Validation of case-mix measures derived from self-reports of diagnoses and health

Abstract 

Self-reported chronic diseases and health status are associated with resource use. However, few data exist regarding their ability to predict mortality or hospitalizations. We sought to determine whether self-reported chronic medical conditions and the SF-36 could be used individually or in combination to assess co-morbidity in the outpatient setting. The study was designed as a prospective cohort study. Patients were enrolled in the primary care clinics at seven Veterans Affairs (VA) medical centers participating in the Ambulatory Care Quality Improvement Project (ACQUIP). 10,947 patients, ⩾ 50 years of age, enrolled in general internal medicine clinics who returned both a baseline health inventory checklist and the baseline SF-36 who were followed for a mean of 722.5 (±84.3) days. The primary outcome was all-cause mortality, with a secondary outcome of hospitalization within the VA system. Using a Cox proportional hazards model in a development set of 5,469 patients, a co-morbidity index [Seattle Index of Co-morbidity (SIC)] was constructed using information about age, smoking status and seven of 25 self-reported medical conditions that were associated with increased mortality. In the validation set of 5,478 patients, the SIC was predictive of both mortality and hospitalizations within the VA system. A separate model was constructed in which only age and the PCS and MCS scores of the SF-36 were entered to predict mortality. The SF-36 component scores and the SIC had comparable discriminatory ability (AUC for discrimination of death within 2 y 0.71 for both models). When combined, the SIC and SF-36 together had improved discrimination for mortality (AUC = 0.74, p-value for difference in AUC < 0.005). A new outpatient co-morbidity score developed using self-identified chronic medical conditions on a baseline health inventory checklist was predictive of 2-y mortality and hospitalization within the VA system in general internal medicine patients.

Keywords:  Co-morbidity, Questionnaires, Quality of life, Confounding factors, ROC curve, Proportional hazards model