Prediction of Attendance to the "Law of 60 Days" in Breast Cancer Patients using Machine Learning Classifiers
Sandra Gioia, Renata Galdino, Lucia Brigagão, Antonio Valadares, Fernando Secol, Sandra San Miguel, Alexandra Bukowski, Lindsay Krush, Paul E. Goss
Abstract
An applied study was conducted on how the use of machine learning techniques can help in the process of identifying compliance with the "Law of 60 Days", which states that all patients with cancer within the public system must initiate the treatment within 60 days after the diagnosis of cancer. Within the Patient Navigation Program (PNP) for breast cancer in Rio de Janeiro, the study aims to construct a model that accurately predicts whether or not a patient meets the period established in the Law. From August 2017 to May 2018, 105 patients aged 33 -80 years (mean 59 years) were recruited for navigation. Patient Navigator (NP) applied questionnaires to collect clinical, psychosocial, and patient satisfaction information. The follow-up was by phone, email, or text message. For the development of the statistical analysis, three learning models were used: AdaBoost, Decision Tree and Gaussian NB. AdaBoost learning model had superior results in relation to accuracy and f-score (0.8889 and 0.8333, respectively) and with good performance in relation to the prediction times. We identified 38 important attributes that contribute 95% of the importance of all the attributes present in the data. We identified 38 important attributes for compliance with the Law, which simplifies the information required for model learning. In the Brazilian context, the PNP may represent an opportunity to adequately implement existing legislation and, as such, would have great potential for integration into federal, state, and local health systems.