Flood exposure and pregnancy loss in 33 developing countries

Catastrophic floods have intensified in severity, duration, and frequency due to climate change, sea-level rise, infrastructure, and population dynamics. Floods are now the most prevalent natural disaster, affecting billions and causing substantial economic losses. While direct flood impacts (injuries, infections, psychological stress) are visible, indirect health risks via compromised water, food, shelter, and healthcare are less well characterized, especially among vulnerable groups such as pregnant women in low- and middle-income countries (LMICs). These regions bear the majority of flood-related mortality and stillbirths and face compounding vulnerabilities including informal settlements, inadequate WASH infrastructure, and socioeconomic constraints that reduce resilience. Prior evidence on maternal flood exposure and adverse pregnancy outcomes is limited by small samples, single events, or restricted geographies, and potential modifiers by socioeconomic and living conditions are poorly understood. Therefore, this study aimed to evaluate the association between gestational flood exposure and pregnancy loss (miscarriage and stillbirth) across multiple developing countries by linking maternal records from Demographic and Health Surveys (DHS) with spatial-temporal flood databases, and to quantify excess pregnancy losses attributable to gestational flood exposure from 2010 to 2020.

Existing studies document a range of flood-related health impacts in the general population, including psychological disorders, infectious and gastrointestinal diseases. Limited research has specifically examined pregnant women, although some studies suggest increased risks of adverse outcomes (e.g., miscarriage, stillbirth, and mental health effects) following disaster-related stress or specific flood events. Indirect mechanisms include compromised safe water, food security, and sanitation. However, prior epidemiologic data often rely on small samples, single incidents, or localized studies, leaving gaps regarding broader multi-country assessments and effect modification by socioeconomic and living conditions in developing countries.

Design: A multi-country matched case-control study leveraging repeated pregnancy outcomes from the same woman to control for time-invariant individual confounders. Data sources: Individual-level pregnancy histories were obtained from multiple Demographic and Health Surveys (DHS) conducted in developing countries. Flood exposure was derived primarily from the Dartmouth Flood Observatory (DFO) database (global verified flood events since 1985 with dates, affected areas, and main causes: heavy rainfall/monsoon, tropical cyclones, levee/dam failure, and snowmelt [excluded from subgroup analyses due to rarity]). For sensitivity analyses, the Global Flood Database (GFD; satellite-derived inundation extents, 2006–2014) was used as an alternative exposure source. Climate covariates (temperature and precipitation) during gestation were obtained from ERA5 reanalysis (0.25°×0.25° resolution). Exposure assessment: DHS clusters (villages/neighborhoods) were spatially matched to flood-affected areas. Gestational flood exposure (binary) was defined when a flood event’s affected area overlapped the cluster location during the pregnancy period. For matched controls (successful deliveries from the same mother), exposure windows were aligned to the gestational length of the corresponding case. Pre-conception exposure was also assessed in three windows: 0–3 months, 3–6 months, and 6–9 months before conception, excluding overlaps into the subsequent window to avoid contamination. Statistical analysis: Conditional logistic regression (R, function clogit/glclogit; version 4.0.3) estimated odds ratios (ORs) and 95% CIs for pregnancy loss (overall, miscarriage, stillbirth) associated with gestational flood exposure. Models adjusted for maternal age at delivery year; month and year of conception (to control seasonality and long-term trends); number of previous pregnancy losses; and gestational-period mean temperature and precipitation modeled with natural splines (3 df each). Subgroup analyses examined effect modification by maternal age (<21, 21–25, 25–35, >35 years), pregnancy stage (early <4 months, mid 4–7 months, late ≥8 months), socioeconomic status (wealth quintiles), education (none/primary, secondary, higher), living conditions (urban/rural; water source: surface, intermediate, piped/tap; floor material: natural/rudimentary/finished), and flood duration (e.g., <6 days vs ≥6 days). Sensitivity analyses included: using GFD exposure; covariate reduction; leave-one-region-out analyses; and restricting to mothers residing ≥10 years at the same address. Burden estimation: Excess pregnancy losses attributable to gestational flood exposure were calculated per 10,000 deliveries for 2010–2020 by linking gridded pregnancy counts from WorldPop (1 km aggregated to 10 km) to DFO flood events. Excess losses for each grid-year were estimated using the OR as an approximation of relative risk (given low prevalence of pregnancy loss), multiplied by the number of flood events and the grid’s pregnancy counts, and then aggregated by country and flood type (heavy/monsoon rains, tropical cyclones, levee/dam failure).

