A Risk Scoring System Utilizing Machine Learning Methods for Hepatotoxicity Prediction One Year After the Initiation of Tyrosine Kinase Inhibitors

Ji Min Han, Jeong Yee, Soyeon Cho, Min Kyoung Kim, Jin Young Moon, Dasom Jung, Jung Sun Kim, Hye Sun Gwak

Research output: Contribution to journalArticlepeer-review

Abstract

Background: There is currently no method to predict tyrosine kinase inhibitor (TKI) -induced hepatotoxicity. The purpose of this study was to propose a risk scoring system for hepatotoxicity induced within one year of TKI administration using machine learning methods. Methods: This retrospective, multi-center study analyzed individual data of patients administered different types of TKIs (crizotinib, erlotinib, gefitinib, imatinib, and lapatinib) selected in five previous studies. The odds ratio and adjusted odds ratio from univariate and multivariate analyses were calculated using a chi-squared test and logistic regression model. Machine learning methods, including five-fold cross-validated multivariate logistic regression, elastic net, and random forest were utilized to predict risk factors for the occurrence of hepatotoxicity. A risk scoring system was developed from the multivariate and machine learning analyses. Results: Data from 703 patients with grade II or higher hepatotoxicity within one year of TKI administration were evaluated. In a multivariable analysis, male and liver metastasis increased the risk of hepatotoxicity by 1.4-fold and 2.1-fold, respectively. The use of anticancer drugs increased the risk of hepatotoxicity by 6.0-fold. Patients administered H2 blockers or PPIs had a 1.5-fold increased risk of hepatotoxicity. The area under the receiver-operating curve (AUROC) values of machine learning methods ranged between 0.73-0.75. Based on multivariate and machine learning analyses, male (1 point), use of H2 blocker or PPI (1 point), presence of liver metastasis (2 points), and use of anticancer drugs (4 points) were integrated into the risk scoring system. From a training set, patients with 0, 1, 2-3, 4-7 point showed approximately 9.8%, 16.6%, 29.0% and 61.5% of risk of hepatotoxicity, respectively. The AUROC of the scoring system was 0.755 (95% CI, 0.706-0.804). Conclusion: Our scoring system may be helpful for patient assessment and clinical decisions when administering TKIs included in this study.

Original languageEnglish
Article number790343
JournalFrontiers in Oncology
Volume12
DOIs
StatePublished - 8 Mar 2022

Keywords

  • hepatotoxicity
  • machine learning
  • prediction
  • risk scoring system
  • tyrosine kinase inhibitor

Fingerprint

Dive into the research topics of 'A Risk Scoring System Utilizing Machine Learning Methods for Hepatotoxicity Prediction One Year After the Initiation of Tyrosine Kinase Inhibitors'. Together they form a unique fingerprint.

Cite this