(248) Application of Machine Learning methods for Prediction of Tyrosine kinase inhibitor-induced hepatotoxicity in Cancer Patients

Publication Number: 1239

Background: Although several studies have performed tyrosine kinase inhibitor (TKI)-induced hepatotoxicity, the relationship and risk factors between TKI and the incidence of hepatotoxicity are still unclear.

Objectives: This study aimed to predict the risk of TKI-induced hepatotoxic adverse reactions through machine learning methods using health insurance claim data, which is large-scale real-world data.

Methods: This case-control study used population-based representative claims data from the National Health Insurance Service in Korea from 2002-2019. Cancer and hepatotoxicity were defined using the ICD-10 code. The cases were defined as the occurrence of hepatotoxicity at least once after cancer. The controls were randomly 1:1 matched to cases by age, sex, and index date.
Based on the significant odds ratio in multivariate analysis, input features used for machine learning were constructed. Machine learning methods, 10-fold cross-validated multivariate logistic regression, ridge, lasso, elastic net, and random forest were used to predict risk factors for the occurrence of hepatotoxicity. The area under the receiver-operating curve (AUROC) was analyzed to assess clinical performance.

Results: A total of 24,468 cancer patients were included in the analysis. The patient with TKIs had an increased risk of hepatotoxicity compared with the patients without TKIs by about 2.5-fold (AOR 2.45, 95% CI 1.41-4.25).
The TKI duration, cumulative dose of TKI, and average cumulative dose of TKI were associated with the increased risk of hepatotoxicity by 2.45-3.29 folds.
Among comorbidities, liver-related diseases, including chronic liver disease, nonalcoholic fatty liver disease (NAFLD), hepatitis B, liver cancer, and pancreatitis, increased the incidence of hepatotoxicity approximately 1.6-2.5 times.
The concomitant use of medications, including anticancer agents except for TKIs, antipyretic agents, antimicrobials, psychotropic agents, antituberculosis agents, etc was associated with the increased risk of hepatotoxicity by 1.1-2.5 folds.
Each patient with EGFR and VEGF was associated with an increased risk of hepatotoxicity by 2.77 fold(EGFR: AOR 2.77 95% CI 1.11-6.93), 6.71 fold(VEGF: AOR 6.71 95% CI 1.57-28.75)
The AUROC values of machine learning methods ranged between 0.639-0.644 using the input features.

Conclusions: This study revealed that the use of TKIs in cancer patients was associated with an increased risk of hepatotoxicity. The performance of the prediction models showed sufficient accuracy. This prediction model may be helpful for clinical decision-making with the administration of TKIs in cancer patients.