ReDiaSafe: A Novel Approach for Predicting 30-Day Diabetes Patient Readmission Risk

Diabetes is an ailment where life becomes dangerous because most readmissions occur within 30 days since patients do not have any means of awareness towards controlling this health issue. Hospital readmission prevention, especially regarding inpatient or outpatient care, has always been very important in improving patient care, satisfaction, and cost benefits in health care. This paper is on ReDiaSafe as a machine learning tool for predicting the probability of 30-day hospital readmissions in a diabetic patient. It intended to identify useful data attributes and build predictive models that are able to classify readmission risks with accuracy. Thorough research on user input data had been conducted and feature analysis done to elicit significant parameters that influence readmission likelihood. Several machine learning technologies, notably LightGBM and logistic regression; have been applied to classify patients into risk categories: low, medium, and high risk. Our results showed several features with strong relationships to readmission risk, including demographic variables, past admissions, lab results, and medication use patterns. Out of all models tested, the one found most promising with regard to predictions was LightGBM, as this learned well to differentiate risk categories. Furthermore, the combining with an interactive AI chatbot provides individual healthcare suggestions and recommendations based on personalized risk assessments to better engage and support the users. This is further enhanced by the ReDiaSafe tool, which goes a long way in harnessing state-of-the-art machine learning techniques for appalling and really meaningful hospital readmission risk estimates for diabetic patients. This way, better quality of life for the patients is achieved with this method towards life preservation. However, this will also potentially offer general healthcare practitioners segmented interventions towards a much more reduced readmission rate.