OBJECTIVES: This study aimed to calculate the mapping algorithm from modified Rankin Scale (mRS) to EQ-5D-3L for Chinese ischemic stroke (IS) patients.

METHODS: Data were collected from a longitudinal multicenter post-market trial, including sociodemographic information, baseline clinical characteristics, mRS levels and EQ-5D-3L records. HRQoL data of Chinese IS patients were recorded from 4 series visits. Correlation analysis and influential factors analysis were applied to test the feasibility to map the mRS to EQ-5D-3L. Least ordinal square (OLS) model, Tobit model, ordered Logistic model, multinomial Logistic model and mixed effect model were used to build the mapping algorithm. 10-fold cross-validation and several indicators such as Akaike information criterion, Bayesian information criterion, mean absolute error and root mean squared error were conducted to assess the model prediction.

RESULTS: A total of 9788 patients were included in the study. The feasibility of mapping EQ-5D-3L from mRS was proved according to the high correlation among the 2 scales. Tobit model was preferred when the data collected at admission visit were used for the mapping construction while OLS model was superior when both the data collected at discharge visit and pooled 4 visits data were used for the mapping construction. When taking into account the internal correlation among longitudinal data thus applying mixed liner model and mixed multinomial Logistic model to build the mapping algorithms, both models exhibited good property of prediction. Based on the internal and external estimation results, the mixed multinomial Logistic model was recommended as the best model for the relatively small errors and high consistency of estimated and observed utility density curves.

CONCLUSIONS: Mixed effect models have superior prediction properties for longitudinal data. Due to the low prediction errors and the ability to predict patient-level utilities, mixed multinomial Logistic model is recommended as the best model for the mapping algorithm for mRS and EQ-5D-3L.