Data-Driven Prediction of Complications Risks in Cancer Patients: Machine Learning based approach

Abstract

Cancer chemotherapy involves drugs that interfere with cellular functioning and lead to cell destruction. These cytotoxic drugs have narrow therapeutic index and in most of cases, their potential side effects concern directly and significantly non tumor cells. These adverse effects may be apparent in different forms of symptoms such as headache, nausea, breathing difficulty, tiredness, etc. In real cases, medical staff is facing difficulties to identify patients’ state due to a lack of medical data. In order to limit chemotherapy related side effects and to support the medical staff in the clinical decision process, effective toxicity prediction and assessment structure are crucial. In this paper, we propose to assist treating physicians by predicting the toxicity level of each patient after each chemotherapy session. Thus, they early decide which drug adjustment is required and then prevent any further complication. Our support approach is based on machine learning techniques and relies on predefined toxicity levels for predicting chemotherapy complications. Multi-classification methods are considered and trained on real medical data that were collected during the treatment phase of cancer patients in Tunisia. An assessment of the proposed approach is performed through an experimental study to show the effectiveness and the performance of learning methods.