“Looking Under the Hood” of Anchor-Based Assessment of Clinically Important Change: A Machine Learning Approach

Objectives: The Global Assessment of Change (GAC) item has facilitated the interpretation of change in patient-reported outcomes, providing an anchor for computing minimally important differences. Construct validity has been documented via disease-specific patient-reported outcomes change. We examined what domains, sociodemographic characteristics, attributions of change, and cognitive-appraisal processes are reflected in GAC ratings. Methods: This secondary analysis examined data from 1,481 chronically ill patients and caregivers surveyed at baseline and 17 months. Items queried change since baseline in overall disease symptoms (GAC) and in physical, emotional, and social functioning. Candidate predictors included sociodemographic factors, health-related quality-of-life domains, change attributions, and quality-of-life appraisal processes. Least absolute shrinkage and selection operator and bootstrapping tested 77 predictors’ effectiveness and stability. Results: GAC worsening was notably associated with being disabled (β = −0.24) and having difficulty paying bills (β = −0.13). GAC was better explained by the physical domain than the emotional or social (β = 0.67, 0.10, and 0.03, respectively; R²_adj = 0.63) after sociodemographic-covariate adjustment. In a separate model (R²_adj = 0.18), GAC variance was explained by attributions about changing health and changing response of one's health team, goals related to solving healthcare problems and maintaining activities, and appraisal about things getting better (β = −0.14, 0.08, −0.07, 0.05, 0.21, respectively; p_range ~0.0005–0.05) after adjustment. Conclusions: The GAC primarily reflects the physical domain, and the GAC reflects attributions, goals, and patterns of emphasis related to change in health and healthcare. Commonly unmeasured factors have some bearing on GAC scores and can facilitate the interpretation of change.