Abstract
ICU readmission is associated with longer hospitalization, mortality andadverse outcomes. An early recognition of ICU re-admission can help preventpatients from worse situation and lower treatment cost. As the abundance ofElectronics Health Records (EHR), it is popular to design clinical decisiontools with machine learning technique manipulating on healthcare large scaledata. We designed data-driven predictive models to estimate the risk of ICUreadmission. The discharge summary of each hospital admission was carefullyrepresented by natural language processing techniques. Unified Medical LanguageSystem (UMLS) was further used to standardize inconsistency of dischargesummaries. 5 machine learning classifiers were adopted to construct predictivemodels. The best configuration yielded a competitive AUC of 0.748. Our worksuggests that natural language processing of discharge summaries is capable tosend clinicians warning of unplanned 30-day readmission upon discharge.