Analysis of Randomized Controlled Trials with Missing Response

When women suffer from nausea and vomiting during pregnancy (NVP), Diclectin is commonly recommended as the prescribed medication. Recently, some doctors questioned its effectiveness. Therefore we started this practicum to test whether Diclectin is effective compared with placebo. This research study was based on a double-blind randomized controlled trial (RCT) with missing response. Due to non-negligible amounts of missing data in the longitudinal study, we will first apply Complete case analysis (CC), Last observation carried Forward (LOCF), replace with mean (MEAN) and Multiple Imputation (MI) to our dataset, then fit Linear Regression (LR), Linear Mixed Model (LMM) and Generalized Estimating Equation Model (GEE) to assess how inferences regarding treatment effects vary as a function of the particular method chosen. Our results showed that under the pregnancy unique quantification of emesis (PUQE) score scale, there was a greater improvement with Diclectin compared with placebo when LOCF was used for missing data at α = 0.05 level; But the improvement was not statistically significant when CC and MEAN were applied. Also, inferences vary for MI dataset if we use different modelling. Therefore, our statistical inferences about Diclectin efficacy depends on the missing data methods and models, and magnitude of the difference suggests the benefit of Dicelctin is not clinically important under the pre-specified minimal clinically important difference of 3 points. Our findings provide evidence on the inefficacy of the Diclectin and will serve as useful information for the Health Canada for providing better public health support to the pregnant women and new-born children.