Machine Learning Application to Predict the Length of Stay of type 2 Diabetes Patients in the Intensive Care Unit

Abstract

The Intensive Care Units (ICU) are costly units in hospitals catered to a small group of critically ill patients who are monitored round-the-clock, and cost as much as six times that of normal wards. Shortage of beds is a prevalent issue in many countries, including Singapore. This problem is worsened by the aging population which has led to rising healthcare demands and costs. It is a serious problem as there may be delays in care due to the lack of beds. Reducing patients’ length of stay (LOS), especially for the ICU, is one of the key priorities for hospitals in a bid to save cost and manage hospital resources more efficiently. One way to do this is to identify patients that are at risk of having prolonged hospital stay at

the start of their hospitalisation. Accurate identification of such patients allow early planning of treatment and provision of more intensive care to speed up their recovery. As a result, these patients may be discharged earlier, hence reaping cost saving benefits to hospitals and mitigate the bed shortage problem. This project is focused on Type 2 Diabetes Mellitus, which is recognised as a global epidemic due to increasing prevalence and the potentially serious complications resulting from this disease. This project first identifies risk factors for prolonged ICU LOS for patients with Type 2 Diabetes Mellitus, and then makes use of these factors to develop an accurate machine learning model to identify these patients. Finally, incentives to reduce LOS is explored, and hospital cost savings, specifically for Singapore, is calculated from reducing LOS in ICUs.