Causal inference and effect estimation using observational data

Causal analyses in longitudinal observational studies in oral health: A scoping review

OBJECTIVE

Causal analysis including causal inference and causal mediation is pivotal to inform effective interventions. In modern epidemilogy, causal analysis involves four key steps: formulating causal questions, employing directed acyclic graphs (DAGs), conducting data management and selecting statistical strategies. Our objective was to conduct a scoping review to assess how longitudinal observational studies (LOSs) in dental field have integrated these four steps to contribute leverage evidence that inform oral public health interventions.

METHODS

LOSs focusing on determinants of dental caries published from 2012 to 2024 were systematically retrieved from five major databases. The Joanna Briggs Institute-scoping review guidance and the Covidence application were employed to identify eligible LOSs for being reviewed.

RESULTS

Out of the 85 eligible LOSs, none formulated causal hypothesis by applying 'what if' question or investigated mediation across three levels of the determinants of oral health. A minority (18 studies, ~21.2%) employed DAGs to visualise relationships among study variables, while only one third (33 studies, ~39%) clearly defined confounders. The majority (64 studies, ~75%) incorporated a time-varying feature of their data, yet only a few (11 studies) fully leveraged this advanced aspect. Among these studies that fully utilised time-varying data, more than half encountered challenges in employing robust statistics to address confounders arising from such data dynamics.

CONCLUSIONS

Dental LOSs have, to date, mostly focused on investigating associations over causality, often neglecting the four-step causal analysis and not fully utilising time-varying data. Researchers necessitate to shift their focus to causal inference and prioritise building capacity in causal analysis with a consistent four-step approach to advance the field. Studies exploring mechanisms linking determinants of dental caries across levels and leveraging time-varying data are strongly encouraged.

Challenges and Opportunities in Big Data Science to Address Health Inequities and Focus the HIV Response

PURPOSE OF REVIEW

Big Data Science can be used to pragmatically guide the allocation of resources within the context of national HIV programs and inform priorities for intervention. In this review, we discuss the importance of grounding Big Data Science in the principles of equity and social justice to optimize the efficiency and effectiveness of the global HIV response.

RECENT FINDINGS

Social, ethical, and legal considerations of Big Data Science have been identified in the context of HIV research. However, efforts to mitigate these challenges have been limited. Consequences include disciplinary silos within the field of HIV, a lack of meaningful engagement and ownership with and by communities, and potential misinterpretation or misappropriation of analyses that could further exacerbate health inequities. Big Data Science can support the HIV response by helping to identify gaps in previously undiscovered or understudied pathways to HIV acquisition and onward transmission, including the consequences for health outcomes and associated comorbidities. However, in the absence of a guiding framework for equity, alongside meaningful collaboration with communities through balanced partnerships, a reliance on big data could continue to reinforce inequities within and across marginalized populations.

Temperature Control Parameters Are Important: Earlier Preinduction Is Associated With Improved Outcomes Following Out-of-Hospital Cardiac Arrest

STUDY OBJECTIVE

Temperature control trials in cardiac arrest patients have not reliably conferred neuroprotective benefit but have been limited by inconsistent treatment parameters. To evaluate the presence of a time dependent treatment effect, we assessed the association between preinduction time and clinical outcomes.

METHODS

In this retrospective, single academic center study between 2014 and 2022, consecutive out-of-hospital cardiac arrest (OHCA) patients treated with temperature control were identified. Preinduction was defined as the time from hospital arrival to initiation of a closed-loop temperature feedback device [door to temperature control initiation time], and early door to temperature control device time was defined a priori as <3 hours. We assessed the association between good neurologic outcome (cerebral performance category 1 to 2) and door to temperature control device time using logistic regression. The proportion of patients who survived to hospital discharge was evaluated as a secondary outcome. A sensitivity analysis using inverse probability treatment weighting, created using a propensity score, was performed to minimize measurable confounding.

RESULTS

Three hundred and forty-seven OHCA patients were included; the early door to temperature control device cohort included 75 (21.6%) patients with a median (interquartile range) door to temperature control device time of 2.50 (2.03 to 2.75) hours, whereas the late door to temperature control device cohort included 272 (78.4%) patients with a median (interquartile range) door to temperature control device time of 5.18 (4.19 to 6.41) hours. In the multivariable logistic regression model, early door to temperature control device time was associated with improved good neurologic outcome and survival before [adjusted odds ratio (OR) (95% confidence interval) 2.36 (1.16 to 4.81) and 3.02 (1.54 to 6.02)] and after [adjusted OR (95% confidence interval) 1.95 (1.19 to 3.79) and 2.14 (1.33 to 3.36)] inverse probability of treatment weighting, respectively.

