Objective: This study aimed to test multiplicative modelling with EQ-5D-3L time-trade-off (TTO) and visual analogue scale (VAS) values.
Methods: A multi-stage sampling design was adopted for the study and data collection took place in three phases in 2010, 2011, and 2012 in the Northern region of Malaysia. Face-to-face interviews involved respondents answering both 13 TTO and 15 VAS valuation tasks were carried out. Both additive and multiplicative model specifications were explored using the valuation data. Model performance was evaluated using out-of-sample predictive accuracy by applying the cross-validation technique. The distribution of the model values was also graphically compared on Bland-Altman plots and kernel density distribution curves.
Results: Data from 630 and 611 respondents were included for analyses using TTO and VAS models, respectively. In terms of main-effects specifications, cross-validation results revealed a slight superiority of multiplicative models over its additive counterpart in modelling TTO values. However, both main-effects models had roughly equal predictive accuracy for VAS models. The non-linear multiplicative model with I32 term, MULT7_TTO, performed best for TTO models; while, the linear additive model with N3 term, ADD11_VAS, outperformed the other VAS models. Multiplicative modelling neither altered the dimensional rankings of importance nor did it change the distribution of values of the health states.
Conclusion: Using EQ-5D-3L valuation data, multiplicative modelling was shown to improve out-of-sample predictive accuracy of TTO models but not of VAS models.
Keywords: EQ-5D-3L; Modelling; Multiplicative; Time-trade-off; Value set; Visual analogue scale.