Hybrid genetic predictive modeling for finding optimal multipurpose multicomponent therapy

摘要

This article proposes a new approach to calculating optimal therapy for the treatment of chronic diseases. For all patients, this approach can calculate drug combinations with the best probability of achieving treatment goals. This approach is based on the combination of machine learning methods, probability graph models, classical statistical modeling tools, and the authors' algorithms. The paper shows the advantages of the proposed approach for the practical medical task of selecting drug combinations for type 2 diabetes mellitus treatment. The authors created useful tools for calculating drug combinations for compensating carbohydrate metabolism for diabetes patients. For treatment goals and values, the research team used the main carbohydrate metabolism indicator - glycated hemoglobin, the lipid profile indicator - low-density lipoprotein cholesterol, and also arterial blood pressure which is an important indicator of a patient's condition. In the virtual implementation, the method showed higher quality in comparison to results obtained without using this approach. For validation, classic metrics and experts' knowledge were used. Both types of validation showed that the method was of high quality and did not contradict fundamental medicine knowledge. Therefore, this method can be used as part of a decision support system for medical specialists who work with type 2 diabetes mellitus patients. This paper is an extended version of the investigation 1. The short version also proposed an approach for calculating optimal therapy. Although this approach is of good quality, it had some drawbacks, so some improvements were made in the new version of the algorithm. The authors used Bayesian networks to reduce the search space by selecting only essential groups of drugs. Moreover, a genetic algorithm was used to find a more accurate solution.