CONCLUSION

In our study of OHCA patients, a shorter preinduction time for temperature control was associated with improved good neurologic outcome and survival. This finding may indicate that early initiation in the emergency department will confer benefit. Our findings are hypothesis generating and need to be validated in future prospective trials.

Estimating the causal effects of income on health: how researchers' definitions of "income" matter

BACKGROUND

There is a well-established cross-sectional association between income and health, but estimates of the causal effects of income vary substantially. Different definitions of income may lead to substantially different empirical results, yet research is often framed as investigating "the effect of income" as if it were a single, easily definable construct.

METHODS/RESULTS

The aim of this paper is to introduce a taxonomy for definitional and conceptual issues in studying individual- or household-level income for health research. We focus on (1) the definition of the income measure (earned and unearned; net, gross, and disposable; real and nominal; individual and household; relative and absolute income) and (2) the definition of the causal contrast (amount, functional form assumptions/transformations, direction, duration of change, and timing of exposure and follow-up). We illustrate the application of the taxonomy to four examples from the published literature.

CONCLUSIONS

Quantified estimates of causal effects of income on health and wellbeing have crucial relevance for policymakers to anticipate the consequences of policies targeting the social determinants of health. However, much prior evidence has been limited by lack of clarity in distinguishing between different causal questions. The present framework can help researchers explicitly and precisely articulate income-related exposures and causal questions.

An AI assistant to help review and improve causal reasoning in epidemiological documents

Drawing sound causal inferences from observational data is often challenging for both authors and reviewers. This paper discusses the design and application of an Artificial Intelligence Causal Research Assistant (AIA) that seeks to help authors improve causal inferences and conclusions drawn from epidemiological data in health risk assessments. The AIA-assisted review process provides structured reviews and recommendations for improving the causal reasoning, analyses and interpretations made in scientific papers based on epidemiological data. Causal analysis methodologies range from earlier Bradford-Hill considerations to current causal directed acyclic graph (DAG) and related models. AIA seeks to make these methods more accessible and useful to researchers. AIA uses an external script (a "Causal AI Booster" (CAB) program based on classical AI concepts of slot-filling in frames organized into task hierarchies to complete goals) to guide Large Language Models (LLMs), such as OpenAI's ChatGPT or Google's LaMDA (Bard), to systematically review manuscripts and create both (a) recommendations for what to do to improve analyses and reporting; and (b) explanations and support for the recommendations. Review tables and summaries are completed systematically by the LLM in order. For example, recommendations for how to state and caveat causal conclusions in the Abstract and Discussion sections reflect previous analyses of the Study Design and Data Analysis sections. This work illustrates how current AI can contribute to reviewing and providing constructive feedback on research documents. We believe that such AI-assisted review shows promise for enhancing the quality of causal reasoning and exposition in epidemiological studies. It suggests the potential for effective human-AI collaboration in scientific authoring and review processes.

Gut and Urinary Microbiota in Cats with Kidney Stones

Upper urinary tract urolithiasis is an emerging disease in cats, with 98% of kidney stones composed of calcium oxalate. In humans, disturbances in the intestinal and urinary microbiota are suspected to contribute to the formation of calcium oxalate stones. We hypothesized that similar mechanisms may be at play in cats. This study examines the intestinal and urinary microbiota of nine cats with kidney stones compared to nine healthy cats before, during, and after treatment with the antibiotic cefovecin, a cephalosporin. Initially, cats with kidney stones displayed a less diverse intestinal microbiota. Antibiotic treatment reduced microbiota diversity in both groups. The absence of specific intestinal bacteria could lead to a loss of the functions these bacteria perform, such as oxalate degradation, which may contribute to the formation of calcium oxalate stones. This study confirms the presence of a distinct urobiome in cats with kidney stones, characterized by greater richness and diversity compared to healthy cats. These findings highlight the potential of microbiota modulation as a strategy to prevent renal lithiasis in cats.

