Predicting utility scores for prostate cancer: mapping the Prostate Cancer Index to the Patient-Oriented Prostate Utility Scale (PORPUS)

Abstract

The Prostate Cancer Index (PCI) is a health profile instrument that measures health-related quality of life with six subscales: urinary, sexual, and bowel function and bother. The Patient-Oriented Prostate Utility Scale (PORPUS-U) measures utility (0=dead and 1=full health). Utility is a preference-based approach to measure health-related quality of life, required for decision analyses and cost-effectiveness analyses. We developed a function to estimate PORPUS-U utilities from PCI scores. The development data set included 676 community-dwelling prostate cancer (PC) survivors who completed the PCI and PORPUS-U by mail. We fit three linear regression models: one used original PORPUS-U scores and two used log-transformed PORPUS-U scores, one with a hierarchy constraint and one without. The model selection was performed using stepwise selection and fivefold cross validation. The validation data included 248 PC outpatients with three assessments on the PCI and PORPUS-U. Scores were retransformed for validation, with Duan's smearing estimator applied to correct potential bias. The predictive ability of the models was assessed with R(2), root mean square error (RMSE) and by comparing predicted and observed utilities. The best-fitting model used the log-transformed PORPUS-U with no hierarchy constraint. The R(2) was 0.72. The RMSE ranged from 0.040 to 0.061 for the three validation data sets. Differences between predicted and observed utilities ranged from 0.000 to 0.006 but predicted utilities overestimated the lowest 5% of observed PORPUS-U scores and underestimated the highest observed scores. Our algorithm can calculate PORPUS-U utility scores from PCI scores, thus supplementing descriptive quality of life measures with utility scores in PC patients. Utilities derived from mapping algorithms are useful for assigning utility to groups of patients but are less accurate at predicting utility of individual patients. We are exploring statistical methods to improve the mapping of utilities from descriptive instruments.