, and a Google Scholar query for each theme, added eight more — giving 47.

The point is the box at the bottom. Every database string requires the word Matlab in the title, abstract or keywords. But a lot of Matlab research is published as "a study in rural Bangladesh". All eight of the extra studies were titled that way. So anyone searching these databases for Matlab climate work will systematically miss a sixth of it.

Each study is graded A, B or C on design strength — that is about design, not quality of the paper.

05 One outcome takes a third of the literature

One outcome takes a third of the literature

The 47 studies by exposure–outcome theme

Climate variability and cholera

14

Flooding, flood control and salinity

7

Household air pollution

6

Seasonality of mortality

6

Heat and season in pregnancy, migration

5

Temperature, weather and mortality

3

Childhood diarrhoea and rotavirus

3

Childhood pneumonia

3

Each study counted once, at its primary theme

Speaker notesslide 5

Here is the shape of the literature.

Cholera is 14 of the 47 — close to a third — and that reflects the site's origins as a cholera research station rather than any judgement about what matters most for health today.

Flooding and flood control is seven. Household air pollution and seasonality of mortality are six each.

The three themes at the bottom — temperature and mortality, childhood diarrhoea, childhood pneumonia — are three studies each. Those are the ones using the daily time-series design that the dataset is actually best suited to.

So the depth is uneven, and it is deepest where the historical interest was, not where the data is strongest.

06 Every study, by what was measured against what

Every study, by what was measured against what

Exposure class against outcome class. The bottom row is the finding.

Exposure Mortality Cholera Diarrhoea Pneumonia Birth, growth Other All
Temperature, heat and heatwaves 322311 12
Season and the agro-climatic cycle 5—1—31 10
Ocean and climate indices —9———— 9
Flood, river flow and salinity 231—1— 7
Household air pollution 1———14 6
Rainfall, humidity and river level —2———— 2
Local water and plankton ecology —1———— 1
Ambient air pollution — nothing, anywhere 000000 0
Studies in column 11174366 47

Each study at its primary exposure and outcome · a dash means no study

Speaker notesslide 6

This is the whole review on one slide: every study placed at its primary exposure and its primary outcome.

Read the columns first. Cholera has 17 studies. Mortality has 11. Everything else is in single figures.

Read the rows and you see where the exposures come from. Nine of the cholera studies arrive on ocean and climate indices — sea-surface temperature, ENSO, the Indian Ocean Dipole — measured hundreds of kilometres away, not on anything recorded in Matlab.

Then the bottom row. Ambient air pollution: zero against every single outcome. Not thin — empty. Matlab sits downwind of Dhaka, inside the Indo-Gangetic winter haze belt, with daily mortality since 1983, and nobody has put those two things together.

That is the row I would want a new analysis to fill.

07 The surveillance continues. The analysis stopped.

The surveillance continues. The analysis stopped.

Last year of Matlab data actually analysed, for the 34 studies giving explicit calendar years

up to 1990

6

1991–2000

3

2001–2010

15

2011–2019

10

2020 onwards

0

Not one study analyses a Matlab record after 2019 — so nothing in this literature covers the hottest years the site has ever recorded. The mortality time series stop at 2009.

13 further studies give a duration (“10 years”) or no period at all

Speaker notesslide 7

This is the slide I would put most weight on for a new analysis.

Take the 34 studies that say exactly which calendar years they analysed. Fifteen of them stop somewhere in 2001 to 2010. Ten reach into the 2010s. And after 2019, nothing at all.

The three temperature–mortality time series — the ones with the strongest designs in the whole review — stop at 2009.

So the entire climate–health literature on Matlab predates the hottest years on record at the site, and predates the period everyone is actually worried about. The surveillance system kept running; the analysis did not keep up.

Extending those series to the present, with a distributed-lag non-linear model and cause-specific verbal-autopsy outcomes, would tell you whether a heat arm has emerged. Nobody has done it.

08 Five findings more than one study supports

What holds up

Five findings more than one study supports

Cold, not heat, kills

Each 1 °C fall in daily mean temperature raises all-cause mortality by about 3.2%. Three independent time series agree, and no heat effect is detectable anywhere in the record.

Heat raises infection

Cholera, childhood diarrhoea and childhood pneumonia all rise with heat — pneumonia admissions on the same day, diarrhoea with counts of hot days rather than averages.

Cholera tracks the ocean

ENSO, the Indian Ocean Dipole, Bay of Bengal temperature and chlorophyll all lead cholera by weeks to months — the basis for forecasting work.

Biomass smoke harms lungs

All three studies testing cooking fuel against obstructive lung disease find it — though every one infers exposure from fuel type rather than measuring it.

Flood control saved children

Unprotected villages had up to 29% higher mortality at ages 1–4 after the 1986 embankment — but see the next slide.

Speaker notesslide 8

What can you actually say from 47 studies?

Five things hold up, in the sense that more than one study supports them.

First, and this surprises people: in Matlab it is cold that kills, not heat. Three independent time-series analyses agree — about 3.2% more deaths per degree of cooling, steepest for cardiovascular and respiratory causes in people over 60. No detectable heat effect. But remember those series stop in 2009.

Second, heat does drive infection — cholera, diarrhoea and pneumonia.

Third, cholera tracks ocean and climate indices at lags of weeks to months, which is what makes forecasting plausible.

Fourth, biomass cooking smoke is associated with obstructive lung disease in all three studies that tested it.

Fifth, the 1986 embankment improved child survival. That one comes with a caveat, which is the next slide.

09 Four disputes and one honest null

What does not

Four disputes and one honest null

Rainfall and cholera

The earliest and longest series — 7,141 cases over 1966–1980 — found no constant relationship to the monsoon or river levels. Every later finding has to be read against that null.

