Development and validation of risk prediction models for COVID-19 positivity in a hospital setting

Area of Science:

• Clinical Medicine
• Epidemiology
• Biostatistics

Background:

• Accurate prediction of coronavirus disease-2019 (COVID-19) positivity is crucial for effective patient management and resource allocation in healthcare settings.
• Developing reliable risk prediction models using readily available clinical parameters can significantly enhance diagnostic capabilities.

Purpose of the Study:

• To develop and validate two risk prediction models for COVID-19 positivity.
• To create nomograms and probability charts for practical clinical application in general hospitals.

Main Methods:

• A cohort of 1330 patients from four Hong Kong hospitals were included, with data randomly split for model development (n=895) and validation (n=435).
• Multivariable logistic regression was employed for model creation, with validation using the Hosmer-Lemeshow (H-L) test and calibration plots.
• Nomograms and probability calculations were generated to assess sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV).

Main Results:

• Two prediction models were developed: Model 1 included age, white blood cell count, chest x-ray, and contact history (AUC=0.911).
• Model 2 excluded contact history but maintained strong predictive power (AUC=0.880).
• Both models demonstrated excellent external validation through H-L tests and calibration plots, confirming their reliability.

Conclusions:

• Two user-friendly, validated nomograms for COVID-19 risk prediction were successfully developed.
• These nomograms, based on accessible parameters, exhibit high accuracy (excellent AUCs) and are suitable for clinical implementation.