The morning session will introduce the challenges with the analysis of PRO data collected over time and equip attendees with an understanding of the nature of missing data and key statistical models used for comparing PRO scores between randomized groups where the missing data is likely to be informative (related to the concept the PRO is measuring).
The afternoon session will then be a practical session where participants can practice applying the models using a clinical trial dataset and one of three software packages. Workshop facilitators will be present to support this hands-on analysis session.
Statisticians or researchers involved in the statistical analysis of clinical trial PROs. A basic understanding of regression modeling is recommended for the morning and afternoon sessions. Prior knowledge of at least one statistical analysis software package (Stata, R, SAS) is recommended for the afternoon session. No prior knowledge of missing data handling is required. Participants should bring their own laptop with statistical software installed.
The analysis of PRO data collected over time is not currently standardized. There are many models that could be used. However, to ensure the assumptions are met and an appropriate model selected, those making the selection require a good understanding of missing data on both its nature and impact. In this workshop, we will first describe what is considered missing data, how it can be mitigated prior to analysis, and discuss the role of intercurrent events. Next, participants will gain knowledge of the types of missing data and how to explore these data. Recent research on the relationship between missing data rates and bias can help participants understand ‘how much is too much’. In the final morning session, a suite of models will then be taught that can be applied to the analysis of longitudinal PRO data from a randomized clinical trial setting. The afternoon practical session will allow participants to work at their own pace, using a software package they are familiar with, analyzing a dataset with support from the facilitators.
1) To understand what constitutes missing data, potential impact and how to explore the missing data. The theoretical background to the field of missing data will be presented, to help participants appreciate the importance of considering this and how to mitigate missing data where possible. The distinction between missing data and intercurrent events will also be clarified. The different mechanisms and patterns of missing data will be introduced, covering practical approaches for exploring and visualizing the missing data prior to fitting analytical models. How the missing data rate impacts bias and efficiency and will show that, under certain conditions, researchers can make valid between-arm comparisons in randomized clinical trials with high missing data rates. Workshop participants will learn why they should document reasons for missing data, use all available data, and conduct analyses that make less restrictive assumptions about the missing data whenever possible.
2) To familiarize participants with the most commonly applied longitudinal models incorporating missing data assumptions. The final morning session will present a suite of longitudinal models for missing data. The mixed model for repeated measures, which assumes data is missing at random, will be covered with the following extensions: multiple imputation, joint models and pattern-mixture models. A focus of the presentation will be estimating the effect of treatment in a randomized clinical trial and equipping participants with the necessary tools for the afternoon session.
3) To gain practical experience of exploring missing data and fitting analytic models to randomized clinical trial data. Participants will apply the knowledge gained in the morning sessions to data from a clinical trial administering PRO data over time. The workshop facilitators will support participants in conducting an analysis with either SAS, R or Stata software, depending on each participant’s preference and access to software. This session will allow participants to gain a greater appreciation of the decisions to be made in practice, and any problems that can arise during analysis. The practical session will finish with a summary and discussion of the challenges found by participants and limitations of the models. The final session will briefly highlight other analysis strategies not covered in the workshop with a focus on future directions in the field.
|Period||19 Oct 2022|
|Event title||International Society for Quality of Life Research (ISOQOL) 29th Annual Conference: Redefining boundaries – breaking new ground in patient-centered outcomes research|
|Location||Prague, Czech Republic|