Stillbirth, which is often defined as the birth of a fetus with no signs of life, is a critical public health concern. The World Health Organization (WHO) uses a pragmatic definition for international reporting of stillbirths, which is a foetal death at more than equal to 28 completed weeks' of gestation or birthweight more than equal to 1000 g if gestational age is unknown. Global estimate of burden is 1.9–2.0 million stillbirths per year using the WHO definition (more than equal to 28 weeks' gestation), translating to roughly one stillbirth every 16 seconds.^1,2 Majority of the global stillbirth burden occurs in LMICs, concentrated in sub-Saharan Africa and South Asia. Progress in reducing stillbirth rates has consistently lagged behind reductions in neonatal and maternal mortality, although attention and SDG targets have improved.³

Many high-income country registries and public health statistics record foetal deaths from 20 weeks' gestation (or 350–500 g birthweight) and above, which produces a larger count of foetal deaths and different rate denominators. The International Classification of Diseases (ICD) and UN guidelines have therefore proposed to improve harmonisation where gestational age information is missing. Differences like these in 20, 22, or 28 weeks create a major source of divergence between data sources and complicate international and within-country comparisons.⁴

Clinically and legally, many systems treat pregnancy losses before 20 weeks as miscarriage/abortion, and those at or after 28 weeks as stillbirth. But losses occurring between 20 and 28 weeks are largely uncharacterised. That middle window includes many foetuses that may be viable with better antenatal care, timely detection and timely interventions, particularly when intensive neonatal support is available. In LMICs, much of this mid-pregnancy loss goes unrecorded and unmeasured. Furthermore, misclassification of early neonatal deaths as stillbirths and vice versa occurs in settings with limited capacity to assess signs of life at birth.⁵

Stillbirths have multi-factorial aetiologies, including maternal, foetal, placental, and environmental factors. Infections, hypertensive disorders of pregnancy, and foetal growth restriction are common causes. Socio-economic determinants, such as maternal age, education, wealth, and access to quality healthcare, play a significant role. Modifiable risk factors like maternal obesity, tobacco use, and short interpregnancy intervals have also been identified. In India, factors such as poor maternal nutrition, inadequate antenatal care, and home deliveries without skilled birth attendants contribute to the high stillbirth burden.⁶

IGME)Every newborn estimate, Lancet modelling) often yield higher numbers. Some reasons include inconsistent definitions of gestational-age cutoffs, misclassification of early neonatal deaths as stillbirths (or vice-versa), incomplete capture of foetal deaths in facility deaths, and survey questionnaire limitations that group miscarriages/abortions/stillbirths or do not reliably capture gestational age.^7,8 However, in the absence of complete civil registration and vital statistics (CRVS) in India, National Family Health Survey (NFHS) stands out as it is the only available nationally representative sample survey that also systematically captures pregnancy outcomes, including stillbirths, along with information on pregnancy duration reported by women, enabling gestational-age-specific classification at the population level. Additionally, the standardized electronic data collection process further enhances its usability when compared to other sources.⁹

In this study we aim to estimate stillbirths at different gestational time points, specifically during the sixth (more than equal to 20 weeks), seventh (more than equal to 24 weeks), and eighth (more than equal to 28 weeks) months of pregnancy using data from the NFHS. By disaggregating stillbirths across these gestational periods, we aim to capture the burden that may be missed by conventional cut-offs. Additionally, the study assesses the contextual maternal, socio-demographic, and health-system risk factors contributing to stillbirths in India, thereby providing evidence to inform targeted prevention strategies and improve perinatal outcomes.

The study used multiple rounds of nationally representative sample survey data in India, known as National Family Health Survey (NFHS: year 2005–06, 2015–16 and 2019–21) for the trend analysis whereas the study leveraged recent round i.e., 2019–21 for risk factor assessments. NFHS is a large-scale, multi-round survey conducted on representative samples from all states and union territories in India. It included a total of 724,115 (approximately 97% response rate), 699,686 (approximately 97% response rate) and 124,385 (approximately 95% response rate) women aged 15–49 years during 2019–21, 2015–16 and 2005–06 respectively.¹⁰ This data was obtained from the DHS data program portal following prior registration and permission. The dataset is publicly available and contains no identifiable information.¹¹

