A critical re-evaluation of the regression model specification in the US D1 EQ-5D value function

Area of Science:

• Health Economics
• Psychometrics
• Regression Analysis

Background:

• The EQ-5D is a widely used health-related quality of life instrument for cost-effectiveness analyses.
• EQ-5D health states are valued using regression models, typically including a constant term representing utility loss from perfect health.
• A recent study replaced this constant with a D1 variable, signifying impaired dimensions beyond the first.

Purpose of the Study:

• To illustrate the problems associated with using the D1 variable instead of a constant term in EQ-5D regression models.
• To compare the implications of the D1 variable versus a constant term on model coefficients, multicollinearity, and interpretation.

Main Methods:

• Comparison of the original D1 regression model with a mathematically equivalent model featuring a constant term.
• Analysis of coefficient magnitudes and statistical significance.
• Assessment of multicollinearity using variance inflation factors (VIFs).
• Evaluation of calculation steps for tariff values and interpretation consequences.

Main Results:

• The D1 variable shifted coefficients and significantly increased multicollinearity (max VIF 113.2 vs. 21.2).
• The number of calculation steps required to determine health state values increased.
• The D1 model obscured statistically nonsignificant main effect dummy variables.

Conclusions:

• Using the D1 variable unnecessarily complicates EQ-5D valuation models.
• This approach hinders clear interpretation and can bias the resulting tariff algorithm.
• A constant term provides a more straightforward and interpretable approach to EQ-5D health state valuation.