Nevin Manimala Statistics

Is EQ-5D-5L Better Than EQ-5D-3L Over Time? A Head-to-Head Comparison of Responsiveness of Descriptive Systems and Value Sets from Nine Countries

Pharmacoeconomics. 2022 Aug 5. doi: 10.1007/s40273-022-01172-4. Online ahead of print.


OBJECTIVES: The aim of this study was to compare the responsiveness of EQ-5D-3L (3L) with EQ-5D-5L (5L) descriptive systems and value sets in two independent samples (rehabilitation and stroke patients).

METHODS: Descriptive system results were compared cross-sectionally, and descriptive responsiveness was tested by calculating changed level responses (‘moves’) from baseline to follow-up, proportion of improved patients, Paretian Classification of Health Change (PCHC), and probability of superiority (PS). Responsiveness of values based on nine country-specific value sets was assessed by standardized response mean (SRM) and standardized effect size (SES). Relative efficiency of 5L over 3L was assessed by calculating ratios of the SRM and SES statistics.

RESULTS: Descriptive comparisons confirmed earlier evidence and showed a consistent overestimation of health problems in 3L. Descriptive responsiveness improved with 5L in terms of moves per respondent, proportions of improved patients and PS, whereas PCHC showed mixed results. Better value responsiveness statistics were observed for 5L in rehabilitation patients for all value sets. In stroke patients, 3L showed better responsiveness statistics compared with 5L. Relative efficiency results were moderately to strongly better with 5L for rehabilitation, and slightly to moderately better with 3L for stroke.

CONCLUSIONS: Descriptive results were the main driver of 3L-5L responsiveness differences. Responsiveness of 3L was influenced by the ‘confined to bed’ label and the overestimation bias of 3L, which affected all responsiveness results. This may impact quality-adjusted life-year (QALY) estimations, leading to over- or underestimations of QALYs gained, depending on the condition and condition severity. QALY calculations based on 5L data will result in more accurate estimates.

PMID:35930137 | DOI:10.1007/s40273-022-01172-4

By Nevin Manimala

Portfolio Website for Nevin Manimala