Evaluation of artificial neural network techniques in improving prescription of anticoagulant therapy

To evaluate whether dose-response schemes of anticoagulant therapy can be improved by using artificial neural network (NN) techniques, a NN was built for each of 50 randomly selected patients based on their clinical history. Five records, each representing five subsequent sets of prescribed dosage (D) and resulting measured coagulant level (TT), were used for training and the subsequent record for testing in each patient. An additional NN was trained using 1200 records derived from 200 randomly selected patients. By comparing the difference between physician intended and in practice achieved TT and the difference between TT achieved and that predicted by the NN, no evidence was found for the hypothesis that an artificial neural network could improve the prescription of anticoagulant therapy. Thus, prescription of anticoagulant therapy by the human expert is optimal