Extent, duration and predictors of exclusive breastfeeding in a longitudinal study: adjusting for missing data using an accelerated failure time model and multiple imputation

Background: The World Health Organization recommends at least 6 months of exclusive breastfeeding (EBF). Longitudinal studies facilitate estimation of EBF duration, but often suffer from loss to follow-up and missing information. The study estimates the prevalence of EBF, duration and predictors of EBF duration while adjusting for missing data using multiple imputation (MI). Methods: A longitudinal study was conducted on all women giving birth between September 2009-February 2010 in selected hospitals (N=2119). Data on EBF and socio-demographic and other characteristics were collected at birth, and at 2, 6, 12 and 24 months. Information on EBF status and duration was missing for 29%. To deal with missing data, we generated multiple datasets using logistic regression-based MI to impute missing EBF practice, and an accelerated failure time (AFT) model to impute missing duration of EBF. The latter model also identified factors associated with EBF duration. Results: The observed 64% of women practicing EBF (95%CI; 62%-66%) was adjusted, after imputation, to 62% (95%CI; 60%-65%). After imputation, the estimated median time of EBF among women practicing EBF was 4.9 months. Predictors of EBF duration were stated intention to breastfeed, religious observance, and giving formula milk while in hospital. Conclusion: Adjusting estimates of EBF practice and duration using MI is feasible and potentially important. Using an AFT model for EBF duration enables the execution of MI in such studies and allows direct interpretation of the impact of various factors on EBF duration.