The 2023–24 Nigeria Demographic and Health Survey (NDHS) was conducted during a major macroeconomic shock in Nigeria—marked by fuel subsidy removal and currency devaluation. The authors examined whether the timing of fieldwork during this period was associated with antihypertensive treatment non‑uptake among adults previously diagnosed with hypertension. They report that timing of interviews was a significant predictor for women but not for men, indicating that contemporaneous economic conditions may have influenced treatment access or continuation for female respondents.
Analyses used the NDHS Individual Recode (women's) file and the Men's Recode file for 2023–24. The analytic sample included 2,975 women and 529 men aged 15–49 who reported a prior diagnosis of hypertension. The authors note that the NDHS data are available from the DHS Program under access restrictions to protect respondents' confidentiality; registration and approval are required to obtain the datasets.
The investigation followed a two‑stage approach. First, survey‑weighted logistic regression models were fit separately by sex. A pooled model tested sex‑by‑predictor interactions formally. Second, the study compared four predictive approaches—survey‑weighted logistic regression, elastic net, random forest, and gradient boosting (XGBoost)—all implemented with appropriate survey weighting. Model comparison relied on held‑out test set area under the receiver operating characteristic curve (AUC).
Timing of interview (measured as month during fieldwork) emerged as a statistically significant risk factor for non‑uptake among women (reported OR=1.14 per month; 95% CI 1.06–1.22; p<0.001). In pooled analyses, household wealth was associated with lower odds of non‑uptake (OR=0.59; p=0.012), with this association appearing to be driven primarily by women (women‑only OR=0.62; p=0.026). The authors extracted additional context from published literature and WHO reports but present the NDHS estimates as the basis for observed associations.
The study explicitly tested sex interactions with geopolitical zone. Results showed notable sex‑reversed patterns: zones that were protective for women were significant risk factors for men, with odds ratios in the reported range of 2.28–5.77 for men in those zones. Interaction testing across zones reached statistical significance (all reported p≤0.007), indicating meaningful heterogeneity in non‑uptake patterns by both sex and region.
Diabetes status behaved differently from other risk markers: it was protective for women (reported OR=0.34; p<0.001) and did not show a significant interaction with sex in the pooled model (interaction p=0.170). Wealth associations, as noted above, were significant in pooled samples and particularly evident among women. The report does not provide additional causal inference beyond these associations.
All three machine learning algorithms—elastic net, random forest, and XGBoost—produced similar discrimination for women, with held‑out test AUC values in the range reported by the authors (0.61–0.63). For men, predictive performance was lower and close to chance levels (AUC ~0.56–0.59). The authors attribute the poorer performance for men in part to the much smaller number of diagnosed men in the sample, which constrains model learning and validation.
From the NDHS‑based analyses the authors conclude there is a large disparity in hypertension treatment and management across sex, geopolitical zones, and macroeconomic conditions. They argue that a single, uniform national intervention strategy is unlikely to be effective given this heterogeneity. Proposed programmatic responses include designing interventions stratified by sex and zone, integrating hypertension screening into chronic disease platforms and maternal health touchpoints, and investigating the resilience of the pharmacy supply chain during economic shocks.
The authors state that the NDHS datasets used are accessible via the DHS Program subject to registration and approval. They report that all relevant ethical guidelines were followed, that participant consent and necessary approvals were obtained, and that clinical identifiers were protected. The authors declared no competing interests.
Notes on limitations and scope
All results, effect estimates, and comparative model metrics reported here are those provided by the source article. The preprint presents observational associations from a nationally representative household survey and model comparisons using the available NDHS sample; causal inference or external validation beyond the NDHS dataset is not reported in the source.