The primary analysis was by intention to treat (ITT), comparing BMI (kg/m2) between study groups at 12 months postpartum. For all primary and secondary outcomes, multilevel linear models fitted cluster (maternity unit) and individual effects. BMI data were log-transformed for all regression analyses and baseline BMI as well as variables used to balance the randomisation included as covariates. The intervention effect for BMI was therefore interpreted as the percentage difference between groups. Two-level logistic models were used for categorical outcomes. The impact of individual demographic factors as well as theoretical mediators on the intervention effect was investigated (self-efficacy, social support, intrinsic motivation and self-regulation) [20 (link)]. Pre-specified subgroups were examined formally using interaction terms for parity, social class, ethnicity and smoking status. A complier average causal effect (CACE) analysis investigated the effect of intervention group attendance on the primary outcome. The influence of missing data was assessed using multiple imputation under a missing at random assumption. Sensitivity analyses were examined for departures towards missing not at random. All primary and secondary analyses were performed in IBM SPSS Statistics 20 and STATA v13, imputation was performed in STATA v13.
Free full text: Click here