How well do mothers recall their own and their infants’ perinatal events? A two-district study using cross-sectional stratified random sampling in Bihar, India

OBJECTIVE: Global monitoring of maternal, newborn and child health (MNCH) programmes use self-reported data subject to recall error which may lead to incorrect decisions for improving health services and wasted resources. To minimise this risk, samples of mothers of infants aged 0–2 and 3–5 months are sometimes used. We test whether a single sample of mothers of infants aged 0–5 months provides the same information. DESIGN: An annual MNCH household survey in two districts of Bihar, India (n=6 million). PARTICIPANTS: Independent samples (n=475 each) of mothers of infants aged 0–5, 0–2 and 3–5 months. OUTCOME MEASURES: Main analyses compare responses from the samples of infants aged 0–5 and 0–2 months with Mantel-Haenszel-Cochran statistics using 51 indicators in two districts. RESULTS: No measurable differences are detected in 79.4% (81/102) comparisons; 20.6% (21/102) display differences for the main comparison. Subanalyses produce similar results. A difference detected for exclusive breast feeding is due to premature complementary feeding by older infants. Measurable differences are detected in 33% (8/24) of the indicators on Front Line Worker (FLW) support, 26.9% (7/26) of indicators of birth preparedness and place of birth and attendant, and 9.5% (4/42) of the indicators on neonatal and antenatal care. CONCLUSIONS: Differences in FLW visits and compliance with their advice may be due to seasonal effects: mothers of older infants aged 3–5 months were pregnant during the dry season; mothers of infants aged 0–2 months were pregnant during the monsoons, making transportation difficult. Useful coverage estimates can be obtained by sampling mothers with infants aged 0–5 months as with two samples suggesting that mothers of young infants recall their own perinatal events and those of their children. For some indicators (eg, exclusive breast feeding), it may be necessary to adjust targets. Excessive stratification wastes resources, does not improve the quality of information and increases the burden placed on data collectors and communities which can increase non-sampling error.