Analysing intensive longitudinal data after summarization at landmarks: an application to daily pain evaluation in a clinical trial

Abstract

The paper addresses some of the key issues to be considered in analysing intensive longitudinal data after summarization at scheduled landmarks (i.e. prespecified times). In this context, the derivation of outcomes requires rigorous rules and the selection of covariates should be based on a thorough data exploration. To guide the choice of statistical approaches for inferences, we study the missingness mechanism by using a specific dropout model. Then, we compare and contrast statistical approaches based on direct modelling and on multiple imputation applied either to the raw data or to the derived outcomes. The results are interpreted in the light of the model constraints and the missingness mechanism assumption. We show that some statistical approaches based on multiple imputation applied to the raw data are particularly well adapted to our context as they avoid any loss of available information for missing data imputation. We also show that the influence of subjects with incomplete profiles can be described by using individual estimations given by appropriate statistical models. The motivating data set was collected in a double-blind placebo-controlled clinical trial to assess the effect on pain of a new compound in subjects suffering from fibromyalgia.