Files
Abstract
Healthcare spending has been increasing in the last few decades. One of the main reasons for this increase is hospital readmissions, which is defined as a rehospitalization of a patient after being discharged from a hospital within a short period of time. The excessive amount of money spent every year on hospital readmissions and the urge to enhance healthcare quality make reducing hospital readmissions a necessity. The approach used in this work is entirely novel and was designed specifically to reduce the number of readmissions by applying the concept of personalization and actionable patterns to guide the health domain experts in their decision-making process. Our goal is not to build a system that replaces physicians, but a system that provides them with new insights discovered from the H-CUP medical dataset.First, we investigate a two-fold problem that predicts the risk of mortality and hospital readmission for newly admitted patients. Several machine learning algorithms are used on our medical dataset to build an accurate classifier. In addition to that, feature selection techniques and boosting were applied to enhance the prediction accuracy and utilize the processing performance.Second, we build the procedure graph, which shows all possible procedure paths that a new patient may undertake during the course of treatment. Additionally, we cluster patients into subgroups that exhibit similar properties in order to improve the predictability of the next procedures. We further devise a metric system that evaluates the level of desirability for procedures along procedure paths, which we would subsequently map to a metric system for the extracted clusters. Finally, we present a novel algorithm that discovers actionable knowledge (medical recommendations) that can be provided to physicians to put patients on a treatment path that would result in optimal reduction of the number of readmissions on average case.Third, we predict the primary medical procedure for a newly admitted patient according to the similarities with the other patients in our medical dataset. This procedure might differ from the primary medical procedure assigned by a physician. We propose three new approaches to identify the patients, from the dataset, that are similar to the newly admitted patient. Finally, we find the procedures that are highly correlated with the primary medical procedure, and provide them as recommendations to physicians to enhance the final status of patients.