Acta Biomedica Scientifica (Dec 2024)

Predicting the risk of developing drug-induced liver injury during remdesivir therapy in COVID-19 patients using machine learning

  • Yu. V. Shevchuk,
  • I. I. Shamigulov,
  • I. V. Sychev,
  • A. V. Kryukov,
  • I. I. Temirbulatov,
  • K. B. Mirzaev,
  • N. P. Denisenko,
  • Sh. P. Abdullaev,
  • S. N. Tuchkova,
  • V. I. Vechorko,
  • O. V. Averkov,
  • D. A. Sychev

DOI
https://doi.org/10.29413/abs.2024-9.6.6
Journal volume & issue
Vol. 9, no. 6
pp. 52 – 62

Abstract

Read online

Background. The antiviral drug Remdesivir has been widely used for etiotropic treatment of COVID-19. The incidence of adverse reactions during Remdesivir therapy reaches 66.2 %, the most common one being an increase in hepatic transaminases.The aim. To develop a machine learning model for predicting the risk of drug-induced liver damage in patients with COVID-19 when prescribing Remdesivir therapy.Materials and methods. This prospective open-label observational study was conducted between November 2021 and February 2022, including 154 patients receiving Remdesivir therapy. Patients were divided into two groups: group 1 (n = 45), in which patients developed signs of liver damage during Remdesivir therapy; group 2 (n = 109) – patients without this adverse reaction. All patients underwent pharmacogenetic study and retrospective analysis of medical histories, database with the results of the conducted studies was formed, basing on which machine learning models for predicting the risk of drug-induced liver damage were trained.Results. The main prognostic factors included body mass index (relevance – 12.03 %) and carriage of AG genotype at polymorphic marker rs776746 of CYP3A5 gene (relevance – 10.04 %). Subsequently, for all obtained factors and based on Сategorical boosting a model for predicting the development of drug-induced liver damage with 57.8 % sensitivity and specificity of 80.7 % was developed.Conclusions. A risk model for the development of drug-induced liver damage during remdesivir therapy was built using machine learning. Body mass index and carriage of AG genotype at polymorphic marker rs776746 of CYP3A5 gene turned out to be key markers. To improve the accuracy of the model, an increase in the proportion of patients with adverse reactions in the training sample is required. Further studies will improve the quality of the model and integrate it into clinical practice.

Keywords