- Gestational flood exposure was associated with increased pregnancy loss risk: OR 1.08 with 95% CI reported as 1.04–1.11 in the abstract and 1.01–1.11 in results text. - Miscarriage specifically showed elevated odds: OR 1.05 (95% CI: 1.02–1.08). Stillbirth risk also increased (text indicates positive association). - Pre-conception flood exposure was associated with increased pregnancy loss when occurring 0–3 months (OR 1.03, 95% CI: 1.01–1.06) and 3–6 months (OR 1.03, 95% CI: 1.01–1.06) before pregnancy; no significant association for 6–9 months prior (OR 1.03, 95% CI: 0.85–1.21). - Flood duration modified risk: prolonged exposure (≥6 days) OR 2.00 (95% CI: 1.83–2.13); brief exposure (<6 days) OR 1.02 (95% CI: 0.76–1.30; not significant). - Maternal age modified risk: higher ORs for <21 years (OR 1.12, 95% CI: 1.02–1.24) and >35 years (reported as 35–37, OR 1.17, 95% CI: 1.07–1.28); no significant associations for 21–25 years (OR 1.11, 95% CI: 0.98–1.26) and 25–35 years (reported OR 1.25, 95% CI: 0.95–1.09; not significant). - Pregnancy stage: highest risks during mid-pregnancy (OR 1.07, 95% CI: 1.02–1.15) and late pregnancy (OR 1.05, 95% CI: 1.01–1.15); early pregnancy showed no significant risk (OR range 0.93–1.08). - Socioeconomic and living conditions: higher ORs among women with lowest wealth (OR 1.12, 95% CI: 1.01–1.23) and lowest education (OR 1.11, 95% CI: 1.01–1.20). Elevated risk for households relying on surface water and with rudimentary flooring compared with intermediate/piped water and finished floors. - Geographic patterns and burden: Highest proportion of excess pregnancy losses in South Asia; proportionally large burdens in Central America and the Caribbean, and South America. - Flood type contributions to excess losses (2010s): heavy/monsoon rains 89.84%; tropical cyclones 9.47%; levee/dam failure 0.70%. - Temporal trend (2010–2020): consistent upward trend in annual excess pregnancy losses associated with flood exposure. - Estimated annual excess pregnancy losses across 33 developing countries in the 2010s: approximately 53,944 to 148,345 per year (abstract); country- and grid-level maps show spatial heterogeneity.

The study demonstrates that gestational flood exposure increases the risk of pregnancy loss, addressing a key evidence gap regarding flood-related maternal and fetal health in LMICs. The observed effect modification highlights mechanisms by which floods exacerbate vulnerabilities: poorer housing quality (rudimentary flooring) may be structurally compromised during floods, increasing injury and hazardous exposure; reliance on unsafe water sources suggests pathways via water contamination and infections; and socioeconomic disadvantages (lower income and education) likely impede access to adequate prenatal care and resources during disasters. Age-related susceptibility (<21 and >35 years) aligns with known biological and healthcare access vulnerabilities, while heightened risk in mid-to-late pregnancy may reflect greater challenges in mobility, relocation, and sustained stress. Longer and multiple flood exposures likely compound physiological and psychological stress via displacement, resource loss, and healthcare disruption. The association with pre-conception exposure underscores persistent, indirect effects of floods on maternal health. Collectively, these findings reinforce the need for targeted flood preparedness, resilient WASH and housing infrastructure, and prioritized prenatal care for at-risk populations, particularly in regions with frequent flood events.

This multi-country matched case-control analysis provides robust evidence that gestational flood exposure significantly elevates the risk of pregnancy loss (miscarriage and stillbirth) in developing countries. By integrating DHS health records with flood databases and climate covariates, the study identifies vulnerable subgroups (younger and older mothers, low-SES households, unsafe water sources, rudimentary housing) and quantifies excess pregnancy losses across 2010–2020, showing a rising trend and dominant contribution from heavy/monsoon rains. These results underscore disparities in maternal and child health intensified by climate-related floods and support the development of localized, targeted protective measures, including strengthened WASH, resilient housing, and prioritized prenatal services during and after floods. Future research should refine country-specific risk estimates, improve exposure assessment accuracy, and evaluate the effectiveness of interventions to protect pregnant women during flood events.

- DHS cluster coordinates are intentionally displaced to protect privacy, potentially introducing exposure misclassification when matching to flood-affected areas. - Potential residual confounding from unmeasured time-varying maternal factors (e.g., chronic conditions, medications, changes in income, healthcare access, housing). - Uniform odds ratios were applied across countries for burden estimation due to limited country-specific sample sizes, which may mask geographic heterogeneity in risk. - Assumed constant annual counts of pregnant women per grid (2010–2020) because only a single year of gridded pregnancy data was available from WorldPop, potentially affecting temporal burden estimates. - Satellite-based GFD inundation data (used in sensitivity analyses) may miss events due to cloud cover and limited spatial coverage, though not used for primary estimates.