Potential Pathological, Clinical, and Symptomatic Findings of COVID-۱۹ to Predict Mortality in Positive PCR Individuals Using Data Mining

Introduction: COVID-۱۹ has placed immense burdens on healthcare systems and medical staff. To avoid spread, the statistician’s role and the use of appropriate predictive models -prediction of survivors versus non-survivors- is highly relevant. This study aimed to apply a model which avoids overfitting and selection bias towards selecting predictors to predict COVID-۱۹ mortality.   Materials and Methods: The Conditional Inference Tree (CIT) model was used. Data from ۵۹,۵۶۴ hospitalized individuals with positive polymerase chain reaction (PCR) test results were collected from February ۲۰, ۲۰۲۰, to September ۱۲, ۲۰۲۱, in the Khorasan Razavi province, Iran.   Results: The sensitivity and specificity of the model were ۸۸.۷% and ۸۸.۱%, respectively, the accuracy was ۸۸.۲%, and the area under the curve (AUC) was ۷۳.۰% on test data. Therefore, the model had considerable accuracy in prediction. The potential predictors involved in predicting survivors versus non-survivors were intubation, age, PO۲ level, decreased consciousness level, presence of distress, anorexia, drug use, and kidney diseases.   Conclusion: According to the findings, the CIT model showed high accuracy by avoiding overfitting and selection bias toward selecting predictors. Thus, the results of this study and the efforts of healthcare systems to stop the spread of this pandemic prove helpful.

Keywords:

Conditional Inference Tree , COVID-۱۹ , Data mining , Decision Trees , Machine Learning