Statistical Methods for Analyzing EQ-5D in Randomized Clinical Trials - A Systematic Literature Review.
Journal Articles
Overview
Research
Identity
Additional Document Info
View All
Overview
abstract
OBJECTIVE: We conducted a systematic literature review to summarize the application of statistical methods for analyzing treatment effect on EQ-5D in RCTs. METHOD: We searched two electronic databases (MEDLINE and EMBASE, from inception through 2021) and www. CLINICALTRIAL: gov. Eligible studies were RCTs that analyzed post-baseline EQ-5D data by treatment group. Information on trial characteristics, EQ-5D data characteristics and statistical methods were extracted. Descriptive statistics were used to summarize results by dimension response, EQ VAS, and EQ-5D utility. RESULTS: A total of 2125 trials met the eligibility criteria. EQ-5D was commonly considered a secondary (n=1219, 57.4%) or exploratory (n=775, 36.5%) endpoint in RCTs. EQ-5D utilities were the most analyzed. Both utilities and EQ VAS were primarily analyzed in numerical format. The most common statistical models for analyzing utilities were linear fixed-effect model for single post-baseline (192/589, 32.6%) and linear mixed-effect model for multiple post-baselines (338/984, 34.3%). Of the 2054 studies that analyzed numerical EQ-5D, 221 (10.8%) examined model assumptions and 438 (21.3%) adjusted for the baseline score. Missing data were explicitly assessed in 661 trials, among which 347 (52.5% of 661) applied imputations, with the two most used imputation methods being multiple imputations (n=200, 57.6% of 347) and last observation carried forward (n=106, 30.5% of 347). CONCLUSION: This review found that health utilities are the most frequently analyzed EQ-5D data collected in clinical trials, followed by EQ VAS. Significant variation was observed in the selection of models, with most trials lacking adjustments for baseline data and appropriate methods for handling missing data.
