AUTHOR=Wang Mingyuan , Qian Yiyi , Yang Yaodong , Chen Haobin , Rao Wei-Feng 

TITLE=Improved stacking ensemble learning based on feature selection to accurately predict warfarin dose

JOURNAL=Frontiers in Cardiovascular Medicine

VOLUME=Volume 10 - 2023

YEAR=2024

URL=https://www.frontiersin.org/journals/cardiovascular-medicine/articles/10.3389/fcvm.2023.1320938

DOI=10.3389/fcvm.2023.1320938

ISSN=2297-055X

ABSTRACT=With the rapid advancement of artificial intelligence, the prediction of warfarin dosage using machine learning has garnered increasing attention. However, due to the presence of both linear and nonlinear factors in dosage prediction, traditional machine learning algorithms struggle to effectively address such complexities. Improved stacking ensemble learning can solve such problem and achieve higher prediction accuracy. We collected data from 641 patients in southern China who had achieved a stable state on warfarin. The dataset encompassed information including: demographic details, medical history, genotypes, and co-medication status and was randomly divided the dataset into a training set (90%) and a test set (10%) to evaluate the predictive capabilities of the model. Furthermore, feature selection methods were employed to identify factors associated with warfarin dosage. Our newly proposed heuristic-stacking ensemble learning outperforms traditional stacking ensemble learning in key metrics. Specifically, the accuracy in predicting the ideal dose improved to 73.44% (compared to 71.88% in traditional stacking method), with mean absolute errors of 0.11mg/day (compared to 0.13mg/day), root mean square errors of 0.18mg/day (compared to 0.20mg/day), and an R 2 value of 0.87 (compared to 0.82). The developed heuristic-stacking ensemble learning can satisfactorily predict warfarin dose with high accuracy. A relationship between hypertension, a history of severe preoperative embolism, and warfarin dose is found, which provides a useful reference for the warfarin dose administration in future.