Computing PROPr Utility Scores for PROMIS® Profile Instruments.

Area of Science:

• Health outcomes research
• Psychometrics
• Health economics

Background:

• Patient-Reported Outcomes Measurement Information System® (PROMIS) Profile instruments assess health status across 8 domains.
• The PROMIS-Preference (PROPr) score summarizes health states using 7 PROMIS domains, incorporating a preference-based utility value.
• PROPr and Profile instruments share 6 domains, with PROPr uniquely including Cognitive Function-Abilities and Profile including Anxiety and Pain Intensity.

Purpose of the Study:

• To develop an equation for calculating PROPr utility scores using data from PROMIS Profile instruments.
• To compare different methods for estimating the missing Cognitive Function-Abilities score within the PROMIS framework.

Main Methods:

• Utilized data from 3982 US online survey panel members with scores across 9 PROMIS domains.
• Employed a 70%/30% data split for model training and validation.
• Compared five models for predicting the Cognitive Function score: population average, excellent/poor function scores, linear regression, and Bayesian neural network, evaluating using root-mean-square error and mean error.

Main Results:

• Linear regression and Bayesian neural network models demonstrated the lowest mean errors (0.018) in predicting PROPr utility scores.
• Root-mean-square errors for linear regression (0.095) and Bayesian neural network (0.094) were comparable and superior to other methods.
• The Bayesian neural network model yielded the best root-mean-square error, but linear regression performed nearly as well.

Conclusions:

• Linear regression provides a nearly optimal and more user-friendly approach for calculating PROPr utility scores from PROMIS Profile data compared to more complex models.
• The developed equation and recommended linear model facilitate the estimation of PROPr utility scores, enhancing the utility of PROMIS Profile data in health research and clinical applications.