Sistema de suporte à decisão para a escolha do protocolo terapêutico para pacientes com leucemia mieloide aguda

Abstract

Acute Myeloid Leukemia is a chronic and disabling disease. It is heterogeneous and presents in different ways. After the diagnosis, the patient receives a risk prognosis of outcomes usually divided into three groups: favorable, intermediate, and adverse. Specialists frequently use this classification to customize therapeutic decisions. In recent decades, the standard treatment has been intensive therapy with a combination of cytarabine and anthracycline. Current risk classification is conservative and often requires specialists to resort to more information, such as the results of other exams and analyses, to select the appropriate treatment, even with little or no evidence of efficacy. This process can delay the start of treatment and worsen the patient's clinical status. In this study, we have investigated the behavior of the disease and its implications. Specifically, we carried out a systematic mapping of the literature to categorize and analyze the existing treatment guides and, mainly, to find out if any employ Machine Learning techniques. The therapeutic protocols were grouped according to the intensity of each current treatment. The disease manifests in different ways in patients; therefore, deciding on a generic therapeutic course is challenging. Thus, therapeutic strategies have become increasingly personalized and isolated for individualized clinical realities. Several variables can influence the choice of treatment guidelines, such as the patient's age, relapses, and drugs that inhibit protein actions. However, combining these and other criteria is difficult in a manual analysis, especially when genetic data are used. Therefore, it is important to resort to tools that can perform these analyzes automatically to assist specialists in choosing the best treatment protocols for their patients. In this context, this study proposes a decision support system that aims to automatically recommend sets of appropriate treatments for patients with acute myeloid leukemia by automatically predicting their mortality/survival. We used clinical and genetic data (mutation and gene expression) from two public-domain databases. The proposed system is formed by two classification committees composed of the best prediction models obtained. The results indicate that the proposed system is promising and can be used as a decision support tool for specialists, with the potential to reduce subjectivity and time in the processes of choosing treatments, resulting in more strong recommendations with fewer adverse effects, which may contribute to increasing the survival and quality of life of patients.