Predicting Hospital Readmissions by CatBoost to Improve Monitoring the Post-Operated Cardiac Patients

Pietro Fernandes Magaldi1, Thaynara Matos1, Júlia Ferreira2, Jasmine Battestin Nunes3, Pietro Colonna Carlotto de Oliveira Martins4, Camila Rodrigues Moreno5, Guilherme de Castro Machado Rabello5, Ahmad A Almazloum6, Emely P da Silva7, Anderson Rocha7
1State University of Campinas, 2Institute of Computing - University of Campinas (UNICAMP), 3Universidade Estadual de Campinas, 4Instituto do Coração (Heart Institute), Hospital das Clinicas HCFMUSP, Faculdade de Medicina, Universidade de São Paulo, São Paulo, Brazil Departamento: InovaInCor, 5Instituto do Coracao, Hospital das Clinicas HCFMUSP, Faculdade de Medicina, Universidade de Sao Paulo, 6Recod.ai Lab, Institute of Computing, University of Campinas (UNICAMP), 7Unicamp


Abstract

Following recent advances in wearable devices and AI classifier models, a system using the CatBoost classifier model to analyze data provided by Smartwatches and cellular devices through remote monitoring system was proposed, in order to improve the accuracy of making the decision in such systems. The input data for each participant were consisted of the patient's medical history along with the patient's vital signals, and statistical features extracted from the signal time series. Vital signals were collected mainly using smartwatches. The model performed binary classification (N=49) across a dataset split into 3 folds, using cross-validation. The Optuna algorithm was used to optimize the model. It scored (91.88 ± 7.40)% balanced accuracy, (83.81 ± 3.30)% F1-score and with (95.18 ± 6.27)% ROC-AUC. Overall, the system showed promising results towards classifying high/low risk patients, given the low number of samples and high evaluation scores. Possible improvements in the project include a higher number of samples and model calibration to enhance the reliability of risk scores.