PURPOSE: To compare several models which combine multiple health conditions to predict preference-based health-related quality of life (HRQoL) summary scores. The best fitting model should be used for the construction of catalogs from large datasets which list the impact of health conditions on health utility for use in cost-effectiveness analyses.

METHODS: We compare four commonly suggested models for the impact of multiple health conditions on HRQoL: additive, count, minimum, and multiplicative. We compare these models to fit HUI Mark3 scores from the Canadian Community Health Survey 1.1 (CCHS), SF-6D scores from the Medicare Health Outcomes Survey 1998-2004 (HOS), and EQ-5D and SF-6D scores from the Medical Expenditure Panel Survey 2000-2002 (MEPS). Summary scores were modeled using 15 health conditions (CCHS, HOS, MEPS) or 25 health conditions (CCHS, MEPS), with and without adjustment for age, sex, education, and income. All models were fit using WinBUGS 1.4.1 and model fit was assessed using Deviance Information Criterion.

RESULTS: We use 116,908 observations from CCHS, 95,191 observations from HOS, and 50,565 observations from MEPS. For all three datasets, the additive and multiplicative models fit better than the count and minimum models. The additive model provides the best fit for HUI Mark3 in CCHS. The multiplicative model provides the best fit for SF-6D in HOS. Without adjustment, the best fitting model for SF-6D and EQ-5D scores in MEPS is multiplicative, but the additive model fits these data best when adjusted for age, sex, education, and income.

CONCLUSIONS: The best fitting model is dependent on HRQoL summary score, dataset, and covariates included in the model. Analysts should test several models when developing catalogs of condition impact on HRQoL from large datasets.