Predicting the length of hospital stay after surgery for perforated peptic ulcer

Abstract

The management of peptic ulcer disease usually implies an urgent surgical procedure with the need of a patient’s hospital admission. By predicting the length of hospital stay of patients, improvements can be made regarding the quality of services provided to patients. This paper focuses on using real data to identify patterns in patients’ profiles and surgical events, in order to predict if patients will need hospital care for a shorter or longer period of time. This goal is pursued using a Data Mining process which follows the CRISP-DM methodology. In particular, classification models are built by combining different scenarios, algorithms and sampling methods. The data mining model which performed best achieved an accuracy of 87.30%, a specificity of 89.40%, and a sensitivity of 81.30%, using JRip, a rule-based algorithm and Cross Validation as a sampling method.