Probabilistic graphical models for finding optimal multipurpose multicomponent therapy
Vladislav V. Pavlovskii, Ilia Vladislavovich Derevitskii, Daria A. Savitskaya · Procedia Computer Science · 2021
This paper suggests a probabilistic graphical models’ approach for finding optimal therapy. This approach is based on creating a network of dependencies using statistics of patient treatment. We used Bayesian networks for describing diabetes mellitus treatment. 4 networks were created, one of them with expert knowledge, and the other was created using different algorithms. Treatment outcomes include a set of treatment-goal values and a combination of drugs. Networks were trained and validated by the treatment dataset. Results of validation showed that this approach was high-quality for cases that had a wide representation of using medication. Most of the predictions were equal with the expert’s opinion, therefore models could be used as part of Decision Support Systems for medical experts who work with patients suffering from T2DM (Type 2 Diabetes Mellitus).