OBJECTIVES: Chronic obstructive pulmonary disease (COPD) is highly associated with a risk of rehospitalization following a hospitalization. This study proposes a model to predict risk of rehospitalization and identifies important factors contributing to rehospitalizations for patients with COPD.

METHODS: This study used a longitudinal retrospective cohort. The database included health care claims for employees and their dependents from a single, large self-insured employer group from Jan. 2010 through Dec. 2013. COPD cohort was generated by using ICD-9 diagnosis codes defined by the Center for Disease Control and Prevention (CDC). Patients in the analytic sample were required to have a hospitalization claim during the study period. Classification Random Forest (RF) was applied to build a risk prediction model of rehospitalization and identify the most important risk factors. The RF model was fitted with 21 input attributes, built with a training set (75 % of final sample), and validated using the remaining data. The final prediction was averaged from 10,000 tree models.

RESULTS: The COPD cohort included 252 patients ≥18 years of age; 73 patients with hospitalizations were included in the final sample. There were 33 (42.5 %) females; 43 (58.9 %) were over 50 years of age. Comorbidity status, post-discharge COPD-related prescription, pre-discharge care management, initial hospital stays, gender, and emergency visits were important risk factors for rehospitalization. Probability of rehospitalization for the final sample were CONCLUSIONS: This study applied an advanced approach to predict rehospitalization and identify important risk factors for patients with COPD. These findings suggest a focus on patient assessment both pre- and post-discharge to reduce potentially preventable rehospitalizations.