engleski

Prediction of unfractioned heparin effect using deep learning approach

Background and Goal of Study: Unfractioned heparin (UFH) is the anticoagulant of choice during carotid endarterectomy surgery (CEA). Activated partial thromboplastin time (aPTT) has been the primary laboratory test used to monitor effect of UFH. Range of 1.5-2.5 times baseline value has gained wide acceptance as therapeutic range in daily clinical practice. Despite being cornerstone of anticoagulation, UHF is limited by its unpredictable pharmacokinetic profile and is difficult to dose accurately. We hypothesized that deep learning, a type of machine learning based on a set of algorithms to model high level abstractions in data using multiple linear and nonlinear transformations, could better interpret the dose-response relationship of UFH. Materials and Methods: We studied 63 consecutive patients undergoing elective carotid endarterectomy in superficial cervical block. Heparin sodium 50 U/kg of total body weight was administered 3 minutes before application of carotid artery cross-clamp and aPTT was measured 30 minutes after heparin administration. We built a model for prediction of aPTT of 1.5 to 2.5 times the baseline value using artificial neural network (ANN). ANN was composed of three layers: input layer, the middle layer and output layer. Input variables included the pharmacokinetic-pharmacodynamic covariates (age, sex, weight, height, previous antiplatelet therapy, complete blood count, biochemistry panel, baseline coagulogram, total heparin dose). Twenty percent of the dataset was used as a testing dataset, and the remaining were used for model training. Results and Discussion: The activation function used for both ANN layers was sigmoid. aPTT of 1.5 to 2.5 times the baseline value at 30 minutes occurred in 12 patients (19.0%). The area under the receiving operator curve (AUROC) was 0.752 with deep learning algorithm prediction accuracy of 90.4%. None of the patients were bellow therapeutic range of aPTT. Conclusion(s): Deep learning approach can be used for calculating better dosage of used drugs to deliver safer, more efficient and more cost-effective doses. Adding more clinical data into this model and testing various models might further improve prediction of heparin effect.

anticoagulants ; Pharmacology ; dose-response ; Model ; statistical

nije evidentirano