Background: Statistical models that identify patients at elevated risk of death or hospitalization have focused on population subsets, such as those with a specific clinical condition or hospitalized patients. Most models have limitations for clinical use. Our objective was to develop models that identified high-risk primary care patients.
Methods: Using the Primary Care Management Module in the Veterans Health Administration (VHA)’s Corporate Data Warehouse, we identified all patients who were enrolled and assigned to a VHA primary care provider on October 1, 2010. The outcome variable was the occurrence of hospitalization or death during the subsequent 90 days and 1 year. We extracted predictors from 6 categories: sociodemographics, medical conditions, vital signs, prior year use of health services, medications, and laboratory tests and then constructed multinomial logistic regression models to predict outcomes for over 4.6 million patients.
Results: In the predicted 95th risk percentiles, observed 90-day event rates were 19.6%, 6.2%, and 22.6%, respectively, for hospitalization, death, and either hospitalization or death, compared with population averages of 2.7%, 0.7%, and 3.4%, respectively; 1-year event rates were 42.3%, 19.4%, and 51.3%, respectively, compared with population averages of 8.2%, 2.6%, and 10.8%, respectively. The C-statistics for 90-day outcomes were 0.83, 0.86, and 0.81, respectively, for hospitalization, death, and either hospitalization or death and were 0.81, 0.85, and 0.79, respectively, for 1-year outcomes.
Conclusions: Prediction models using electronic clinical data accurately identified patients with elevated risk for hospitalization or death. This information can enhance the coordination of care for patients with complex clinical conditions.
*VA Puget Sound Health Care System
Departments of †Medicine
‡Health Services, University of Washington, Seattle, WA
§VHA Office of Informatics and Analytics, Analytics and Business Intelligence, Washington, DC
Supplemental Digital Content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal's Website, www.lww-medicalcare.com.
The views expressed in this manuscript are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs. This work was conducted entirely for operations; findings reported were not derived, in whole or in part, from activities constituting research.
The authors declare no conflict of interest.
Reprints: Charles Maynard, PhD, VA Puget Sound Health Care System, HSR&D 1100 Olive Way Suite #1400, Seattle, WA 98101. E-mail: firstname.lastname@example.org.