Mortality risk assessment of COVID-۱۹ Patients using Cause-specific hazard regression models: The Khorshid COVID-۱۹ Cohort study

Background and aims: Simple tools are needed so that clinicians can identify high-risk fromlow-risk patients during hospitalization. Predicting the clinical course of COVID-۱۹ patientshelps manage patients in the hospital and make correct and timely clinician decisions. It has beenproven that Machine learning and survival analysis methods effectively predict the mortality ofCOVID-۱۹ patients.Method: Using Cause-specific hazard regression models on the Khorshid Cohort COVID-۱۹(KCC) data set, a color risk chart was presented using the following variables at hospital admission:age, gender, oxygen saturation, Charlson comorbidity index, Blood Urea Nitrogen, Creatinine,Bicarbonate (HCO۳), Partial Pressure of Carbon Dioxide (PCO۲), Lymphocyte, and bloodplatelet count. The evaluation of the model was performed on the validation data, and the area underthe receiver operator curve (AU-ROC) and the calibration curve with the related parameters,including calibration-in-the-Large, calibration in Slope, and R-Square, was provided.Results: Age over ۷۵ years (HR=۳.۱۰ [CI۹۵%: ۰.۵۹-۱۶.۲۱]) was the most effective variable in themortality of COVID-۱۹ patients. The risk analysis method in the validation set had excellent diagnosticaccuracy and suitable coverage for the entire risk values regarding the calibration curve(Calibration in the Large =۰, Calibration on Slope =۱). In the validation data, = ۰.۷۲ and AUC =۰.۹۷ [CI۹۵%: ۰.۸۶-۰.۹۱] were reported.Conclusion: The proposed risk chart can help with resource allocation and management of medicalcenters during the pandemic as part of the pandemic preparedness. The external validation ofthe proposed model is the focus of our future activity.

کلیدواژه ها:

Machine Learning ، COVID-۱۹ ، Mortality ، Prognosis ، Risk assessment ، Risk chart ، Calibration