Improving methods for specifying target differences in sample size calculations for randomised trial of treatments for osteoarthritis

点击翻译按钮获取中文摘要

The first stage of the project will involve a systematic review of the existing literature. The review will examine the basis for the choice of target difference as reported in protocols of trials evaluating treatments of musculoskeletal conditions. Trial protocols provide a higher level of detail on sample size calculations than is reported in published trial results. It will provide evidence for whether trialists are using target differences much lower or higher than the MCID which may suggest that trialists are taking other factors into account, such as non-compliance. It will also highlight whether and how trialists take time into account when calculating important differences.During the second stage, the impact of time on the assessment of important differences in disease specific patient reported outcomes will be examined using time trade-off methodology. I will conduct a focus group of patients with musculoskeletal conditions in order to assess their opinions about whether their interpretation of whether a difference in important would depend on the time point of assessment. Time trade-off techniques involve presenting participants with scenarios which give different ways in which a treatment could influence the time course of a patient's condition. The findings will provide estimates of patients relative valuations of treatment benefits of varying duration.The third phase of the project will involve (a) assessing the stability over time of standard single time point approaches to calculating important differences and (b) comparing methods to calculate target differences which incorporate longitudinal data. First, existing single time-point methods will be applied to datasets from randomised controlled trials of a cognitive-behavioural approach in low back pain patients (BeST) and an exercise programme in patients with hand problems due to rheumatoid arthritis (SARAH) (Lamb 2010, Lamb 2015). Second, a simulation study will compare approaches for incorporating target difference in the sample size calculation and correspondinganalysis which utilise data from multiple time points. The approaches will be (1) single time point analysis (e.g. ANCOVA) separately for each time point, (2) mixed linear regression models, and (3) area-under-the-curve (AUC) analysis. These approaches will be compared regarding the level of power and significance, the effect of missing data from different levels of loss to follow-up, and the ease of interpretation. I will produce recommendations on specifying the target difference and reporting sample size calculations for randomised trials in musculoskeletal conditions using disease specific patient reported outcomes.