Validity of self-reported receipt of iron supplements during pregnancy: implications for coverage measurement

Background: Iron-deficiency anemia during pregnancy is an underlying cause of maternal deaths, and reducing risk through routine iron supplementation is a key component of antenatal care (ANC) programs in most low- and middle income countries. Supplementation coverage during pregnancy is estimated from maternal self-reports in population-based household surveys, yet recall bias and social desirability bias lead to errors of unknown magnitude. Methods: We linked data from household and health facility surveys from 16 countries to estimate input-adjusted coverage of iron supplementation during pregnancy. We assessed the validity of reported receipt of iron supplements in client exit interviews using direct observation as the gold standard across 9 countries with a recent Service Provision Assessment (SPA). Using a sample of 227 women who participated in the Nepal Oil Massage Study (NOMS), we also assessed the validity of self-reported receipt of iron folic acid (IFA) supplements. We used Poisson regression models to explore the association between client and health facility characteristics and agreement of self-reported receipt of iron supplements compared to direct observation. Results: Across the 16 countries, iron supplements were in supply at most of the 9215 sampled health facilities offering ANC services (91%). We estimated that between 48 and 93% of women attended at least one ANC visit at a health facility with iron supplements available. The specificity of recall of receipt of iron supplementation immediately following a visit was 79.3% and the sensitivity was 88.7% for the entire sample. Individual-level accuracy was high (Area under the curve > 0.7) and population bias low (0.75 < inflation factor < 1.25) across all countries. By contrast, in the NOMS sub-study, the accuracy of self-reported receipt of IFA supplements after 1-2 years was poor (sensitivity 86.1%, specificity 34.3%). Adjusted regression analyses indicated that older age and higher level of education were associated with poorer agreement between self-reports and direct observation. Conclusions: These findings suggest the need for caution when using self-reported measures with an extended recall period. Further validation studies using conditions similar to widely used population-based household surveys are warranted.