Interpretable Topic Features for Post-ICU Mortality Prediction.

摘要: Electronic health records provide valuable resources for understanding the correlation between various diseases and mortality. The analysis of post-discharge mortality is critical healthcare professionals to follow up potential causes death after a patient discharged from hospital give prompt treatment. Moreover, it may reduce cost derived readmissions improve quality healthcare. Our work focused on ICU prediction. In addition features physiological measurements, we incorporated ICD-9-CM hierarchy into Bayesian topic model learning extracted medical notes. We achieved highest AUCs 0.835 0.829 30-day 6-month prediction using baseline proportions Labeled-LDA. our emphasized interpretability which facilitates investigation complexity diseases.