AI-BASED PREDICTION OF OTOTOXICITY AND HEARING COMPLICATIONS IN ICU PATIENTS RECEIVING AMINOGLYCOSIDES OR CHEMOTHERAPY

Abstract

One of the serious and under-recognized side effects of aminoglycosides and/or chemotherapy in the intensive care unit, particularly in high dose units, in those who are treated for a prolonged period, and who have renal dysfunction and are exposed to multiple risk factors at the same time, is ototoxicity. Early detection of hearing problems is critical to avoiding permanent hearing loss and better patient outcomes. The present paper proposes a methodology based on Artificial Intelligence (AI) to predict the risk of hearing-related complications and ototoxicity in patients admitted to the Intensive Care Unit (ICU) using clinical, pharmacological, laboratory, and auditory parameters. The machine learning models were created using patient-level information, including length of intensive care unit (ICU) stay, use of aminoglycosides, chemotherapy, cumulative dose of drug, renal function markers, baseline hearing, inflammatory markers, and treatment intensity. These supervised learning algorithms, including logistic regression, random forest, support vector machine, gradient boosting machine, and XGBoost, were trained and tested, and accuracy, precision, recall, F1-score, and area under the receiver operating characteristic (ROC) curve were used to measure them. The ensemble-based models were more effective at predicting the class labels than the traditional classifiers, indicating the ability of the ensemble-based models in capturing the complex nonlinear relationships between drug exposure, physiological instability, and hearing deficits. Cumulative aminoglycoside dose, exposure to chemotherapy, serum creatinine level, length of stay in the intensive care unit, age, and baseline auditory threshold were the most important factors identified by the feature importance analysis as determining ototoxicity. Patients with a high cumulative dose of ototoxic drugs and combined nephrotoxicity indicators had the highest risk for developing hearing complications, as determined from risk stratification results. The framework proposed aims to be based on artificial intelligence with a useful clinical decision support system to detect ototoxicity in critical patients. The findings of this research can be utilized to include anticipatory analytics in monitoring systems utilized in the ICU to assist with clinical medication safety and in customizing treatment plans and auditing for a specific individual.