The potential impact fraction of population weight reduction scenarios on non-communicable diseases in Belgium: application of the g-computation approach

BACKGROUND

Overweight is a major risk factor for non-communicable diseases (NCDs) in Europe, affecting almost 60% of all adults. Tackling obesity is therefore a key long-term health challenge and is vital to reduce premature mortality from NCDs. Methodological challenges remain however, to provide actionable evidence on the potential health benefits of population weight reduction interventions. This study aims to use a g-computation approach to assess the impact of hypothetical weight reduction scenarios on NCDs in Belgium in a multi-exposure context.

METHODS

Belgian health interview survey data (2008/2013/2018, n = 27 536) were linked to environmental data at the residential address. A g-computation approach was used to evaluate the potential impact fraction (PIF) of population weight reduction scenarios on four NCDs: diabetes, hypertension, cardiovascular disease (CVD), and musculoskeletal (MSK) disease. Four scenarios were considered: 1) a distribution shift where, for each individual with overweight, a counterfactual weight was drawn from the distribution of individuals with a "normal" BMI 2) a one-unit reduction of the BMI of individuals with overweight, 3) a modification of the BMI of individuals with overweight based on a weight loss of 10%, 4) a reduction of the waist circumference (WC) to half of the height among all people with a WC:height ratio greater than 0.5. Regression models were adjusted for socio-demographic, lifestyle, and environmental factors.

RESULTS

The first scenario resulted in preventing a proportion of cases ranging from 32.3% for diabetes to 6% for MSK diseases. The second scenario prevented a proportion of cases ranging from 4.5% for diabetes to 0.8% for MSK diseases. The third scenario prevented a proportion of cases, ranging from 13.6% for diabetes to 2.4% for MSK diseases and the fourth scenario prevented a proportion of cases ranging from 36.4% for diabetes to 7.1% for MSK diseases.

CONCLUSION

Implementing weight reduction scenarios among individuals with excess weight could lead to a substantial and statistically significant decrease in the prevalence of diabetes, hypertension, cardiovascular disease (CVD), and musculoskeletal (MSK) diseases in Belgium. The g-computation approach to assess PIF of interventions represents a straightforward approach for drawing causal inferences from observational data while providing useful information for policy makers.

Psychological burden associated with incident persistent symptoms and their evolution during the COVID-19 pandemic: a prospective population-based study

BACKGROUND

Identifying factors that predict the course of persistent symptoms that occurred during the COVID-19 pandemic is a public health issue. Modifiable factors could be targeted in therapeutic interventions.

OBJECTIVE

This prospective study based on the population-based CONSTANCES cohort examined whether the psychological burden associated with incident persistent symptoms (ie, that first occurred from March 2020) would predict having ≥1 persistent symptom 6-10 months later.

METHODS

A total of 8424 participants (mean age=54.6 years (SD=12.6), 57.2% women) having ≥1 incident persistent symptom at baseline (ie, between December 2020 and February 2021) were included. The psychological burden associated with these persistent symptoms was assessed with the Somatic Symptom Disorder-B Criteria Scale (SSD-12). The outcome was having ≥1 persistent symptom at follow-up. Adjusted binary logistic regression models examined the association between the SSD-12 score and the outcome.

FINDINGS

At follow-up, 1124 participants (13.3%) still had ≥1 persistent symptom. The SSD-12 score at baseline was associated with persistent symptoms at follow-up in both participants with (OR (95% CI) for one IQR increase: 1.42 (1.09 to 1.84)) and without SARS-CoV-2 infection prior to baseline (1.39 (1.25 to 1.55)). Female gender, older age, poorer self-rated health and infection prior to baseline were also associated with persistent symptoms at follow-up.

CONCLUSIONS

The psychological burden associated with persistent symptoms at baseline predicted the presence of ≥1 persistent symptom at follow-up regardless of infection prior to baseline.

CLINICAL IMPLICATIONS

Intervention studies should test whether reducing the psychological burden associated with persistent symptoms could improve the course of these symptoms.

