OBJECTIVES: Studies have shown the disutility of having multiple health problems is usually smaller than the disutility sum of individual component health problems. However, existing model specifications for predicting utility values of the multi-dimensional EQ-5D health states do not reflect this phenomenon. This study aimed to evaluate the two-way interaction effects between health dimensions on modelling of EQ-5D values.

METHODS: We tested 10 two-way dimensional interactions with 16 EQ-5D-5L valuation datasets, each derived from a different country/district using the same study protocol. We generated interaction terms by treating dimensional problem levels of EQ-5D-5L health states as continuous variables, and added them into a 20-parameter (incremental) main-effects model for predicting composite time trade-off (cTTO) values. Model performance was assessed in terms of: i) parameter statistical significance; ii) predicted health-state values’ logical consistency; and iii) out-of-sample prediction accuracy using mean square error (MSQE), and Pearson’s R (R), also compared with that of the main-effects model.

RESULTS: In 15 countries, there was minimally one significant interaction term, with the median (interquartile) of significant interaction terms being 3 (1 to 4). In 9 countries, including interactions increased the number of significant main-effect terms. Most significant interactions occurred between the pain/discomfort and anxiety/depression dimensions (n= 9 countries) and interactions mainly exhibited diminished joint disutility. Incorporating interaction terms resulted in lower MSQE (mean reduction: 24.6%; SE: 5.74) with higher R (mean increase: 0.6%; SE: 0.20) in 13 of 15 countries. Additionally, logically inconsistent predictions for health-state pairs with dominance relationship was minimal, ranging from 0% to 0.48%.

CONCLUSIONS: Significant dimensional interaction effects existed in EQ-5D-5L health-state valuation. The interactions existed, behaved as expected, and improved data modelling in most countries studied. Future valuation studies for multi-dimensional health descriptive systems such as EQ-5D should consider exploring interaction effects between health dimensions.