Feature analysis and prediction of complications in ostomy patients based on laboratory analytical data using a machine learning approach

BACKGROUND: Stoma complications are commonly related to inadequate surgical preparation and changes in abdominal perimeter configuration. Preventive identification of such complications would improve patient care quality of life.OBJECTIVE: To carry out an approximate study on the prediction of complications in ostomy patients, based on laboratory analytical data using a machine learning approach.METHODS: A total of 20 patients participate in the study. Following the data preprocessing stage, 16 patients with 14 features were selected. The cross-validation method with k = 4 was used to obtain the results of 3 classifiers trained using all the features. Moreover, the classification process was also applied to the most important features, selected according to a committee composed of feature importance algorithms.RESULTS: The results showed that the best outcomes were obtained using the logistic regression classifier, with the features selected by the attribute importance committee.CONCLUSIONS: It is not possible to use the system developed as a diagnostic tool at this point, due to the small amount of data. However, the results were good in terms of classification. Furthermore, the approach of using a committee to select the most important clinical features improved the results. The results obtained in this project can serve as an initial starting point for future research on diagnostic systems for ostomy patients.