Estimating Joint Health Condition Utility Values

Objectives
To predict health state utility values (HSUVs) for individuals with up to four conditions simultaneously.
Methods
Person-level data from the General Practice Patient Survey, a national survey of adult patients registered with general practices in England. Individuals report whether they have one of 16 chronic conditions and complete the EQ-5D three-level. Four non-parametric methods (additive, multiplicative, minimum and the adjusted decrement estimator) and one parametric estimator (the Linear Index) were used to predict HSUVs for individuals with a joint health condition (JHC). Predicted and actual utility scores were compared for precision using root mean squared error and mean absolute error. Bias was assessed using mean error.
Results
The analysis included 929,565 individuals, of which 30.5% had at least two conditions. Of the non-parametric estimators, the multiplicative approach produced estimates with the lowest bias and most precision for two condition JHCs. For populations with a long-term mental health condition within the JHC, the multiplicative approach overestimated utility scores. All non-parametric methods produced biased results when estimating HSUVs for three or four condition JHCs. The Linear Index generally produced unbiased results with the highest precision.
Conclusions
The multiplicative approach was the best non-parametric estimator when estimating HSUVs for two JHCs. None of the non-parametric approaches to estimate HSUVs can be recommended with more than two JHCs. The Linear Index was found to have good predictive properties but needs external validation before being recommended for routine use.