Estimating the Number of Primary vs Incidental COVID-19 Hospitalizations in Santa Clara County

Area of Science:

• Public Health Surveillance
• Health Informatics
• Machine Learning in Healthcare

Background:

• Accurate monitoring of COVID-19 trends is crucial for public health.
• Differentiating primary COVID-19 hospitalizations from incidental cases is challenging using standard discharge data.
• Local health departments require efficient surveillance tools to manage public health crises.

Purpose of the Study:

• To evaluate the efficacy of International Classification of Diseases, 10th Revision (ICD-10) discharge data in distinguishing primary COVID-19 hospitalizations.
• To explore the application of machine learning algorithms for improving COVID-19 surveillance capabilities within a local public health setting.

Main Methods:

• Utilized discharge data from 5122 Santa Clara County hospitalizations with confirmed SARS-CoV-2.
• Trained multiple machine learning models to classify primary COVID-19 hospitalizations against a chart review gold standard.
• Employed Area Under the Receiver Operating Characteristic Curve (AUROC) to evaluate model performance.

Main Results:

• All trained models demonstrated strong performance, with AUROC values ranging from 0.808 to 0.818.
• Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression was selected for implementation due to comparable performance, transparency, and familiarity.
• The chosen model successfully differentiated primary COVID-19 hospitalizations.

Conclusions:

• International Classification of Diseases, 10th Revision (ICD-10) discharge data can be effectively used for monitoring primary COVID-19 hospitalizations.
• Predictive algorithms offer a low-burden solution for local health jurisdictions to meet surveillance needs.
• This approach minimizes manual effort in public health surveillance, enhancing efficiency.