All women, irrespective of marital status, who had any history of birth or termination in past five years preceding the survey were considered for the burden analysis whereas recent births were considered for risk factor assessments. Those outcomes that were classified as stillbirth following the definition used as presented in Appendix Table S1. A total of 231,519, 252,973, and 57,867 total births at more than equal to 20 weeks of gestation (including live births and stillbirths) were considered for the burden assessment in the periods 2019–21, 2015–16, and 2005–06, respectively (this sample was re-stratified to get sample for more than equal to 24 and more than equal to 28 weeks of gestation). The risk factor assessment, a total of 176,215 recent total births having stillbirth at more than equal to 20-week gestation was included as pregnancy care related information was available only for recent total births in the NFHS.

Based on extensive literature review and availability of data elements in NFHS, this study broadly guided by a conceptual framework-Commission on Social Determinants of Health (CSDH).¹² The application of the CSDH framework have been broadly discussed and elaborated in a systematic review¹³ which highlights that health and mortality are influenced by structural factors like governance, macroeconomic policies, public policies, culture and societal values; as well as intermediary factors like material circumstances, psychosocial factors, behavioral/biological factors, and health systems. We adopted this framework and classified factors into three domains: individual (including healthcare), household, and community related factors that could broadly determine stillbirth. These variables were categorized as modifiable if existing literature/guidelines suggest that target interventions could alter the overall effect on outcome.¹⁴ For NFHS data, those modifiable risk factors and which emerged significant in final model of adjusted regression analyses were considered for further assessment. Appendix Table S2 elucidates all the risk factors.

The burden of stillbirth is calculated at three gestational age cut-offs: more than equal to 20 weeks, more than equal to 24 weeks, and more than equal to 28 weeks of gestation whereas risk factor assessment was done for more than equal to 20 weeks gestation. Although NFHS provides information on stillbirth in 2015 and 2019 variables, yet these two variable classifications were not used, considering the suboptimal quality (Appendix Table S3). Hence we have used vcal (reproductive calendar) variable from NFHS, which provides information from all sampled women about history of birth (B), termination (T), pregnancy (P) and contraceptive usage (0), for what is assigned for each month in the calendar period. Pregnancy following termination (which included all miscarriage, abortion and stillbirth) were reclassified to estimate the gestational age at the time of the outcome. Specifically, pregnancies ending as termination at more than equal to 20 weeks correspond to approximately 5 months of pregnancy, more than equal to 24 weeks correspond to approximately 6 months, and more than equal to 28 weeks correspond to 7 months stillbirths. Although late-trimester abortions/miscarriages represent smaller fraction, some degree of misclassification between foetal deaths and pregnancy terminations may have resulted in minimal residual misclassification bias. In

response to these, sensitivity analysis is done to assess the robustness of the estimations made to alternate stillbirth definition by applying alternative gestational age thresholds using continuous variable (more than equal to 6, more than equal to 7, and more than equal to 8 months). Additionally, a rounding-adjusted specification was implemented by extending to ±1-month gestations to evaluate the impact of rounding and heaping. The proportion of stillbirths reclassified under each scenario was comparable to the primary definition. The complex sampling design was adjusted during all analysis for setting the primary sampling unit and stratification variable using the svyset command in Stata.

Further, quality assessment included judging availability of missing data. Missing values ranged between 1 and 7% for variables such as caste (4.9%), number of antenatal checkups (ANC) (6.7%), timing of ANC (3.3%), height of women (2.5%), anaemia status (3.7%), cooking fuel (1.1%), drinking water (1.0%), floor material (4.8%), and type of toilet facility (5.0%). Considering these as critical variables, multiple imputation techniques were adopted to impute the missing data using non-missing background characteristics namely place of residence, wealth status and education.¹⁵