Veal calves management in Québec, Canada: Part I. Association between passive immunity transfer at arrival and health disorders

Morbidity and mortality remain important concerns for veal production which traditionally involves very young calves being transported directly from dairy farms to calf-raising facilities or sold through auction markets from dairy farms. There are few studies that examine the relationship between transfer of passive immunity and health outcomes in veal production in Québec. The objective of this study was to investigate the risk factors associated with morbidity and mortality in milk and grain-fed veal facilities in Québec. Between October 2017 and December 2018, a prospective cohort study was conducted on 59 batches of milk- and grain-fed veal calves from Montérégie and Centre du Québec regions in Québec, Canada. A total of 30 calves per batch were randomly sampled for the transfer of passive immunity (TPI) status using the Brix refractometer (serum threshold ≥ 8.4%). Throughout the production cycle, arrival, treatment, mortality, and slaughter records of each batch of calves were extracted and used to quantify individual- and group-level risk factors. Morbidity and mortality were modelled through generalized linear mixed models (GLMMs) using a logit link, as function of categorical variables (individual inadequate TPI, arrival season, purchasing sites, and arrival weights) and a continuous variable (proportion of inadequate TPI in the batch). Given that the health and performance data are usually collected retrospectively under commercial settings, incomplete data was recognized as a potential issue. Thus, multiple imputation models were used. From 59 batches where 1729 calves were sampled for measuring TPI, 1084 calves had a serum Brix% < 8.4% giving a prevalence of 62.7% of inadequate TPI. The prevalence of morbidity and mortality in those 59 batches were not associated with the batch-level inadequate TPI prevalence. On the other hand, individual calves with inadequate TPI had higher odds of morbidity (OR: 1.56; 95% CI: 1.2-2.05, P = 0.001) compared to those calves with adequate TPI. Odds of mortality, on the other hand, were not significantly affected by inadequate TPI (OR:1.27; 95% CI: 0.74-2.18, P = 0.3). Calves arriving to the facility during the winter season had lower odds of mortality than those arriving in fall (OR:0.16; 95% CI: 0.03-0.78, P = 0.02). In conclusion, this study suggests that inadequate TPI, as assessed upon arrival in veal farms, remains an important concern for negative health outcomes in veal calves.

Emulating a Target Trial of Shorter Compared to Longer Course of Antibiotic Therapy for Gram-Negative Bacteremia

BACKGROUND

Despite the availability of antimicrobial therapies, gram-negative bacteremia remains a significant cause of morbidity and mortality on a global level. Recent randomized controlled trials support shorter antibiotic treatment duration for individuals with uncomplicated gram-negative bacteremia. The target trial framework using the cloning approach utilizes real-world data but eliminates the issue of immortal time bias seen in observational studies by emulating the analysis of randomized trials with full adherence.

METHOD

A hypothetical target trial allocating individuals with gram-negative bacteremia to either short antibiotic treatment duration (5-7 days) or longer antibiotic treatment duration (8-14 days) was specified and emulated using the cloning, censoring, and weighting approach. The primary outcome was 90-day all-cause mortality. Secondary outcome was a composite endpoint of clinical and microbiological relapse. The emulated trial included individuals from four hospitals in Copenhagen from 2018 through 2021.

RESULTS

In sum, 1040 individuals were included. The median age of the cohort was 76 years, the majority were male (54%), had community-acquired gram-negative bacteremia (86%), urinary tract infection as the source of the infection (78%), and Escherichia coli as the pathogen of the infection (73%). The adjusted 90-day risk difference in all-cause mortality was 1.3% (95% confidence interval [CI]: -.7, 3.3), and the risk ratio was 1.12 (95% CI: .89, 1.37). The adjusted 90-day risk difference in relapse was 0.7% (95% CI: -2.3, 3.8), and the risk ratio was 1.07 (95% CI: .71, 1.45).

CONCLUSIONS

We found comparative outcomes for shorter treatment duration compared to longer treatment duration in patients with gram-negative bacteremia.

Introduction to target trial emulation in rehabilitation: a systematic approach to emulate a randomized controlled trial using observational data

Rehabilitation providers and policymakers need valid evidence to make informed decisions about the healthcare needs of the population. Whenever possible, these decisions should be informed by randomized controlled trials (RCTs). However, there are circumstances when evidence needs to be generated rapidly, or when RCTs are not ethical or feasible. These situations apply to studying the effects of complex interventions, including rehabilitation as defined by Cochrane Rehabilitation. Therefore, we explore using the target trial emulation framework by Hernán and colleagues to obtain valid estimates of the causal effects of rehabilitation when RCTs cannot be conducted. Target trial emulation is a framework guiding the design and analysis of non-randomized comparative effectiveness studies using observational data, by emulating a hypothetical RCT. In the context of rehabilitation, we outline steps for applying the target trial emulation framework using real world data, highlighting methodological considerations, limitations, potential mitigating strategies, and causal inference and counterfactual theory as foundational principles to estimating causal effects. Overall, we aim to strengthen methodological approaches used to estimate causal effects of rehabilitation when RCTs cannot be conducted.

