Development of an Institution-Specific Readmission Risk Prediction Model for Real-time Prediction and Patient-Centered Interventions.

Barnes Jewish Hospital (BJH) has developed intervention programs for patients 65 and older diagnosed with acute myocardial infarction (AMI), congestive heart failure (CHF), chronic obstructive pulmonary disease (COPD), and pneumonia (PNA). Additionally, a model was developed using the TDA Mapper algorithm which divides the patients into sub-groups, and then trains independent models for each group.[6] Model parameters are selected via cross-validation, and SMOTE/ROSE sampling was tested to address the low readmission rate. Eligible patients are high-risk inpatients identified using the LACE Index Scoring Tool,[1] which is the most widely used method to quantify the risk of readmission, but only achieves a C-statistic of 0.59 in our population, compared to range 0.63-0.70 in other systems.[2] Yu et al. reported that "the institution specific readmission risk prediction framework is more flexible and more effective than the one-size-fit-all models like the LACE",[3] and several groups have used machine learning models to improve readmissions predictions.[4],[5] Therefore, in this proof of concept study, we use available variables to better identify at-risk patients (compared to the LACE score) at BJH in order to enroll them into the current readmission reduction programs early in their hospital course. [Extracted from the article]