This analysis was done at national, state and district level. Geographical representation is state/district was graphically presented using QGIS version 3.44.3 software. District level assessment was not done for 2005–06 as no data were available whereas for 2015–16 and 2019–21 a total of 707 and 640 districts were assessed. To ensure comparability, district identifiers from earlier NFHS rounds were meticulously harmonised with the most recent Government of India district boundary framework. A limited reassessment is attached as Appendix Table S4 to enhance transparency and reproducibility. Further clustering in district level SBR was validated by using Moran's I, which is a commonly used measure of spatial autocorrelation. GeoDa software¹⁶ was used to calculate Local Moran's I, which provides insights on the spatial clustering of areas. This method evaluates whether districts with comparable or contrasting levels of SBR prevalence are spatially grouped. Essentially, Moran's I calculate the degree of association between the prevalence in one district and that of its neighbouring districts, thereby quantifying the extent of spatial clustering or dispersion. Local Indicators of Spatial Association (LISA) were utilized to pinpoint the exact locations of these clusters and detect spatial outliers, where areas with high incidence were surrounded by low-incidence areas and vice versa.^17,18 A first-order queen contiguity spatial weights matrix was applied to define neighbourhood structure for spatial autocorrelation analyses. Further, as crude district-level stillbirth rate (SBR) estimates may be unstable for a rare outcome due to small denominators, empirical Bayes (EB) smoothing¹⁹ was applied to district-level SBRs. Sensitivity analyses comparing spatial patterns derived from crude and smoothed rates are presented.

The study leveraged two strong statistical approaches to establish individual level risk factors. Risk factor assessment as exploratory assessments of correlates and prophylactically relevant markers rather than definitive causal determinants. Descriptive statistics were evaluated followed by bivariate logistic regression to present association between each risk factor and stillbirth. Later significant (P < 0.05) and biologically plausible variables were selected for the adjusted analyses. This was followed by a multivariable analysis in 3 different models. Multicollinearity between variables were assessed and variables exhibiting moderate co-variance (variance inflation factor [VIF] ≥ 2.5) (Appendix Table S5) were included within the final models. Results were presented using the standard template for presentation of descriptive, bivariate and multivariable analysis.²⁰

Secondly, we executed propensity score matching (PSM) analysis amongst only significant modifiable exposure variables to appraise regression results findings (as it mimics randomized controlled trials by creating a matched dataset where treated and untreated groups have similar distributions of covariates). PSM is beneficial in scenarios with substantial covariate imbalance or rare outcomes, and is often used as a bounding analysis alongside regression models.^21,22 Nearest neighbour matching is used. The reporting and interpretation for PSM is explained in detail in Appendix Table S6.

We obtained certificate of exemption (COE) from the Ethics committee of IIHMR, New Delhi (COE number: IIHMR/ERC/ 2023 /10) as in full compliance with the international guidelines for Human Research protection. Data used in this study are de-identified and publicly available upon request through the Demographic and Health Surveys (DHS) Program website (https://dhsprogram.com), subject to registration and approval.

None.

A total of 542,359 women who delivered in the five years preceding the three NFHS survey rounds (2005–06, 2015–16, and 2019–21) yielded stillbirth rates (SBR) of 12.8 (95% CI 10.7, 15.2), 16.2 (95% CI 13.6, 18.8), and 22.0 (95% CI 19.4, 24.7) per 1000 total births at more than equal to 28, more than equal to 24, and

more than equal to 20 weeks of gestation in India, respectively. Trend analysis revealed that the stillbirth rate during 2005–06 was 12.3 (95% CI 11.1, 13.5) per 1000 total births at more than equal to 28 weeks gestation, compared to 12.4 (95% CI 11.8, 13.0) and 13.8 (95% CI 13.0, 14.6) per 1000 in 2015–16 and 2019–21, respectively (Fig. 1; Appendix Table S7).

The relative reduction in stillbirths was highest (36.0%) between 2005–06 and 2015–16, while it declined modestly (8.4%) in the subsequent round (2015–16 to 2019–21). A similar pattern of decline was observed across all selected gestational age thresholds. Further a state/UT wise exploration of stillbirth burden at more than equal to 28 weeks gestation revealed that high stillbirth rate was reported mostly in larger states during 2005–06 (Fig. 2). During 2019–21 the high stillbirth rates were reported from states as depicted in Fig. 2; Appendix Table S8.

