Antenatal Iron-Folic Acid Supplementation Gaps in Nigeria: A Predictive Machine Learning Analysis of Non-Initiation and Sub-Optimal Duration Using NDHS 2023–24
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
Although iron-folic acid (IFA) supplementation is a known, low-cost intervention to prevent anemia in pregnancy, the Nigeria Demographic and Health Survey (NDHS) 2023–24 shows that 33% of women do not begin IFA supplementation, and the majority of women who start IFA supplementation do not complete it for the recommended duration. No nationally representative dataset from Nigeria has formally tested whether these are one or two distinct problems.
Methods
This is a cross-sectional secondary analysis of the NDHS Individual Recode - Women aged 15-49 (2023-24). Two gaps were identified: Gap A: Non-initiation of antenatal IFA supplementation (n=13,940); Gap B: Inadequate duration (<90 days) among those who initiated the supplementation (n=8,431). We compared the survey-weighted logistic regression model with elastic net, random forest, and XGBoost models, all of which included the normalised DHS sampling weight, and tested them on an equal 80:20 test-set split held out in Stata. The factors that were found to predict included: education, wealth, residence, geopolitical zone, religion, ethnicity, health insurance, media exposure, access barriers, parity, ANC visit count, timing of the ANC visits, and type of facility. We cross-checked self-reported IFA status with hemoglobin and anemia biomarkers from the same survey.
Results
Non-initiation prevalence was 33.8% and among initiators, 59.8% consumed iron for < 90 days. The single factor that dominated Gap A was women with no antenatal contact, who were 21 times more likely not to have initiated than those receiving home-based ANC (95% CI 11.6–38.6), while both public- and private-sector facility care were protective. Education, wealth, and ANC visit count were independently protective for both outcomes. Gap B showed a more diffuse predictor profile and weaker or reversed geographic effects relative to Gap A. XGBoost discrimination was higher in both cases (AUC 0.922 and 0.721) than in survey-weighted logistic regression (AUC 0.915 and 0.678), in contrast to the best comparator study in the literature, which found XGBoost to underperform simpler models for a different health outcome. The directionality of self-reported IFA status relative to the biomarkers (hemoglobin and anemia) was the same for both outcomes (p<0.05).
Conclusions
The IFA supplementation shortfall in Nigeria can be attributed to two issues: a large, highly concentrated access gap, primarily driven by the absence of antenatal contact, and a smaller, more diffuse adherence gap not adequately explained by survey variables. Plausibly, two different interventions are needed to close this gap: expanding the reach of antenatal services for non-initiation, and investigating side effects, supply continuity, and household barriers for non-adherence among antenatal service initiators, likely in a qualitative format.