Do patients with prediabetes managed with metformin achieve better glycaemic control? A national study using primary care medical records

AIMS

To estimate the effectiveness of metformin on glycaemic parameters among participants with incident prediabetes attending Australian general practices.

METHODS

This retrospective cohort study used electronic health records of regular participants (3+ visits in two consecutive years) attending 383 Australian general practices (MedicineInsight). Participants with 'incident' prediabetes (newly recorded diagnosis between 2012 and 2017) and their glycaemic parameters (haemoglobin A1c [HbA1c] or fasting blood glucose [FBG]) at 6-, 12-, and 18-24 months post diagnosis (unexposed) or post-management with metformin (treatment) were identified from the database. We estimated the average treatment effect (ATE) of metformin management on glycaemic parameters using both linear regression and augmented inverse probability weighting.

RESULTS

Of the 4770 investigated participants with 'incident' prediabetes, 10.2% were managed with metformin. Participants on metformin had higher HbA1c levels at the baseline than those unexposed (mean 45 mmol/mol [6.2%] and 41 mmol/mol [5.9%], respectively), but no differences were observed at 6-12 months (mmol/mol ATE 0.0, 95% CI -0.4; 0.7) or 12-18 months (ATE -0.3, 95% CI -1.2; 0.3). However, participants on metformin had lower mean HbA1c mmol/mol at 18-24 months (ATE -1.1, 95% CI -2.0; 0.1) than those unexposed. Consistent results were observed for FBG (ATE at 6-12 months -0.14 [95% CI -0.25; -0.04], 12-18 months 0.02 [95% CI -0.08; 0.13] and 18-24 months -0.07 [95% CI -0.25; 0.12]).

CONCLUSION

The higher HbA1c and FBG baseline levels among participants with 'incident' prediabetes managed with metformin improved after 6-12 months of starting pharmacological management, and the effect persisted for up to 24 months. Management with metformin could prevent further deterioration of glycaemic levels.

Impact of intensive adherence counseling on viral load suppression and mortality among people living with HIV in Kampala, Uganda: A regression discontinuity design

Intensive adherence counseling (IAC) is recommended for people living with HIV (PLHIV) with viral load (VL) ≥1,000 copies/ml after ≥6 months of anti-retroviral therapy (ART). We evaluated the effect of IAC on VL suppression and all-cause mortality among PLHIV on first-line ART with VL ≥1,000 copies/ml after ≥6 months of ART in Kampala, Uganda using regression discontinuity design, a quasi-experimental method for effect estimation when interventions depend on a cut-off. PLHIV just above VL ≥1,000 copies/ml cut-off who received ≥3 IAC sessions formed the intervention group while those just below the cut-off who received routine psychosocial support constituted the control group. Primary outcome was repeat VL suppression defined as VL <1,000 copies/ml approximately 9-12 months following initial VL assessment. Secondary outcome was all-cause mortality. We used logistic regression for causal-effect analysis, reported as odds ratio (OR) with a 95% confidence interval (CI). We performed sensitivity analyses to assess the robustness of findings to varying bandwidths at the cut-off. We found 3,735 PLHIV were started on ART between Nov 2020 and Nov 2021 of whom 3,199 were included in the analysis (3,085 control, 114 intervention). Within an optimal bandwidth, there were 236 participants (222 control, 14 intervention) with similar demographic and clinical characteristics. Repeat VL suppression was lower in the intervention than in the control group (85.7% versus 98.6%, p = 0.021) while all-cause mortality was similar (0% versus 0.5%, p = 1.000). In multivariable analysis, the odds of repeat VL suppression were 91% lower in the intervention than control group (OR = 0.09; 95% CI, 0.01-0.66). Findings are robust to varying bandwidths around the cut-off. We concluded IAC is ineffective in suppressing VL among PLHIV on first-line ART in Kampala, Uganda. Findings suggest a need to investigate the IAC implementation fidelity for successful translation in practice and the reasons for VL persistence beyond the suppression threshold.

