OBJECTIVES:

Health-related quality of life (HRQoL) data are routinely used to generate utilities for health technology assessment, with a range of methods available to calculate health state utility values (HSUVs). Values may depend on the questionnaire, value set, and statistical analysis performed. The Phase II VISION study (NCT02864992) of tepotinib in patients with NSCLC harboring METex14 skipping captured both EQ-5D-5L and EORTC QLQ-C30 responses. This study compares alternative HSUVs generated from VISION.

METHODS: EQ-5D utilities were derived from the UK EQ-5D crosswalk value set, and EORTC QLU-C10D utilities were derived from EORTC CLQ-C30 responses using the UK value set. To account for repeated measures, linear mixed models were fitted. Utilities were based on progression status (defined by investigator [INV] or independent evidence review committee [IRC]) and time to death (TTD). The correlation between paired observations of EQ-5D-5L and EORTC QLE-C10D was assessed using Pearson’s product moment correlation coefficient (PMCC).

RESULTS: From 273 patients with METex14 skipping NSCLC at the February 2021 data cut-off, 1,545 EQ-5D-5L and 1,546 EORTC QLQ-C30 responses were available. In general, EQ-5D produced higher estimated HSUVs than EORTC QLU-C10D. The difference between estimates was <0.02 (two decimal places) for all health states across models. Progression-based utilities were similar across progression definitions (INV, IRC), with differences only appearing at the third decimal place. EORTC QLU-C10D estimated greater utility in TTD categories closest to death (–0.0052 vs –0.0178). When comparing paired values, a statistically significant (p<0.001) correlation was found between EQ-5D and EORTC QLU-C10D HSUVs (PMCC: 0.693; 95% CI: 0.667, 0.718).

CONCLUSIONS:

A range of methods are available to generate HSUVs from HRQoL data. This analysis found high correlation between the well-established EQ-5D and novel QLU-C10D measures. Further research is required to understand when it is appropriate to use each method and the resulting impact on economic modeling.