A Predictive Model To Include Uncertainty In Mean Health Utilities Estimation: Impact Of Social Class, Body Mass Index And Chronic Diseases

Abstract

Including second order uncertainty for utility values when building cost-effectiveness modelling is challenging due to its odds distribution. Our approach enables including uncertainty related to the effect of sociodemographic characteristics or chronic conditions on mean utilities included as input in different models. In Spain, the EQ-5D-5L was incorporated in the National Health Survey carried out between June 2011 and June 2012. The data of the survey were downloaded from the Spanish Statistical Office webpage. The following variables were considered in this study: sex, age, and social class (3 categories), body mass index from self-perceived height and weight and Euroqol 5D 5L questionnaire. For the estimation of the mean utilities we divided the process in two steps. First a logistic regression model was applied in order to estimate percentage (p) of the utilities which were equal to 1 based on individual characteristics. For those with not perfect health state the mean utility value was estimated using generalized linear model. In combination, this two regression models allow to estimate the mean utility value. In addition, using Cholesky decomposition of the variance-covariance matrix uncertainty on these parameters could be included in a mathematical model. This study included 20209 Spanish people, mean age51.3 years, 53.0% women. The mean utility score was 0.91, mean values were significantly lower in obese compared to normal weight in both sexes and social classes. Men with mental health diseases and women with stroke showed the highest disutility values. This methodology enables estimating mean utility values based on individual characteristics of social class, body mass index and chronic diseases and its associated uncertainty. This way it is possible to include more exhaustive second order uncertainty in mean utility values for probabilistic sensitivity analysis in cost-effectiveness modelling.