Abstract
Background/Aims Preventable hospital readmissions are the focus of many performance metrics. However, little attention has been given to preventing index admissions. The likelihood of hospitalization (LOH) is a statistical model (Verisk Analytics®) which uses patient historical clinical data and logistic regression to predict the likelihood of admission in the next six months. One of our goals was to test the accuracy of the LOH model to predict hospitalizations. The second goal was to identify possible common intervenable causes for the hospitalization.
Methods We obtained the list of patients with LOH scores in the top 1% calculated as of July 1, 2012. We then retrospectively identified which of these patients had hospitalizations in the following 6 months. We censored patients who had hospitalizations related to trauma and pregnancy. We performed chart reviews on a subset of the hospitalized patients and classified reasons for the admission into intervenable or non-intervenable causes.
Results Of the patients in the top 1% of LOH (N = 1460; mean LOH = 0.47), 412 (28.2%) were hospitalized within 6 months of the LOH calculation date. The average age and percent female for hospitalized and non-hospitalized patients was 74.1 years, 57.8% and 73.4 years, 60.0%, respectively (not significant). 324 of the hospitalized patients had no emergency department visits in the prior 6 months. Fifty percent of hospitalizations occurred within 33 days of the LOH calculation date. We performed chart reviews on 134 (32.5%) of the hospitalized patients. Twenty-eight patients (20.9%) were classified with intervenable causes for the admission, 19 (67.9%) of which were related to system issues such as inadequate follow-up after procedures or medication adjustments.
Conclusions Identifying patients with intervenable causes for hospitalizations may significantly decrease unnecessary admissions. Risking algorithms may have utility in developing strategies to identify these patients. Risk of Hospitalization (RHO2) uses near real-time clinical data from the electronic medical to calculate risk scores. Future studies will compare the performance characteristics of LOH to RHO2 to identify overlapping cohorts of patients. If RHO2 performs as well as LOH, we will use the near-real time clinical data in RHO2 to flag patients with intervenable causes of hospitalization at the point-of-care.
