Predicting SF-6D Preference-based Score based on the European Organization for Treatment and Research of Cancer Quality of Life Questionnaire in colorectal cancer

Abstract

AIMS: To develop a mapping model for estimating SF-6D preference-based scores from the European Organization for Research and Treatment of Cancer Quality of Life Questionnaire Core module (QLQ-C30) and Colorectal-specific module (QLQ-CR38/CR29) in patients with colorectal Cancers (CRC), with and without adjustment for clinical and demographic characteristics.; METHODS: Ordinary least squares regression model was applied to the model development using a cross-sectional data of 216 patients with CRC collected from outpatient specialist clinic of a regional hospital in Hong Kong. The items responses or scale scores of QLQ-C30 in conjunction with QLQ-CR38/CR29 data, selected demographic (sex and age) and clinical (cancer stage, treatment completion and presence of stoma) characteristics of patients were used to predict the SF-6D scores. The goodness-of-fit of models were examined by exploratory power (R-square and adjusted R-square), Akaike information (AIC) and Bayesian information criterion (BIC) whereas the predictive performances were evaluated by root mean square error (RMSE), mean absolute error (MAE) and Spearman’s correlation coefficients between predicted and observed SF-6D scores. The models were validated against an independent set of 56 patients with CRC to assess the predictive performance.; RESULTS: Both scale and item response models obtained the variation more than 67 % explained in the SF-6D scores. The model using the main and interaction effects of the QLQC30 item scores in supplemented with QLQ-CR29 item responses plus clinical variable of stoma was chosen as the best-performing mapping model because of having the best fit with the highest R-square of 75.9 %, and good predictive performances with Spearman’s correlation coefficient of 0.871 and low predictive errors (RMSE: 0.079; MAE: 0.062). There was over-prediction for poorer health states at the observed SF-6D scores lying below 0.7 but under-prediction was observed for better health states when SF-6D scores were valued at 0.7 or above.; CONCLUSIONS: SF-6D preference-based scores can be predicted using QLQ-C30 and QLQ-CR38/CR29 scores with satisfactory precision in patients with CRC for appraising clinical interventions in economic evaluations.