Technical report: Trial experience and data capture in the Low Birth Weight South Asia Trial, a large cluster-randomised controlled trial in lowland Nepal

Objectives: i) to describe data capture in the Low Birth Weight South Asia Trial (LBWSAT) and factors affecting it; ii) to analyse to what extent differential data capture created bias in the available data. Methods: We describe the context, study design, data collection instruments used and their capture rates. Little of the data available were eligible for trial analyses, so use of the data for secondary analyses is important. Data capture was affected by data collector overload, pressure to enrol women in the food and cash transfer arms, delayed receipt of participant ID cards, enrolment of women at any gestational age (including after delivery at the start), in-migration into the food/cash arms to access transfers, political instability, conflict in the field team, logistical issues, establishment of a run-in period, hiatus of data collection during training, and lack of funds to extend the duration of the study. To assess the extent that differential data capture generated bias we described background characteristics by study arm and in-migration status. Then for each of the main data collection instruments we compared captured and not-captured enrolled women’s age, age at marriage, wealth score and height using t-tests and enrolled women’s and husband’s education using chi squared tests. Using mixed logistic regressions (adjusted for clustering using random effects) we assessed the odd of questionnaire capture in relation to these factors. Results: Small differences between captured and non-captured women were found. In-migrators were more prevalent in the cash/food transfer arms and compared with permanent residents were more likely to be living in their parental homes, younger, primigravida, adolescent, Muslim, slightly poorer and have some education. Analyses of captured and non-captured women by questionnaire revealed small differences in age, age at marriage, wealth score and education but mostly these differences were very small. The largest differences were between captured and non-captured women in the endline cross-sectional survey, when slightly older, less educated, poorer women were more likely to be captured. These women more likely to report to a measuring station in their community. Conclusions: Many challenges in implementing large-scale trials in the plains of Nepal affect rates of data capture, especially when several timebound follow-up data collection occasions are needed. Although in-migrated and permanent residents, and captured and non-captured women differed slightly, the differences were not large enough to be of concern.