Developing a mortality risk prediction model using data of 3663 hospitalized COVID-19 patients: a retrospective...

Since the declaration of COVID-19 as a pandemic, a wide between-country variation was observed regarding in-hospital mortality and its predictors. Given the scarcity of local research and the need to prioritize the provision of care, this study was conducted aiming to measure the incidence of in-hospital COVID-19 mortality and to develop a simple and clinically applicable model for its prediction.

COVID-19-confirmed patients admitted to the designated isolation areas of Ain-Shams University Hospitals (April 2020-February 2021) were included in this retrospective cohort study (n = 3663). Data were retrieved from patients' records. Kaplan-Meier survival and Cox proportional hazard regression were used. Binary logistic regression was used for creating mortality prediction models.

Patients were 53.6% males, 4.6% current smokers, and their median age was 58 (IQR 41-68) years. Admission to intensive care units was 41.1% and mortality was 26.5% (972/3663, 95% CI 25.1-28.0%). Independent mortality predictors-with rapid mortality onset-were age ≥ 75 years, patients' admission in critical condition, and being symptomatic. Current smoking and presence of comorbidities particularly, obesity, malignancy, and chronic haematological disorders predicted mortality too. Some biomarkers were also recognized. Two prediction models exhibited the best performance: a basic model including age, presence/absence of comorbidities, and the severity level of the condition on admission (Area Under Receiver Operating Characteristic Curve (AUC) = 0.832, 95% CI 0.816-0.847) and another model with added International Normalized Ratio (INR) value (AUC = 0.842, 95% CI 0.812-0.873).

Patients with the identified mortality risk factors are to be prioritized for preventive and rapid treatment measures. With the provided prediction models, clinicians can calculate mortality probability for their patients. Presenting multiple and very generic models can enable clinicians to choose the one containing the parameters available in their specific clinical setting, and also to test the applicability of such models in a non-COVID-19 respiratory infection.