Specifying uniform eligibility criteria to strengthen causal inference studies of long-term outcomes of COVID-19

Background Causal interpretation of findings from existing epidemiological studies on long-term clinical outcomes of coronavirus disease 2019 (COVID-19) may be limited by the choice of comparator (control) group. Objective We compare two approaches to control group selection (based on requirement for negative SARS-CoV-2 test for eligibility) in long-term clinical outcomes after COVID-19 in patients with history of heart failure (HF). Design Retrospective cohort study using data from February 1, 2020 to July 31, 2021. Setting: Veteran Health Administration (VHA). Participants We studied two cohorts of Veterans with COVID-19 and history of HF which selected comparison group using two different approaches. In Cohort I, Veterans with HF who tested for positive for SARS-CoV-2 were age, sex, and race matched to Veterans with no evidence of COVID-19 in 1:5 ratio. In Cohort II Veterans with HF who tested positive for SARS-CoV-2 were age, sex, and race matched with Veterans with HF who tested negative for SARS-CoV-2 within +/-15 days of the positive test date within the same VHA facility. Exposure COVID-19 as determined by a positive SARS-CoV-2 test. Main Outcomes and Measures 1-year all-cause mortality and hospital admissions beyond the first 30 days after COVID-19 diagnosis. Adjusted hazard ratios (HRs) accounting for comorbidity and 95% confidence intervals were calculated. Results Cohort I comprised 13,722 Veterans with HF with COVID-19 (mean [SD] age 72.0 [10.2] years, 2.4% female, 71.1% White) and 60,956 matched controls not known to have COVID-19. Cohort II comprised 6,725 Veterans with HF with COVID-19 (mean [SD] age 72.5 [7.5] years, 0.1% female, 80.8% White) and 6,726 matched controls with negative SARS-CoV-2 test. The adjusted HRs for 1-year mortality and hospital admission beyond the first 30 days after diagnosis of COVID-19 were 1.40 (1.32-1.49) and 1.34 (1.28-1.41), respectively, in analysis of Cohort-I (where the comparator group was not required to test negative for SARS-CoV-2). However, in Cohort-II (using the second comparator group specifying negative SARS-CoV-2 test for eligibility), the associations were markedly attenuated; adjusted HRs 1.05 (0.95-1.17) and 1.07 (0.96-1.19), respectively. Conclusions We found significant attenuation of associations between COVID-19 and long-term risk of mortality and hospital admissions beyond the first 30 days among patient with existing HF, when comparing with a control group selected based on a negative SARS-CoV-2 test versus control group not known to have COVID-19. The findings have implications for the design of studies of long-term CVD (and non-CVD) outcome of COVID-19.