Temperature-related neonatal deaths attributable to climate change in 29 low- and middle-income countries

Anthropogenic greenhouse gas emissions have driven unprecedented warming, with recent global mean temperatures ~0.95–1.2 °C above pre-industrial levels. Newborns are physiologically vulnerable to thermal stress due to immature thermoregulation, high surface-area-to-mass ratio, and susceptibility to infections, making both heat and cold potentially lethal. While neonatal deaths concentrate in LMICs, most temperature–neonatal mortality evidence comes from high-income, temperate settings, and no prior work quantified climate change’s contribution to temperature-related neonatal mortality. This study aims to attribute the burden of temperature-related neonatal deaths in 29 LMICs (2001–2019) to observed climate change using an impact attribution framework, combining DHS individual-level mortality data with multiple factual and counterfactual temperature reanalyses. The research assesses both overall neonatal (0–28 days) and very early neonatal (day 0) mortality, recognizing potentially distinct exposure pathways.

Prior studies document neonatal vulnerability to hypothermia and hyperthermia, and associations of maternal heat exposure with preterm birth and pregnancy complications. However, empirical evidence linking ambient temperature to neonatal mortality is mixed, and LMIC-specific analyses are scarce (only a few studies addressed LMICs). Impact attribution research has largely focused on natural systems, with comparatively limited attention to human health—particularly neonatal outcomes in LMICs. No previous work quantified climate change’s contribution to temperature-related neonatal mortality, highlighting a critical evidence gap this study addresses.

Design: Two-stage analysis within the ISIMIP impact attribution framework. Stage 1 estimated exposure–response functions between ambient temperature and neonatal mortality; Stage 2 quantified temperature-attributable burdens under factual versus counterfactual climate. Data sources: Individual-level neonatal deaths (0–28 days) from DHS-VII/VIII (2014–2021 rounds), restricted to events within 15 years before interview and with precise day/month/year of birth and death; PSU GPS coordinates used for exposure linkage. Total: 40,073 neonatal deaths across 29 LMICs (15,027 very early neonatal deaths). Temperature: Three ISIMIP3a surface temperature datasets: 20CRV3-ERA5, 20CRV3-W5E5, and GSWP3-W5E5 at 0.5° resolution. Counterfactual series created via detrending (ATTRICI) removing long-term climate change signals while preserving day-to-day variability and event timing. Absolute temperatures converted to PSU-specific percentiles to account for local adaptation. Stage 1 model: Time-stratified case-crossover with conditional logistic regression. Case days matched to 3–4 control days on same weekday within the same month/year. Exposure–lag–response modeled via DLNMs with natural splines: for neonatal mortality, temperature spline with one knot at 20th percentile; for very early neonatal mortality, one knot at 10th percentile; lag splines with one internal knot over a 0–2 day lag window (sensitivity up to 7 days). Minimum mortality temperature (MMT) identified from model predictions; temperatures below/above MMT classified as cold/hot. Air pollution and humidity not included due to limited confounding in prior literature and lack of daily data. Model selection via AIC and sensitivity analyses varying knot placement and recall period windows (5 and 10 years) showed robust results. Stage 2 attribution: For each country and dataset, computed attributable fractions (AF) for non-optimal temperatures using the backward attributable risk approach compatible with DLNMs, separating hot and cold components and further stratifying factual temperature into six ranges (extremely cold, moderately cold, mildly cold, mildly hot, moderately hot, extremely hot). Compared AFs and burdens between factual and counterfactual climates (2001–2019) to derive climate change-attributable excess (heat) and averted (cold) fractions and numbers. Converted AFs to rates per 100,000 live births using country-specific neonatal mortality rates (UNICEF). Very early neonatal rates derived by applying DHS-based shares to national neonatal mortality statistics. Uncertainty quantified via 10,000 Monte Carlo draws of model coefficients (multivariate normal), combining across the three datasets; 95% uncertainty intervals from the 2.5th and 97.5th percentiles.

