A weighted predictive modeling method for estimating thresholds of meaningful within-individual change for patient-reported outcomes.

Abstract

Purpose: Calculating the threshold for meaningful within-individual change (MWIC) is essential for interpreting patient-reported outcomes (PRO). However, traditional methods of determining MWIC threshold yield varying estimates and lack a standardized approach. We aim to propose a novel method for more accurate MWIC threshold estimation.

Methods: We developed a weighted predictive modeling method. The weighting involved using the rank difference between PRO score change and the anchor of each individual. A Monte Carlo simulation was conducted to compare the performance of the new method and that of existing state-of-the-art methods. Simulation parameters included distributions of PRO score changes, sample sizes, improvement proportions, and correlation strengths. Statistical performance was assessed using relative bias (rbias), coefficient of variation (CV), and relative root mean squared error (rRMSE).

Results: Distribution-based methods had the largest rbias and rRMSE among all methods. Existing anchor-based methods except for the Terluin 2022 method were biased when the correlation strength was weak or when the improvement proportion was not 50%. The Terluin 2022 method requires estimating an important reliability parameter, and this method had highest CV compared to other predictive modeling methods. The new weighted method demonstrated the smallest rRMSE across most simulation settings. It also maintained relatively high accuracy under weak correlation strength or imbalanced improvement proportion. Similar results were presented under normal or skewed distributions of PRO score changes.

Conclusion: This novel method offers a simple and feasible alternative to existing predictive modeling methods for estimating MWIC threshold, which can facilitate the application of PRO.