📄 medicine

Intimate partner violence during pregnancy and risk of stillbirth, low birth weight, and early neonatal mortality: a pooled design-corrected analysis of 40 low- and middle-income countries

This pooled analysis of 40 low- and middle-income countries reveals that intimate partner violence during pregnancy is significantly associated with an increased risk of stillbirth, though not with low birth weight or early neonatal mortality, underscoring the need for design-aware inference and routine IPV screening in antenatal care.

Original authors: Xiaoyu Wu, Benqing Zheng

Published 2026-07-07

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Xiaoyu Wu, Benqing Zheng

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the world's health data as a massive library containing millions of stories about mothers and babies from 40 different countries. Usually, researchers read just one book (one country) or a few books from the same shelf (hospitals). But this study decided to read 40 different books at once to see if there's a common pattern.

The story they were looking for involves a very dark chapter: Intimate Partner Violence (IPV) during pregnancy. The researchers wanted to know: Does this violence hurt the baby before or right after birth? Specifically, they looked at three tragic outcomes:

Stillbirth: A baby born without life after 7 months of pregnancy.
Low Birth Weight: A baby born too small.
Early Neonatal Mortality: A baby dying within the first week of life.

Here is the breakdown of their findings, using simple analogies:

1. The "Heavy Backpack" Analogy (The Main Finding)

The researchers found a clear link between domestic violence during pregnancy and stillbirth.

Think of pregnancy as a journey up a mountain. When a mother experiences violence, it's like someone suddenly adding a heavy, jagged rock to her backpack.

The Result: The more rocks (types of violence) they add, the harder the climb becomes, and the higher the risk that the journey ends in tragedy (stillbirth) before reaching the top.
The Math: If a mother experienced violence, her odds of having a stillbirth were about 1.7 times higher than a mother who didn't.
The "Rock Count": It wasn't just about any violence; it was about the amount. If a mother faced physical, sexual, and emotional abuse all at once (three rocks), the risk was even higher than if she faced just one type.

2. The "Blurry Camera" Analogy (Why Other Results Changed)

The study also looked at Low Birth Weight and Early Neonatal Mortality. Here is where the story gets interesting.

In previous studies (reading just one book at a time), researchers thought violence made babies smaller or more likely to die in the first week. But this study used a special, high-definition camera lens called "design-corrected analysis."

The Old Way: Imagine looking at a crowd through a foggy window. You might think you see a pattern that isn't really there because the fog (statistical errors from how the survey was done) distorts the view.
The New Way: This study wiped the fog off the glass. When they looked clearly, the link between violence and low birth weight or early death disappeared. It turned out the "pattern" was mostly an illusion caused by how the data was collected, not a real biological cause-and-effect in this specific dataset.

3. The "Safety Net" Check (Is the result real?)

Because this is such a serious topic, the researchers played "Devil's Advocate" to make sure their finding about stillbirth wasn't a fluke.

The "One Country" Test: They took the data from one country out, then another, then another (like removing one puzzle piece at a time). The picture of the danger remained the same. The result wasn't just driven by one specific country; it was a global trend across the 40 nations.
The "Unseen Thief" Test (E-value): They asked, "How strong would a hidden, unmeasured factor have to be to explain away this result?" The answer was: "It would have to be a very powerful thief." This suggests the link between violence and stillbirth is robust and likely real, not just a coincidence.

4. The "Context" Check (Does the country matter?)

They wondered if living in a country where women have fewer rights (measured by the Gender Inequality Index) made the violence more dangerous for the baby.

The Finding: It didn't seem to matter. The risk of stillbirth was high whether the country had high or low gender inequality. The violence itself was the danger, regardless of the country's laws or culture.

Summary of the Story

What is true: In 40 low- and middle-income countries, if a pregnant woman is abused by her partner, her baby is significantly more likely to be stillborn. The more types of abuse she suffers, the higher the risk.
What is NOT true (in this specific study): The study did not find a statistically significant link between this violence and babies being born too small or dying in the first week, once they corrected for how the survey data was gathered.
The Takeaway: The study suggests that doctors and nurses should gently and safely ask pregnant women about violence during check-ups, because the risk of stillbirth is real. However, they also warn that when mixing data from many countries, you have to be very careful with your math (the "foggy window" analogy) so you don't mistake a statistical error for a medical fact.

