Mapping treatment effects on generic health-related quality of life scales from disease-specific measures using bivariate meta-analytic models and summary data.

BACKGROUND: Missing treatments effects on generic measures of health related quality of life (HRQoL) is a common limitation in economic evaluation. Mapping from disease-specific measures of HRQoL bridges this evidence gap. We aimed to estimate treatment effects on generic measures of HRQoL in ankylosing spondylitis (AnSp) and non-radiographic axial Spondylarthritis (nr-axSpA) and highlight the benefits for economic evaluation. METHODS: Summary measures of HRQoL were extracted from 25 randomised control trials (RCTs) of treatments for AnSp and nr-axSpA. We identified correlated pairs of disease-specific and generic measures. We mapped disease-specific to generic measures using multivariate meta-analytic surrogate endpoint models. Where generic measures were missing, they were predicted from the disease-specific measures. We compared predictions from bivariate random-effects meta-analysis (BRMA) models, the fixed-effects model by Daniels-Hughes (D&H), and the BRMA in product-normal formulation (BRMA PNF). Furthermore, we compared the pooled generic measures including and excluding predictions of the missing treatment effects. RESULTS: Prediction intervals for the generic measures - the 36-Item Short Form Survey Physical Component Summary (SF36-PCS) and the Mental Component Summary (SF36-MCS) - overlapped with 90% of the reported observed estimates. Estimates of the missing treatment effects were more precise with BRMA-PNF compared to D&H model. Pooling reported and predicted treatment effects using BRMA-PNF reduced uncertainty by up to 38% compared to univariate models. CONCLUSION: Aggregate level mapping with BRMA-PNF model provided the best fit to data and gave more precise estimates of relative treatment effects on the generic measures of HRQoL for this case study in AnSp and nr-axSpA.