Mapping MacNew Heart Disease Quality of Life Questionnaire onto country-specific EQ-5D-5L utility scores: a comparison of traditional regression models with a machine learning technique.

Autor: Gao L; Deakin Health Economics, Institute for Health Transformation, Deakin University, Geelong, VIC, Australia. lan.gao@deakin.edu.au., Luo W; School of Information Technology, Deakin University, Geelong, VIC, Australia., Tonmukayakul U; Deakin Health Economics, Institute for Health Transformation, Deakin University, Geelong, VIC, Australia., Moodie M; Deakin Health Economics, Institute for Health Transformation, Deakin University, Geelong, VIC, Australia., Chen G; Centre for Health Economics, Monash University, Caulfield East, VIC, Australia.
Jazyk: angličtina
Zdroj: The European journal of health economics : HEPAC : health economics in prevention and care [Eur J Health Econ] 2021 Mar; Vol. 22 (2), pp. 341-350. Date of Electronic Publication: 2021 Jan 13.
DOI: 10.1007/s10198-020-01259-9
Abstrakt: Background: This study aims to derive country-specific EQ-5D-5L health status utility (HSU) from the MacNew Heart Disease Health-related Quality of Life questionnaire (MacNew) using both traditional regression analyses, as well as a machine learning technique.
Methods: Data were drawn from the Multi-Instrument Comparison (MIC) survey. The EQ-5D-5L was scored using 4 country-specific tariffs (United States, United Kingdom, Germany, and Canada). The traditional regression techniques, as well as a machine learning technique, deep neural network (DNN), were adopted to directly predict country-specific EQ-5D-5L HSUs (i.e. a direct mapping approach). An indirect response mapping was undertaken additionally. The optimal algorithm was identified based on three goodness-of-fit tests, namely, the mean absolute error (MAE), mean error (ME) and root mean square error (RMSE), with the first being the primary criteria. Internal validation was undertaken.
Results: Indirect response mapping and direct mapping (via betamix with MacNew items as the key predictors) were found to produce the optimal mapping algorithms with the lowest MAE when EQ-5D-5L were scored using three country-specific tariffs (United Kingdom, Canada, and Germany for the former and United Kingdom, United States, Canada and Germany for the latter approach). DNN approach generated the lowest MAE and RMSE when using the Germany-specific tariff.
Conclusions: Among different approaches been explored, there is not a conclusive conclusion regarding the optimal method for developing mapping algorithms. A machine learning approach represents an alternative mapping approach that should be explored further. The reported algorithms from response mapping have the potential to be more widely used; however, the performance needs to be externally validated.
Databáze: MEDLINE
Externí odkaz: