A Clinical Decision Support System to Help the Interpretation of Laboratory Results and to Elaborate a Clinical Diagnosis in Blood Coagulation Domain

Hemophilia is a rare hemorrhagic disorder caused by clotting factor deficiencies that leads to a less efficient coagulation system. Treatments of this pathology rely on a patient’s subjective assessment which reflects a need for a laboratory assay able to predict the clinical patient phenotype. According to the literature, global assays such as thrombin generation (TG), are good predictors of bleeding episodes and therefore seem to be good candidates to fit this need. Nevertheless, the result of the TG assay, known as thrombogram, is difficult to interpret for non-expert clinicians. In this paper, we present a machine learning-based clinical decision support system which goal is to help clinical decision making. In doing so, we have adopted several approaches in order to evaluate well-known machine learning algorithms, in terms of accuracy and robustness, on a thrombogram database generated using numerical simulations. Obtained results, 95.57% of accuracy using a cascade of a SVM and MLPs to classify all categories and 98.10% of accuracy for the binary case hemophilia A/B, prove that our proposal can efficiently diagnose hemophilia.