- Temperature–mortality association: U-shaped exposure–response for both overall and very early neonatal mortality. Overall neonatal mortality risk was higher at lower temperatures; very early neonatal mortality risk increased more steeply at higher temperatures. Optimal temperatures (percentiles): neonates ~52nd–53.6th; very early neonates ~40.3rd–41.3rd. Absolute MMT spanned ~9–29 °C (neonates) and ~5–28 °C (very early), higher in warmer/equatorial locations. - Burden in factual climate (2001–2019, all 29 countries): 4.3% (95% UI: 1.7–6.8%) of all neonatal deaths were associated with non-optimal temperatures; heat 1.5% (0.2–2.6%), cold 2.9% (1.5–4.1%). Moderately hot and moderately cold temperatures dominated the burden. Countries with highest temperature-related neonatal mortality rates (>160 per 100,000): Pakistan, Mali, Sierra Leone, Nigeria. - Climate change attribution (overall neonatal): 32% (country range 19–79%) of heat-related neonatal deaths attributable to climate change, equalling 175,133 additional neonatal deaths (95% UI: 7,806–353,516) and representing a 46% increase versus a counterfactual without climate change. Largest proportions in Philippines (79%), Haiti (79%), Rwanda (70%). Climate change reduced cold-related neonatal deaths by 30% (10–63%), equalling 457,384 fewer deaths (95% UI: 170,106–868,519). - Rates (overall neonatal): Largest increases in heat-related neonatal mortality rates (>30 per 100,000) in Sierra Leone, Ethiopia, Liberia, Haiti. Largest reductions in cold-related rates (>110 per 100,000) in Liberia, Ethiopia, Sierra Leone, Uganda, Guinea. As fractions of all neonatal deaths: heat-related due to climate change ranged from 0.2% (Armenia) to 1.1% (Haiti); cold-related averted ranged from 0.3% (Albania, Nepal, Tajikistan) to 4.6% (Philippines). - Very early neonatal deaths (factual climate): 4.1% (95% UI: 1.6–6.5%) attributed to heat and 1.9% (0.2–3.5%) to cold. Countries with high very early temperature-related mortality rates included Liberia, Angola, Timor-Leste; moderately hot/cold ranges dominated. - Climate change attribution (very early): 29% (8–72%) of heat-related very early neonatal deaths attributable to climate change, totalling 168,835 additional deaths (95% UI: 48,835–296,467). Climate change reduced cold-related very early neonatal deaths by 35% (10–69%), equalling 141,322 fewer deaths (95% UI: 2,377–339,337). Largest increases in heat-related very early rates (>32 per 100,000) in Liberia, Timor-Leste, Sierra Leone, Ethiopia, Angola; cold-related averted >37 per 100,000 in Liberia, Rwanda, Uganda.

The study directly addresses the question of how much observed climate change has contributed to temperature-related neonatal mortality in LMICs. By combining DHS individual-level data with factual and counterfactual temperature series, the authors show that climate change has already increased heat-related neonatal deaths while reducing cold-related deaths. The dominance of moderately hot/cold exposures reflects their greater frequency versus extremes. The differential vulnerability profile—greater cold sensitivity for neonates overall but stronger heat sensitivity on day 0—aligns with distinct causal pathways: prematurity and intrapartum complications (linked to heat exposure during pregnancy) for very early deaths, versus infection-related causes (often associated with hypothermia) later in the neonatal period. Impacts were largest where baseline neonatal mortality is high and warming has been pronounced, notably in parts of sub-Saharan Africa. Findings are consistent with broader adult heat-attribution studies and extend the evidence base to newborns in LMICs with population-specific exposure–response functions. Public health implications include the need for improved thermal care practices, targeted education, context-appropriate equipment, and infrastructure and design adaptations to reduce thermal stress. With continued warming, heat-related burdens are likely to grow and may outweigh any reductions in cold-related mortality.

This study provides the first quantitative attribution of climate change impacts on temperature-related neonatal mortality across 29 LMICs, demonstrating increased heat-related and decreased cold-related neonatal deaths due to observed warming. Using robust epidemiologic models and multiple temperature datasets, it derives country-level burdens and uncertainties, highlighting substantial impacts in high-mortality, strongly warming settings. The results underscore the urgency of mitigation to limit future warming and of adaptation measures to protect pregnant women and newborns—improved postnatal thermal practices, context-suitable thermal equipment, kangaroo mother care, and heat-resilient homes and health facilities. Future research should project neonatal risks under warming scenarios, refine location-specific exposure–response functions with higher-resolution vital data, and further disentangle anthropogenic forcing contributions to observed impacts.

Key limitations include reliance on self-reported DHS birth histories with potential omissions, misclassification (e.g., very early neonatal deaths vs stillbirths), and recall errors; however, the case-crossover design mitigates time-invariant confounding, and recall-related misreporting is likely non-differential with respect to daily temperature. Pooling across countries using temperature percentiles accounts for local adaptation but limits flexibility in exposure–response shape across locations and may underestimate extreme heat impacts. Potential temporal changes in susceptibility or adaptation were not modeled. Air pollution and humidity were not included, though prior literature suggests modest or minimal confounding for temperature–mortality associations. Generalizability is affected by a large share of observations from India and underrepresentation of some regions (e.g., Latin America). Sampling weights could not be applied in the comparative risk assessment. The attribution follows IPCC impact attribution (climate-driven changes) but does not separate anthropogenic forcing from other drivers of climate trends, which would require additional modeling.