UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Impact of informative censoring on estimated treatment effects for confirmed disability worsening in multiple sclerosis clinical trials Lobay, Rachel

Abstract

In multiple sclerosis (MS) clinical trials, any elevation of a patient’s baseline Expanded Disability Status Scale (EDSS) score at a follow-up visit that is sustained or increased at the next visit is known as confirmed disability worsening (CDW). It is not clear how to handle patients who have an initial EDSS elevation (known as initial disability worsening (IDW)) immediately preceding dropout or at the last visit of the trial as they lack a confirmatory visit. Our objective was to assess the impact of different approaches to impute those missing CDW statuses on the estimated treatment effect in a time to first CDW analysis. Guided by a MS clinical trial dataset of 718 patients, we constructed a larger simulated dataset where all patients had EDSS scores from baseline to the last study visit. From this complete dataset, we obtained several samples of 300 patients and imposed dropouts on them by using probabilities that depended on whether IDW occurred at the immediately preceding visit. Then, 3 investigations were conducted to ascertain in what circumstances and to what extent different strategies for the imputation of the CDW statuses led to clearly different treatment effect results. First, we investigated the impact of magnifying the probabilities of dropout alone. In our next 2 investigations, we magnified the dropouts in conjunction with magnifying the treatment effect by modifying the EDSS data simulation procedure. In the first, we used Markov chain models to simulate the EDSS data, while in the second we incorporated more patient-specific information in the simulation through use of mixed-effects proportional odds models. We found that magnifying the dropouts alone had little impact on the estimated treatment effects. In contrast, under extreme treatment and dropout factors, the imputation strategies that accounted for treatment group led to estimated treatment effects that were clearly stronger than those from the other strategies in both the Markov chain and proportional odds model-based investigations. Since this was only observed under extreme circumstances, the imputation strategies are not expected to lead to clearly different estimated treatment effects in a time to first CDW analysis of a similar MS clinical trial.

Item Media

Item Citations and Data

Rights

Attribution-NonCommercial-NoDerivatives 4.0 International