Current practices in studies applying the target trial emulation framework: a protocol for a systematic review

INTRODUCTION

Observational studies represent an alternative to estimate real-world causal effects in the absence of available randomised controlled trials (RCTs). Target trial emulation is a framework for the application of RCT design principles to emulate a hypothetical open-label RCT (the hypothetical target trial) using existing observational data as the primary data source as opposed to the prospective recruitment and measurement of randomised units. The aim of this systematic review is to investigate the practices of studies applying the target trial emulation framework to evaluate the effectiveness of interventions.

METHODS AND ANALYSIS

We will systematically search in Medline (via Ovid), Embase (via Ovid, entries from medRxiv are included), PsycINFO (via Ovid), SCOPUS, Web of Science, Cochrane Library, the ISRCTN registry and ClinicalTrials.gov for all study reports and protocols which used the trial emulation framework (without time restriction). We will extract information concerning study design, data source, analysis, results, interpretation and dissemination. Two reviewers will perform study selection, data extraction and quality assessment. Disagreements between reviewers will be resolved by a third reviewer. A narrative approach will be used to synthesise and report qualitative and quantitative data. Reporting of the review will be informed by Preferred Reporting Items for Systematic Review and Meta-Analysis guidance (PRISMA).

ETHICS AND DISSEMINATION

Ethical approval is not required as it is a protocol for a systematic review. Findings will be disseminated through peer-reviewed publications and conference presentations.

Foundation models for generalist medical artificial intelligence

The exceptionally rapid development of highly flexible, reusable artificial intelligence (AI) models is likely to usher in newfound capabilities in medicine. We propose a new paradigm for medical AI, which we refer to as generalist medical AI (GMAI). GMAI models will be capable of carrying out a diverse set of tasks using very little or no task-specific labelled data. Built through self-supervision on large, diverse datasets, GMAI will flexibly interpret different combinations of medical modalities, including data from imaging, electronic health records, laboratory results, genomics, graphs or medical text. Models will in turn produce expressive outputs such as free-text explanations, spoken recommendations or image annotations that demonstrate advanced medical reasoning abilities. Here we identify a set of high-impact potential applications for GMAI and lay out specific technical capabilities and training datasets necessary to enable them. We expect that GMAI-enabled applications will challenge current strategies for regulating and validating AI devices for medicine and will shift practices associated with the collection of large medical datasets.

Prenatal exposure to long-term heat stress and stillbirth in Ghana: A within-space time-series analysis

INTRODUCTION

Few studies examined the association between prenatal long-term ambient temperature exposure and stillbirth and fewer still from developing countries. Rather than ambient temperature, we used a human thermophysiological index, Universal Thermal Climate Index (UTCI) to investigate the role of long-term heat stress exposure on stillbirth in Ghana.

METHODS

District-level monthly UTCI was linked with 90,532 stillbirths of 5,961,328 births across all 260 local districts between 1^st January 2012 and 31^st December 2020. A within-space time-series design was applied with distributed lag nonlinear models and conditional quasi-Poisson regression.

RESULTS

The mean (28.5 ± 2.1 °C) and median UTCI (28.8 °C) indicated moderate heat stress. The Relative Risks (RRs) and 95% Confidence Intervals (CIs) for exposure to lower-moderate heat (1st to 25th percentiles of UTCI) and strong heat (99th percentile) stresses showed lower risks, relative to the median UTCI. The higher-moderate heat stress exposures (75th and 90th percentiles) showed greater risks which increased with the duration of heat stress exposures and were stronger in the 90th percentile. The risk ranged from 2% (RR = 1.02, 95% CI 0.99, 1.05) to 18% (RR = 1.18, 95% CI 1.02, 1.36) for the 90th percentile, relative to the median UTCI. Assuming causality, 19 (95% CI 3, 37) and 27 (95% CI 3, 54) excess stillbirths per 10,000 births were attributable to long-term exposure to the 90th percentile relative to median UTCI for the past six and nine months, respectively. Districts with low population density, low gross domestic product, and low air pollution which collectively defined rural districts were at higher risk as compared to those in the high level (urban districts).

DISCUSSION

Maternal exposure to long-term heat stress was associated with a greater risk of stillbirth. Climate change-resilient interventional measures to reduce maternal exposure to heat stress, particularly in rural areas may help lower the risk of stillbirth.