Flood control, both ways

The same embankment that improved child survival moved cholera into the rainy season and concentrated cases behind it — and its link to diarrhoea its own author could not explain.

Temperature and pneumonia

One study finds warmth mildly protective; another finds heat raising admissions the same day and cold at a four-week lag. The three disagree on sign.

The floods themselves

Once village differences and seasonality were controlled, the 1988 and 2004 floods produced no convincing excess in mortality or nutrition — only acute respiratory infection rose.

The one measurement study

Personal PM2.5, black carbon and carbon monoxide against vascular function in 199 women: null on all three. The only study to measure rather than infer found nothing.

Speaker notesslide 9

Now the other side.

The oldest study in the review, Glass 1982, took 7,141 culture-confirmed cholera cases over fifteen years and found no constant relationship between the epidemic and the monsoon rains or river levels. That null sits underneath everything that came later, and the review is careful to say so.

Flood control cuts both ways — it saved children's lives and it redistributed cholera.

Temperature and pneumonia: the three studies disagree about the direction.

And the floods themselves: once you control properly for which villages flood and for seasonality, the big 1988 and 2004 events produced much less excess mortality than weaker study designs had reported.

The last one matters most for the next slide. The single study that actually measured pollutant concentrations, rather than inferring exposure from what people cook with, found nothing. So the biomass–lung-disease finding rests entirely on fuel type as a proxy.

10 Five questions this dataset can answer and nobody has asked

The point of the review

Five questions this dataset can answer and nobody has asked

01

Ambient air pollution and health. Downwind of Dhaka, inside the winter haze belt, daily mortality since 1983 — and not one study.

02

Heat and pollution together. Every input exists — daily deaths, daily admissions, a weather station, measured particulates. No joint model has been fitted.

03

Weather, water quality, disease. Rainfall and heat predict E. coli in Matlab tubewells. Nobody has carried that through to diarrhoeal illness.

04

Temperature and mortality since 2009. The series stop before the hottest years on record. Has a heat arm emerged?

05

Non-communicable disease under heat. Cardiovascular, renal and mental health — unexamined, in a population whose epidemiological transition the same system documents.

Speaker notesslide 10

This is what I would actually present to a funder.

The review's contribution is not the summary of what is known. It is this list — five questions that this specific dataset is unusually well placed to answer, where nobody has done the work.

Number one is the big one. Ambient air pollution against any health outcome: zero studies. And the usual excuse, that there is no exposure data, has just gone away — a 2026 study measured personal and kitchen PM2.5 and black carbon in 400 women in Matlab and Araihazar. It reports no health outcome, so it is not in this review, but it gives you the ground truth to validate satellite or reanalysis estimates against.

Number three is similar: two studies have now related rainfall and temperature to E. coli in Matlab tubewells. The middle term is established. The health outcome is simply missing.

Each of these is a paper that could be written with data that already exists.

11 The first two are tractable now

What it would take

The first two are tractable now

Ambient PM2.5 and daily mortality

Validate satellite or reanalysis PM2.5 at 1 km against the 2026 ground measurements, then carry it into the same time-series design already used here for temperature.

Exposure layer: exists. Outcome: 40 years of daily deaths. Design: already validated on this site.

Updated temperature–mortality series

Extend the 1983–2009 analyses to the present with a distributed-lag non-linear model and cause-specific verbal-autopsy outcomes.

The single clearest test of whether the cold-dominant finding still holds in a warmer decade.

One caution the review insists on: rural Matlab women carried higher personal and kitchen PM2.5 than peri-urban women. A heat–pollution model here would be testing household smoke against heat, not traffic against heat.

Speaker notesslide 11

If you only take two things away as doable projects, take these.

The first is ambient PM2.5 against daily mortality. The exposure layer is now obtainable — satellite or reanalysis estimates at one-kilometre resolution, validated against the 2026 ground measurements in Matlab itself. The outcome is forty years of daily, cause-coded deaths. And the time-series design has already been validated on this exact population for temperature. It is close to assembly rather than invention.

The second is simply bringing the temperature–mortality series up to date. Those analyses stop at 2009. A distributed-lag non-linear model on the full record, with cause-specific outcomes, directly tests whether the cold-dominant pattern survives in a warmer decade.

The caution at the bottom is worth stating out loud: the 2026 measurements found rural women more exposed than peri-urban women. So in this setting, pollution means household smoke, not traffic. A model that assumes otherwise would be answering the wrong question.

12 Fifty years of data, used mainly to study cholera and the cold

Fifty years of data, used mainly to study cholera and the cold

The gaps are not in the dataset. They are in what has been asked of it — and the two largest can be closed with records that already exist.

The full review

imperialcollegelondon.github.io/
matlab-hdss-climate-health-review

Table 1, five figures, four appendices, every search string and every excluded candidate with its reason.

What it is

PRISMA 2020 structure across PubMed, Scopus and Web of Science, plus citation chasing of 40 reference lists and a themed Google Scholar pass.

What it is not

No registered protocol, no meta-analysis, and screening and grading done once rather than in duplicate. Read it as a mapped scoping of the literature.

Speaker notesslide 12

To close.

Fifty years of some of the best longitudinal health data anywhere in the global South, and the climate literature built on it is concentrated on cholera, and on cold rather than heat, with almost nothing after 2019.

The gaps are not in the dataset. They are in the questions that have been put to it. And the two biggest — ambient pollution, and an updated temperature–mortality series — need no new data collection at all.

The full review is at that link: the complete table of all 47 studies, the five figures, the search strings, and every candidate that was excluded with the reason.

I should be straight about its limits, which the review states itself: there is no registered protocol, no meta-analysis, and screening and grading were done once rather than in duplicate. So it is a mapped, PRISMA-structured scoping of the literature rather than a fully independent systematic review.

Happy to take questions.