Detection of potential drug-drug interactions for drug-induced liver injury using machine-learning algorithms

2023 Fiscal Year Final Research Report

• Project/Area Number: 22K15347
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 47060:Clinical pharmacy-related
• Research Institution: Nihon University
• Principal Investigator: AKIMOTO Hayato 日本大学, 医学部, 助教 (80847658)
• Project Period (FY): 2022-04-01 – 2024-03-31
• Keywords: 薬物間相互作用 / 薬物性肝障害 / 機械学習 / 人工知能 / リアルワールドデータ

Outline of Final Research Achievements

The number of elderly patients receiving polypharmacy tends to increase by the aging of Japan's population. One of the most common adverse drug events is liver injury, and these patients receiving polypharmacy may be unintentionally prescribed drugs that increase the risk of liver injury. In this study, potential drug-drug interactions for increased risk of drug-induced liver injury were identified from electronic medical records using machine learning models. The machine learning models such as the logistic least absolute shrinkage and selection operator (LASSO) model and the gradient boosting decision tree model showed better predictive performance than multiple logistic regression model, one of the traditional statistical models, and suggested that the combination of diclofenac and famotidine significantly has been associated with an increased risk of drug-induced liver injury.

Free Research Field

薬剤疫学

Academic Significance and Societal Importance of the Research Achievements

本研究では電子カルテ情報を機械学習により解析することで, 薬物誘発性肝障害発症リスクに関連する潜在的な薬物間相互作用の検出を試みた. 機械学習アルゴリズムはOTC医薬品としても使用されているジクロフェナクやファモチジンの組み合わせが肝障害のリスクを高めることを示唆し, 医療機関に受診している患者のみならず, セルフメディケーションを実施している国民へも一定の注意を払う必要があると思われる. また従来の薬剤疫学的薬物間相互作用は線形回帰分析や不均衡分析により検出されてきたが, 機械学習を用いた検出はこれらの手法よりも高い汎化性を示しつつ相互作用の検出が可能な手法として今後活用が期待できる.