Development and Validation of Algorithms to Identify COVID-19 Patients Using a US Electronic Health Records Database:...

Area of Science:

• Medical Informatics
• Public Health Surveillance
• Epidemiology

Background:

• Accurate identification of COVID-19 cases in electronic health records (EHR) is crucial for public health surveillance.
• Traditional diagnostic codes may not fully capture the nuances of early COVID-19 detection.
• Evaluating algorithm performance requires comparison against a reliable reference standard, such as RT-PCR testing.

Purpose of the Study:

• To identify and evaluate candidate algorithms for detecting COVID-19 cases within an EHR database.
• To compare the utilization of acute respiratory disease codes before and during the COVID-19 pandemic.
• To assess the sensitivity and specificity of various diagnostic codes for COVID-19 identification.

Main Methods:

• Utilized de-identified EHR data to identify and develop 11 candidate algorithms for COVID-19 patient identification.
• Calculated the proportion of patients with SARS-CoV-2 RT-PCR tests within 10 days of diagnosis codes.
• Assessed algorithm performance using sensitivity, specificity, and likelihood ratios, adjusting for verification bias.

Main Results:

• From January to March 2020, R06 (dyspnea) and R05 (cough) were common COVID-19 related codes.
• Post-April 2020, the U07.1 code demonstrated high adjusted sensitivity (1.00) but low adjusted specificity (0.32) in hospitalized patients.
• No combination of ICD-10-CM codes met satisfactory sensitivity and specificity thresholds when validated against RT-PCR results.

Conclusions:

• Algorithms using the U07.1 code offer high sensitivity for hospitalized COVID-19 patients but suffer from low specificity.
• Current ICD-10-CM code combinations are insufficient for reliably detecting COVID-19 cases in EHRs with high accuracy.
• Further development of robust algorithms is needed for effective COVID-19 surveillance using EHR data.