Comparison of physician referral and insurance claims data-based risk prediction as approaches to identify patients for care management in primary care: an observational study

BACKGROUND: Primary care-based care management (CM) could reduce hospital admissions in high-risk patients. Identification of patients most likely to benefit is needed as resources for CM are limited. This study aimed to compare hospitalization and mortality rates of patients identified for CM either by treating primary care physicians (PCPs) or predictive modelling software for hospitalization risk (PM). METHODS: In 2009, a cohort of 6,026 beneficiaries of a German statutory health insurance served as a sample for patient identification for CM by PCPs or commercial PM (CSSG 0.8, Verisk Health). The resulting samples were compared regarding hospitalization and mortality rates in 2010 and in the two year period before patient selection. No CM-intervention was delivered until the end of 2010 and PCPs were blinded for the assessment of hospitalization rates. RESULTS: In 2010, hospitalization rates of PM-identified patients were 80% higher compared to PCP-identified patients. Mortality rates were also 8% higher in PM-identified patients if compared to PCP-identified patients (10% vs. 2%). The hospitalization rate of patients independently identified by both PM and PCPs was numerically between PM- and PCP-identified patients. Time trend between 2007 and 2010 showed decreasing hospitalization rates in PM-identified patients (−15% per year) compared to increasing rates in PCP-identified patients (+34% per year). CONCLUSIONS: PM identified patients with higher hospitalization and mortality rates compared to PCP-referred patients. But the latter showed increasing hospitalization rates over time thereby suggesting that PCPs may be able to predict future deterioration in patients with relatively good current health status. These patients may most likely benefit from preventive services like CM.