A state/UT wise burden for all the year is summarized in Appendix Table S8. During 2019–21, 19 out of 37 states/UTs (51.4%) reported a single-digit SBR, whereas 18 out of 37 (48.6%) and 0 out of 30 (0%) had SBR in single digit in 2015–16 and 2005–06 respectively. Details of states/UTs having SBR higher than national average are depicted in Appendix Table S9. Additionally, it was observed that approximately 40% (42.5% during 2019–21, 42.4% during 2015–16 and 41.2% during 2005–06) of all stillbirths in India are between 20 and 28 weeks of gestation. Moreover, state/UT wise assessment revealed that the contribution of SBR between more than equal to 20 weeks and less than equal to 28 weeks was higher than the national average in 38.9% (14/36), 45.7% (16/35), and 53.6% (15/28) of states during 2019–21, 2015–16 and 2005–06 respectively (Table 1).

Analysis of SBR at the district level revealed clusters of value observed not only in states/UTs having value above the national average but also scattered across various parts of the country (Fig. 3). A detailed SBR by district is attached as Appendix Table S10. Analysis on district-wise SBR also revealed that 63.9% of the districts have SBR in single digit at more than equal to 28 weeks gestation, whereas only 23.6% are reportedly having SBR in single digit at more than equal to 20 weeks gestation (Table 2).

The distribution of stillbirths occurring between 20 weeks and 28 weeks of gestation showed a similar variance between 2015–16 and 2019–21. Overall, the data showed a modest increase in districts with higher proportions of early stillbirths over the study period (Appendix Table S11).

Further crude estimates of Moran's I indicated statistically significant clustering in approximately 119 districts during 2015–16 and 140 districts during 2019–21 whereas significant clustering was observed in approximately 114 and 115 districts using the smoothed rates during 2015–16 and 2019–21 respectively. The district level Global Moran's I for crude estimates are 0.225 and 0.134 whereas smoothed rates yielded a Global Moran's I of 0.219 and 0.168 respectively during 2015–16 and 2019–21, indicating a moderate positive spatial autocorrelation. This suggests that SBR in one district are influenced by and share similarities to those in adjacent districts (Appendix Table S12).

Bivariate analysis results revealed that the overall direction and relative patterns across socio-

demographic groups and geographies were consistent. Additionally, alternative gestational cutoffs sensitivity analyses yielded minimal, insignificant differences, indicating that the estimates are robust to minor shifts in gestational age classification.

Burden of stillbirth amongst women with various risk factors are summarised in Table 3. Results from multivariable analysis showed that women with high risk fertility behaviour, height less than equal to 145 cm, delivery at private facilities, and those using unclean fuels, residing in urban areas had higher odds of stillbirths (Table 3, Appendix Table S13). aOR of stillbirth at private facility was further assessed and findings revealed that this is often times disorientated to be effect modification by place of residence. The association between place of delivery and stillbirth is significant in rural areas but not in urban areas (Appendix Table S14).

Based on the findings of multivariable regressions, PSM of the selected modifiable variables from final multivariate models except early neonatal deaths were run for individual level risk factor assessment i.e. propensity score matching analysis. Quality of matching and table presenting verification of estimates of PSM are available in Appendix Table S15, Fig. S1, S2. The propensity score matching (PSM) analysis (Appendix Table S16) revealed notable differences in the effect of selected modifiable risk factors. The attenuation of these associations after covariate balancing suggests sensitivity to model specification and residual confounding, indicating that these relationships should be interpreted cautiously as associative rather than causal.

The average treatment effect (ATE) estimates indicate that several factors are associated with small changes in the probability of stillbirth. Exposure to high-risk fertility behaviour increased the probability of stillbirth by 0.0014, while the use of unclean cooking fuel increased it by 0.0041, and small family size by 0.0015. Illiteracy showed a relatively larger positive effect (0.0072), and residing in rural areas increased the probability by 0.0029. Private facility delivery and maternal anaemia was also associated with a higher probability (0.0025, 0.0020 respectively). In contrast desired pregnancies (−0.0032) were associated with a lower average probability of stillbirth. Overall, the estimated population-level effects are modest in magnitude but indicate differential associations across maternal and household characteristics (Appendix Table S16). Common support result showed >99% of the data were on support that were included in the analysis (Appendix Table S17).

