The possibilities of data mining methods for assessing the outcomes of COVID-19 in patients with diseases of the blood system

Abstract

Introduction. Various artificial intelligence technologies are widely used in many areas of medicine with integration into research and practical work, including hematology. The attractiveness of machine learning methods is due to the possibility of excluding the subjective factor both assessment of the patient's condition and examination results. Aim. The construction of a predictive survival model for hematological patients with COVID-19 coronavirus infection. Materials and methods. 144 medical records of patients with malignant and benign diseases of the blood system treated at the Regional Clinical Hospital No. 2 in Vladivostok were retrospectively analyzed. The average age of the studied patients was 64 years. The solid endpoint is the mortality of patients from all causes (46 people or 32%). Indicators such as the type of disease (malignant, benign); the stage of therapy; clinical manifestations of COVID-19 (yes/no); symptoms of infection were used as predictors for constructing predictive models; ECOG status at the time of admission; concomitant diseases; glucocorticosteroids therapy; the use of humidified oxygen and complications of COVID-19. When constructing predictive models with a binary classifier, machine learning methods were used: logistic regression, a decision tree based on “conditional inference” and a “random forest”. Results. 3 predictive models were developed. The choice of the model depended on the number of parameters included. According to the F-measure, the accuracy of the “random forest” model was higher. Based on the selected machine learning methods, the presence of respiratory failure requiring oxygen support was the most significant predictor of forecasting the outcome of COVID-19. Conclusion. Our study allowed us to identify significant predictors of an unfavorable outcome, on the basis of which prognostic models of survival of hematological patients with coronavirus infection were built.