Online Program

Predicting risk of hospitalization among primary care patients (303201)

Keywords: Prediction model; Hospitalization; Primary care; Cox regression; Time-varying effect

A time-to-event prediction model was applied to a retrospective cohort of primary care patients to predict hospitalization among them. The origin of time was the date of first office visit (i.e., index visit) in 2013, and the end of the study was June 30, 2014. Potential predictors included demographics, payer, comorbidities, procedures, medications, health care utilization, and temporal variation of health-related data. Approximately 6.6% of the study population experienced a hospitalization after index visit. Predictors in the final model included older age, being unmarried, previous utilization of inpatient or outpatient service, worsened health or increased utilization 2-year prior to admission, days since last ER, advanced imaging, or major procedure, and several specific comorbidities. The c-statistics were above 0.87 with sensitivities =0.74, specificities =0.84, negative predictive values =0.98 and positive predictive values ranging from 0.055 to 0.309 over the first year of follow-up. The model demonstrated good predictive accuracy. Implementing the model or similar models in primary care could help guide PCPs in managing patients and preventing hospitalizations.