This study has estimated the trend of stillbirth rate at state level (2005 to 20 21) and 28 weeks gestation cutoff, and cross-sectional spatial distribution at the district level. We assessed a total of 542,359 women who delivered in the five years preceding the three survey rounds (2005–06, 2015–16, and 2019–21). The SBR has shown a decline post the 2005–06, although the trend from 2015-16 to 2019-21 has slowed down or reversed at some of the gestational age cut offs with 20–28 weeks consistently contributing to nearly 40% of stillbirths. This decline could be attributed to various national programs aimed at improving the availability, accessibility, affordability and utilisation of healthcare services. Study yielded stillbirth rates (SBR) of 12.8 (95% CI 10.7, 15.2), 16.2 (95% CI 13.6, 18.8), and 22.0 (95% CI 19.4, 24.7) per 1000 total births at more than equal to 28, 24, and 20 weeks of gestation in India, respectively. Additionally, the decline observed from 2005–06 to 2015–16 aligns with broader improvements in maternal and newborn care in India. However, the apparent plateau or rise in the most recent period should be interpreted cautiously. This pattern may partly reflect changes in reporting, recall dynamics, or survey implementation rather than a true epidemiologic reversal, especially given that NFHS-5 fieldwork overlapped with early pandemic-related disruptions. Moreover, other sources such as the Sample Registration System (SRS) generally report

unlikely to have significant effect on SBR without strengthening quality of facility-based intervention irrespective of the SBR. The effectiveness of community level interventions depends on various factors like acceptability of intervention in community, community readiness etc.²⁹ Purbey and colleagues²⁴ reported that burden by districts in northern, eastern and central India states, similar to our observation. These could largely be due to social determinants (poverty, low education, malnutrition) combined with limited access and poor utilization of healthcare services as these states which fall within the category of EAG i.e., Empowered Action Group.³⁰

The quality and availability of data on stillbirth remains suboptimal, highlighting the need for high-quality, granular, and standardized data systems with consistent and accurate reporting of gestational age.^5,31 Such improvements are essential for global comparisons. Further the data must be made more accessible and actionable for evidence-based decision-making at district/state level.

Though NFHS is valuable for describing patterns and associations, and many of the findings from risk factor assessment are plausible and consistent with the wider literature, temporality and residual confounding remain central constraints. Risk factor assessment

revealed High risk fertility behaviour (HRFB), illiteracy, delivering at private healthcare facility, using unclean fuel for cooking, residing in rural area and belonging to scheduled caste as risk factors. High rate at private facility appears to be effect modification by place of residence. The association between type of facility and stillbirths is significant in rural areas but not in urban areas. This suggests that in rural settings, private facility deliveries are associated with higher odds of stillbirth compared to public facilities, whereas in urban areas, the difference is not statistically significant. Further deliberation on this finding should consider possible referral patterns, selection differences, and access-related factors. Residing in joint family emerged as protective factor and High risk fertility behaviour (HRFB)^32,33 emerges as a significant risk factor implying the importance of age, birth spacing, and parity in adverse outcomes. However, after matching the dataset the SBR turned non-significant except. The link between healthcare readiness and the effect may be confounded by underlying socioeconomic and health related factors. Illiteracy also emerged as other social determinant, which could be due to overall poor health literacy during pregnancy and intrapartum period.^34,35 These findings are in line with other published literature^14,36-38 showing the central role of education in empowering women to recognize danger sign and seek timely care at appropriate healthcare facility. Biological factors like short maternal stature (<145 cm) and maternal anaemia emerged as factors influencing stillbirth. Short maternal stature reaffirm the intergenerational nature of risk of stillbirth and other associated adverse outcomes.^24,39,40 India carries a high burden of maternal anaemia with almost half of the pregnant women affected.¹⁰ Compliance to iron supplementation has remained as one of the biggest challenges to address this deficiency effectively.^41,42

Similar to our study existing evidence also report that use of unclean fuel like agricultural waste, cow dung, etc could increase the risk to adverse outcomes like stillbirth.⁴³ This is in reference to pollutants like carbon monoxide, particulate matter etc. that can impair placental function, leading to foetal hypoxia. Interestingly in line with existing literature this study observed that women residing in a joint family (more than equal to 5 members) had significantly lower risk of stillbirth.⁴⁴ It is plausible that women in nuclear families and other limited household in the importance of monitoring foetal movements during the late gestations periods translated into negative that a lack of awareness regarding decreased foetal movement is one of the most single factors women face their experiences stillbirths.⁴⁵ The protective associations observed for joint family structure and delayed pregnancy are particularly noteworthy in the Indian context, where family systems may influence social support, resource pooling, health-seeking behaviour, and buffering during obstetric complications. While exploratory, these findings warrant careful discussion and is indicative of a hypothesis for future research on the social ecology of stillbirth risk.

