PHP72 - Predicting A Patient's Future Medical Visits: A Comparison Of Quality Of Life And Clinical Variables

Abstract

Many clinical measures and patient-reported quality-of-life scores correlate with health care resource use, but their relative predictive value is understudied. This analysis compares a well-established physical-health-related quality of life measure against comorbidity status and current medical visits in their ability to predict the number of doctor visits three months later. Data are drawn from a 2009 study of a general US representative sample in which a subset were reinterviewed after approximately 3 months (n=470). Study measures included a comorbidity checklist, a self-reported count of medical visits in the preceding four weeks, and the SF-36 or SF-12 physical component summary (PCS) measure. We constructed three logistic regression models estimating the probability of higher resource use (self-report of 2+ medical visits in the past four weeks vs. 1 or none) after three months. Model 1 included the number of comorbidities (none; 1; 2–3; 4+) and resource use (count of medical visits) at baseline as independent variables. Model 2 included just PCS at baseline, and model 3 included all predictors from models 1 and 2. Models 1 (AUC=0.77, pseudo-R2=0.16) and 2 (AUC=0.76, pseudo-R2=0.14) have roughly the same explanatory power, suggesting that PCS has predictive ability similar to counts of comorbidities and medical visits. For example, a 5-point lower PCS at baseline is associated with a 50% increase in the odds of having 2+ medical visits three months later (Model 2), an association similar to that of one more medical visit at baseline or being in a higher comorbidity count category (Model 1). Model 3 (AUC=0.80, pseudo-R2=0.19) improves somewhat on both models. Our results suggest that as compared to comorbidity counts and number of medical visits, PCS alone has similar ability to predict having 2+ doctor visits during a four week period three months later.