The long COVID evidence gap: comparing self-reporting and clinical coding of long COVID using longitudinal study data linked to healthcare records.

The term "long COVID" (LC) was coined in spring 2020 by individuals with ongoing symptoms following COVID-19, but it took until December 2020 for clinical codes to be created in order to record persistent post-COVID-19 illness and referrals within electronic health records (EHRs). Analysis of whole-population EHR databases have helped understand the epidemiology of LC; yet concerns exist about the completeness of accessible EHRs for LC. UK longitudinal population studies (LPS) collected self-reported data on COVID-19 and LC from early 2020 and deposited these data in the UK Longitudinal Linkage Collaboration (UK LLC) research database where they are systematically linked to the participants EHRs. Comparisons of LPS reported LC with recorded LC in the EHRs of the same individuals may be helpful in understanding the epidemiology of emerging conditions such as LC. We used data from 10 UK LPS in the UK LLC to investigate whether participants self-reporting LC had a LC diagnosis or referral code in their English EHR after 10 to 22 months of follow up. Of 6412 participants with COVID-19 symptom duration data and linkage to health records, 898 (14.0%) self-reported LC of any severity in LPS surveys. Among these, just 42 (4.7%; 95% CI: 3.5, 6.3) were identified with LC-related codes in EHRs. In individuals reporting debilitating LC, this proportion was only marginally higher (5.6%; 95% CI: 3.7, 8.3). Our data show a striking discrepancy between LC as perceived and reported by participants in LPS and evidence of LC recorded in their EHRs; and that this discrepancy was patterned by ethnicity and possibly by indicators of deprivation. Self-reported symptoms may not be reflected in coded EHRs due to factors including variations in individuals help seeking behaviours, clinician coding practices and the availability of appropriate codes. However, these considerations appear unlikely to provide a complete explanation for the substantial observed reporting discrepancy. These results may indicate substantial unmet clinical need, in keeping with patient reports of difficulties accessing healthcare and sub-optimal recognition of, and response to, their illness when they do. They may also indicate potential shortcomings of epidemiological research on LC based on EHR- or LPS-based ascertainment alone and illustrate the value of triangulation between LPS and EHR data where linked and made available through resources such as the UK LLC.