Place of residence, regional and caste-based inequalities were also evident in our study. Studies emphasize the importance of region-specific strategies to address localized risk factors. Such approaches can bridge gaps in care and ensure equitable access to

essential healthcare services.^46,47 Higher risk reported from private healthcare deliveries reflects a possible referral bias or differences in case-complexity.^38,48,49 Our study noted that there are associations with anaemia, short stature, unclean fuel, rural residence, and caste. These variables also represent a plausible marker of deprivation and structural disadvantage. Thus, social and household characteristics may operate through unmeasured clinical pathways (e.g., undiagnosed/uncontrolled gestational diabetes, maternal infections, or foetal growth restriction, which are not adequately captured in NFHS). Acknowledging these intermediate pathways will help prevent overinterpretation of these correlates as direct biological effects while retaining their programmatic relevance as markers of vulnerability.

This study has provided insights on the burden of stillbirth at different gestational age cutoffs at national, district and state level. Additionally, it demonstrates the proportion of uncounted foetal death that occurring late gestational age stillbirths. The use multi-pronged approach is another key strength of this study. For the burden assessment, study relied on quantifying the clustering of stillbirth burden using methods like Moran’s I. To address the issue of variance in empirical district-level SBR estimation for a relatively rare outcome, the sensitivity spatial analysis depending on EB results is presented. Further for risk stratification, the study did not rely solely on regression models, which provided a foundation for risk factor assessment by also reporting the potential confounders while examining relationship between exposure and outcome. Further, another strength of the study is use of another robust methods like PSM which mimic randomized controlled trial conditions by balancing observed covariates between exposed and unexposed groups.

While there are many strengths to the study, there exist certain limitations. Existing literatures raised questions on the data quality within India when compared between different reporting systems be it a routine or survey dataset. Owing to this limitation, there exist uncounted stillbirths and unclassified live stillbirth. This reclassification may have introduced minor misclassification, as months were aggregated from weeks (e.g., a late-week pregnancy or early-week delivery counted as a full month); however, this is expected to be minimal and overall observations are noted as a limitation. Sensitivity analyses demonstrated robustness to alternative gestational cutoffs, month-level aggregation (from weeks) may have introduced minor misclassification, particularly around boundary values due to rounding and heaping. Additionally, the calendar category "termination" may include induced abortions and miscarriages; while late-trimester abortions constitute a very small proportion in India, some residual misclassification between foetal deaths and pregnancy terminations cannot be entirely excluded. NFHS does not allow distinction between antepartum and intrapartum stillbirths, nor precise differentiation from very early neonatal deaths, potentially affecting outcome specificity. Recall bias in retrospective calendar histories and possible under-reporting of induced abortions due to stigma may have influenced estimates. The absence of detailed clinical covariates (e.g., hypertensive disorders, infections, foetal growth restriction) limits adjustment for biological pathways and residual confounding is likely. Unmeasured state-level factors (e.g. healthcare infrastructure, policy implementation) and district-level compositional factors may have unmeasured impacts on cluster formations. Though spatial harmonisation was undertaken to ensure comparability. Given this cross-sectional and observational design, findings should be interpreted as associative rather than causal. Future studies linking survey findings with prospective clinical data and explicit causal modelling frameworks (e.g., directed acyclic graphs) are needed to better untangle pathways and draw stronger causal inference.

In short, while the study highlights the counting only global recommendations would overlook a substantial burden of missing approximately 40% of total stillbirth burden. The study further reiterates that stillbirth is an important public health concern, with notable spatial and socio-demographic disparities in India. Despite progress, the persistently high stillbirth rates, particularly at earlier gestational ages underscore the need for strengthened maternal and newborn care. District-level clustering suggests concentrated areas of vulnerability requiring targeted interventions. Addressing preventable risk factors through comprehensive maternal and newborn care strategies is also critical for reducing the burden. The study calls for improved data systems with accurate gestational age reporting for evidence-based decision-making at district/state level.