Potential of an artificial intelligence-based predictive model in identifying alcohol delirium risk groups in patients with alcohol withdrawal syndrome

BACKGROUND: Alcohol withdrawal delirium is one of the most severe forms of alcohol withdrawal syndrome, occurring in 5%–12% of patients with alcohol dependence. Delirium can have serious consequences and is associated with a high risk of developing multiple organ pathology, death, pronounced cognitive impairments, or psycho-organic/amnestic syndrome. Therefore, identifying effective approaches to the prevention and treatment of alcohol withdrawal delirium is extremely important for addiction medicine practice. To achieve this goal, the development of reliable tools for predicting the onset of alcohol withdrawal delirium is necessary. A promising direction in this field is the application of machine learning-based algorithms. AIM: This study aimed to develop a method for predicting the onset of alcohol withdrawal delirium in patients with early manifestations of alcohol withdrawal syndrome based on mathematical analysis using artificial intelligence. METHODS: To build predictive models, four laboratory parameters were used: blood platelet count and serum levels of potassium, sodium, and chloride. The probabilistic model was based on the Multilayer Perceptron (MLP) algorithm, while the binary model was based on the Random Forest (RF) algorithm. For model training, an anonymized database of patients with alcohol withdrawal syndrome was used, comprising 498 individuals: 295 patients with developed alcohol delirium and 203 ones with uncomplicated alcohol withdrawal syndrome. RESULTS: Both models demonstrated good results on the test dataset: the average prediction accuracy for the model based on the MLP algorithm was 84%, and 83% for the RF algorithm. The MLP model was validated in an addiction treatment hospital setting, with an accurate prediction rate of 84.4%. CONCLUSION: The use of artificial intelligence opened the possibility to identify a risk group for alcohol delirium based on a scientific approach of objective and rapid laboratory parameters.