Background
Many early COVID-19 outcome studies were based exclusively on electronic health record (EHR) data from single health systems. However, many people in 2020 who would have ordinarily received care in the same system experienced outcomes at other hospitals or at home – 2 common scenarios when COVID-19 temporarily transformed usual practice. We assessed completeness of COVID-19 outcomes among people with HIV (PWH) receiving HIV care in 3 New York City health systems, comparing EHR data alone to the gold standard of public health surveillance data.
Methods
We included adult (18+) PWH treated with antiretroviral therapy from 1/1/2019-3/1/2020 in the Montefiore, Mount Sinai and Northwell health systems. We obtained PCR-confirmed COVID-19 infections, hospitalizations (via diagnosis codes), ICU admissions and all-cause deaths from health system EHRs through 12/31/2020, before COVID-19 vaccines were widely available. We linked data with the New York City HIV and COVID-19 Surveillance Registries, which contain city-wide data on legally mandated reporting of HIV, COVID-19 and deaths. We compared numbers of COVID-19 outcomes identified between data sources. To assess potential bias when limiting analyses of COVID-19 outcomes to EHR data alone, we compared associations of selected characteristics with COVID-19 outcomes by source.
Results
Among 19833 PWH (73% men, 42% Black, 39% Latino, 16% white, median age 53), we identified 1372 (7%) COVID-19 infections, 435 (2%) hospitalizations, 70 (0.3%) ICU admissions and 460 deaths (2%) (Table). Using EHR data alone resulted in a 3-fold underestimation of COVID-19 infections, hospitalizations and ICU admissions and 5-fold underestimation of all-cause mortality. Among those who died, younger PWH were more likely to be missing from EHR after adjusting for site, race/ethnicity, sex, HIV transmission category and HIV viral suppression. In relation to pre-COVID HIV viral suppression, the adjusted odds ratio of all-cause mortality for pre-COVID HIV viremia (200+ cp/mL) was 1.62 (95% CI 0.96-1.70) using EHR data only and 2.20 (95% CI 1.76-2.75) using surveillance data. Adjusted associations of race/ethnicity differed by source across all 4 COVID-19 outcomes.
Conclusions
In 2020, analyses limited to EHR severely underestimated COVID-19 outcomes and resulted in important biases in outcome estimates and measures of association. Interpretation of estimates from EHR-based studies during public health emergencies should require careful consideration of ascertainment bias.
Table or Figure