Technical Summary: Intimate Partner Violence During Pregnancy and Perinatal Outcomes in 40 LMICs

Problem Statement
Intimate partner violence (IPV) during pregnancy is a prevalent issue in low- and middle-income countries (LMICs), yet existing evidence regarding its impact on perinatal outcomes—specifically stillbirth, low birth weight (LBW), and early neonatal mortality—largely relies on single-country studies, facility-based recruitment, or pooled analyses that fail to account for the complex, clustered, and weighted structure of national surveys. While previous meta-analyses suggest associations with adverse outcomes, there is uncertainty regarding whether these associations persist when analyzed at a pooled scale using design-aware inference. Furthermore, the role of country-level gender inequality (GII) as a potential moderator of these associations remains unexplored in the context of perinatal mortality.

Methodology
This study employed a pooled design-corrected analysis of the most recent Demographic and Health Surveys (DHS) from 40 LMICs that included a domestic-violence module (spanning DHS Phases 7 and 8).

Data Source: The analysis utilized women's and children's recode files, applying the module-specific domestic-violence weight ( $d005$ ) to ensure representativeness, as the module is administered to only one randomly selected woman per household.
Exposure Definition: IPV during pregnancy was defined as physical violence by a current or former partner during pregnancy. Cumulative exposure was measured by counting co-occurring forms of violence (physical, sexual, psychological), ranging from 0 to 3.
Outcomes:
- Stillbirth: Reconstructed from the reproductive calendar as a terminated pregnancy lasting $\ge$ 7 months.
- Low Birth Weight (LBW): Defined as measured birth weight < 2,500 g, with recalled size used as auxiliary information for imputation.
- Early Neonatal Mortality: Defined as death within the first 7 days of life.
Statistical Analysis:
- Primary Model: Survey-weighted logistic regression with country-stratified primary sampling units (PSUs) to yield cluster-robust confidence intervals, explicitly accounting for the pooled survey design.
- Complementary Model: Random-intercept multilevel logistic models (two-level for stillbirth; three-level for child outcomes) were fitted in parallel to assess structural consistency.
- Robustness Checks: The study included leave-one-country-out analyses, exposure-response modeling, Population Attributable Fraction (PAF) estimation via g-formula, E-value calculation for unmeasured confounding, multiple imputation for missing birth weights, and cross-level interaction tests with the UNDP Gender Inequality Index (GII).

Key Results

Prevalence: The weighted prevalence of pregnancy IPV across the 40 countries was 5.1%, varying significantly by country (1% to 15%).
Stillbirth Association: After adjusting for maternal age, education, wealth, residence, and parity, IPV during pregnancy was significantly associated with stillbirth in the design-corrected model (Adjusted Odds Ratio [aOR] 1.71; 95% CI 1.06–2.76).
- A dose-response relationship was observed: the odds of stillbirth increased by 18% (aOR 1.18; 95% CI 1.10–1.27) for each additional form of violence.
- The association remained robust across all 37 leave-one-country-out iterations (aOR range 1.41–1.81).
- The E-value for the point estimate was 2.81, suggesting an unmeasured confounder would need a risk ratio of ~2.8 with both exposure and outcome to explain away the result.
- The PAF was estimated at 3.4% (95% CI 0.6–7.3).
LBW and Neonatal Mortality: In contrast to the stillbirth findings, neither LBW (aOR 1.08; 95% CI 0.95–1.23) nor early neonatal mortality (aOR 1.12; 95% CI 0.94–1.34) showed a statistically significant association with IPV when the survey design was correctly modeled.
- Note: Multilevel models without full design correction produced nominally significant estimates for LBW and neonatal mortality, highlighting a methodological divergence where standard multilevel approaches may overstate precision for weaker outcomes.
Moderation: Country-level GII did not significantly modify the association between IPV and stillbirth (interaction $p = 0.93$ ).

Significance and Claims
The paper claims that its primary contribution is demonstrating a statistically significant association between pregnancy IPV and stillbirth across 40 LMICs when using design-aware inference, while simultaneously showing that this association does not extend to LBW or early neonatal mortality under the same rigorous analytical framework.

The authors emphasize a critical methodological lesson: pooled DHS analyses for weaker outcomes (like LBW and neonatal mortality) require design-corrected inference to avoid false positives that can arise from models that inadequately represent clustering and weighting. The findings support the integration of safe IPV enquiry into antenatal care protocols in LMICs. However, the authors maintain a modest stance, noting that the observational nature of the data precludes causal attribution and that the lower bounds of the confidence intervals for the stillbirth association and PAF are close to the null, indicating the association is consistent but moderate. The study concludes that while the link to stillbirth is robust, the lack of association with other outcomes suggests that previous positive findings in the literature may have been influenced by residual confounding or analytical limitations.