Abstract
<title>Abstract</title> <p>Objectives This study empirically compared three recommended statistical models for estimating EQ-5D health state utilities (HSUs) using rheumatoid arthritis (RA) trial data and examined their impact on cost effectiveness analysis. Method Individual patient data from three RA trials were accessed through Vivli. Outcomes including EQ-5D and disease activity score-28 (DAS28) data were collected longitudinally over 24 weeks. Following the guidelines by ISPOR Task Force, we applied linear mixed-effects models (LMM), generalized estimating equations (GEE), and two-part mixed models (TPMM) to estimate HSUs for DAS-defined health states. Each trial data was analyzed separately. Estimated HSUs were incorporated into a published Markov model for RA. Incremental quality-adjusted life-years (QALYs) and incremental cost-effectiveness ratios (ICERs), and probabilistic sensitivity analysis (PSA) results were compared across models. Results The trials included 527, 684, and 584 patients contributing 2,576, 3,541, and 3,217 HSU observations, respectively. Approximately 40% of HSUs were in the moderate disease activity state across trials. LMM and GEE produced HSU estimates differing by < 0.002, resulting in differences of < 0.00005 in incremental QALYs and < 10,000 in ICERs in the Markov model. In contrast, TPMM consistently generated smaller HSUs and incremental QALYs, leading to higher ICERs. PSA showed comparable net monetary benefits and cost-effectiveness probabilities across models at common willingness-to-pay thresholds. Conclusion LMM and GEE produced similar HSU estimates, while TPMM yielded smaller values. Differences had minimal impact on cost-effectiveness conclusions in the selected Markov model. However, the generalisability of the findings was limited by the disease and model type used in our study.</p>