Characterization and Validation of EHR Computable Phenotypes for Long COVID Using Patient-Reported Symptoms: Insights...

Area of Science:

• Computational biology
• Medical informatics
• Public health

Background:

• Long COVID (LC) presents a significant challenge due to its complex nature and the need for effective patient identification methods.
• Advanced computational tools are essential for characterizing LC patients and improving diagnostic capabilities.

Purpose of the Study:

• To develop and validate a computable phenotype algorithm for identifying Long COVID patients using electronic health record (EHR) data.
• To characterize EHR features associated with highly symptomatic LC within the RECOVER-Adult cohort.

Main Methods:

• Utilized FHIR-sourced EHR data from 1,501 RECOVER-Adult cohort participants.
• Trained a machine learning model to identify LC based on diagnoses, medications, lab tests, and vital signs.
• Validated the algorithm against patient-reported symptoms and assessed model performance using AUROC and AUPRC.

Main Results:

• The study identified 376 (25%) participants with highly symptomatic LC.
• Key EHR features associated with LC included shortness of breath, fatigue, cardiac dysrhythmias, and specific medications.
• The developed algorithm achieved an AUROC of 0.80 and AUPRC of 0.58 for identifying highly symptomatic LC patients.

Conclusions:

• A machine learning model effectively identifies patients with Long COVID using EHR data.
• This computable phenotype algorithm can aid in identifying high-probability LC cases within healthcare systems.
• The model facilitates patient screening, recruitment for clinical trials, and etiological research for Long COVID.