SF-6D POPULATION NORMS AND DISUTILITY WEIGHTS FOR SINGULAR AND CO-OCCURRING BEHAVIOURAL RISK FACTORS: EVIDENCE FROM TRADITIONAL AND MACHINE LEARNING MODELS
Author(s)
Nirmali Sivapragasam, MPP;
Queensland University of Technology, Brisbane, Australia
Queensland University of Technology, Brisbane, Australia
OBJECTIVES: To estimate contemporary Short Form Six-Dimension (SF-6D) population norms for Australian adults and quantify health-related quality of life (HRQoL) impacts of major behavioural risk factors, obesity, smoking, physical inactivity, and risky alcohol use, both individually and in combination, including potential non-linear and interaction effects relevant to economic modelling.
METHODS: Data were drawn from the nationally representative Household, Income and Labour Dynamics in Australia (HILDA) Survey. SF-6D utilities were derived from the SF-36. Multivariable regression models were used to estimate disutilities associated with behavioural risk factors and sociodemographic characteristics. Predicted utilities for behavioural health-state profiles defined by combinations of behavioural risks were generated separately for males and females. Gradient boosting and random forest models were estimated to explore potential non-linearities and interaction effects, with predictive performance assessed using cross-validated root mean squared error and mean absolute error.
RESULTS: Physical inactivity, obesity, and smoking were monotonically associated with lower HRQoL. In contrast, risky alcohol use was associated with near-neutral differences in utility after adjustment. Predicted utilities declined monotonically with increasing behavioural risk burden, with the lowest values observed among individuals with multiple co-occurring risk factors. While absolute utility levels differed by sex, the relative ordering and magnitude of utility differences across behavioural risk profiles were consistent for males and females. Machine learning models did not meaningfully improve predictive accuracy relative to regression models, but identified modest interaction effects involving physical activity, employment status, and body mass index.
CONCLUSIONS: This study provides contemporary Australian SF-6D population norms and utility estimates for combined behavioural risk profiles. By enabling direct parameterisation of multi-risk health states, these results support more realistic economic evaluation of lifestyle and prevention interventions.
METHODS: Data were drawn from the nationally representative Household, Income and Labour Dynamics in Australia (HILDA) Survey. SF-6D utilities were derived from the SF-36. Multivariable regression models were used to estimate disutilities associated with behavioural risk factors and sociodemographic characteristics. Predicted utilities for behavioural health-state profiles defined by combinations of behavioural risks were generated separately for males and females. Gradient boosting and random forest models were estimated to explore potential non-linearities and interaction effects, with predictive performance assessed using cross-validated root mean squared error and mean absolute error.
RESULTS: Physical inactivity, obesity, and smoking were monotonically associated with lower HRQoL. In contrast, risky alcohol use was associated with near-neutral differences in utility after adjustment. Predicted utilities declined monotonically with increasing behavioural risk burden, with the lowest values observed among individuals with multiple co-occurring risk factors. While absolute utility levels differed by sex, the relative ordering and magnitude of utility differences across behavioural risk profiles were consistent for males and females. Machine learning models did not meaningfully improve predictive accuracy relative to regression models, but identified modest interaction effects involving physical activity, employment status, and body mass index.
CONCLUSIONS: This study provides contemporary Australian SF-6D population norms and utility estimates for combined behavioural risk profiles. By enabling direct parameterisation of multi-risk health states, these results support more realistic economic evaluation of lifestyle and prevention interventions.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
P46
Topic
Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
No Additional Disease & Conditions/Specialized Treatment Areas