Evidence of Reporting Bias Within the Context of a Randomized Trial of Infant and Young Child Feeding Messages in Rural Kenya

Background Caregiver report of infant and young child feeding (IYCF) practices may be susceptible to reporting bias, particularly during intervention studies; the magnitude of this bias is unknown. Objective This analysis estimates IYCF reporting bias within a community‐based cluster‐randomized trial. Methods We are evaluating the impact of water, sanitation, hygiene (WSH), and nutrition interventions on child growth and development in rural Kenya. Households with consenting pregnant women answered questions about breastfeeding knowledge and attitudes. After enrollment, community clusters were randomized to WSH, nutrition, WSH+nutrition, or control groups; community‐based health promoters were trained on implementation activities and visited with participating households at least monthly to deliver study hardware and provide the behavior change intervention. In the two nutrition groups, standard IYCF practices, including early initiation of breastfeeding and exclusive breastfeeding through 6 months, were promoted during household visits. In a subsample of 1,471 women visited within 6 months after delivery, interviewers asked mothers about breastfeeding and complementary feeding practices. Because women in their second or third trimester could be recruited, some gave birth prior to the start of health promoter activities, introducing random variability in the timing of when women first heard the intervention messages, allowing us to examine potential reporting bias among those who delivered infants prior to hearing messages about initiation of breastfeeding. We compared reported early initiation by randomized groups and by timing of birth relative to the time of the first community meeting. Results The time from participant enrollment until the time of the first community meeting was a median of 7.0 wk (IQR: 6.7–7.8). At the enrollment visit, 78.0% of mothers reported that breastfeeding within the first hour after delivery was important. The reported prevalence of early initiation in the nutrition groups was 76.4% and in the non‐nutrition groups was 49.3% (p<0.001). Among women who delivered before the first community meeting (and thus had not yet heard breastfeeding messages prior to the birth of their child), the reported prevalence of early initiation in the nutrition groups was 75.1% and non‐nutrition groups was 48.6% (p<0.001). Among those who delivered after the first community meeting (and thus could have heard breastfeeding messages), the rates were 77.4% and 50.2%, respectively (p<0.001). Conclusions These data suggest that intervention messages were conveyed and understood in the study communities. Yet, all of the increases in reported early initiation practice could likely be attributed to reporting bias since those who had heard the early initiation messages prior to the birth of their infant were no more likely to have reported the target behavior than those who had not yet heard the messages by the time of the birth. Support or Funding Information Bill and Melinda Gates Foundation