Predicting Inpatient Status for the Next 30/60/90 Days With Machine Learning

Lakshmi Prayaga, Krishna Devulapalli, Chandra Prayaga, Joe Carloni

OAI: oai:igi-global.com:284961 • DOI: 10.4018/IJBDAH.20210701.oa9

In this paper, we report the development of machine learning techniques which can help hospital authorities assess a patients' medical condition and also calculate the probability of readmission of the patient as inpatient, and thus identify patients with higher risks for readmissions. Factor Analysis is performed on patient data to understand the severity of mental health, and Random Forest models are used to determine the probability of a patient becoming an inpatient for the next 30/60/90 days from their last visit to the physician’s office. The Random Forest model fits the data with an overall OOB Error rate of 3.69% and an accuracy of 97.65%. The accuracy on the test data was 96.11%. A web application is also developed to provide a user-friendly interface for physicians and administrators to interact with and obtain relevant information for a given patient and or a group of patients. The web application affords physicians additional inputs to assist in their diagnosis and administrators, a window into anticipating and preparing for future patient needs.