Dear Editor,
Re: Association of Inpatient Use of Angiotensin Converting Enzyme Inhibitors and Angiotensin II Receptor Blockers with Mortality Among Patients With Hypertension Hospitalized With COVID-19 by Pen Zhang et al.
We read with interest the article by Pen Zhang and colleagues 1 who describe an association between in hospital use of Angiotensin Converting Enzyme Inhibitors (ACEi) or Angiotensin II Receptor Blockers (ARB) and all-cause mortality among 1,128 patients diagnosed with COVID-19 in a multicenter study. The authors use epidemiological methods to minimize confounding by indication, including restriction (to subjects with hypertension), propensity score matching to adjust for observed confounding, and e-values to assess the potential impact of residual confounders. After these, Zhang and co-authors reach the striking results that ACEi/ARB use is associated with a 60% reduction in mortality, with adjusted HR 0.42 [95% CI, 0.19-0.92] and propensity-matched HR 0.37 [0.15-0.89]. The accompanying editorial by Dr Ravi Shah et al 2 further scrutinized their results, where limitations including lack of randomization are discussed.
While we are aware of the interest in the estimation of the effect of ACEi/ARB in the current COVID-19 pandemic, we are concerned with two major methodological issues in the design of this study, which if understood correctly, could substantially bias the results and led to misleading findings. First, Zhang et al. do not provide sufficient detail in their manuscript on how they handled patient follow-up in their analysis. We are worried that their definition of index date, which separates the preceding exposure time from the follow-up time, is ill-specified (“Patients with hypertension who received ACEi/ARB during hospitalization were classified as ACEi/ARB group”) and it is unclear where their time zero for follow up starts. In Figure 2, no events are observed in the ‘ACEi/ARB group’ during at least 1 week after start of follow up, making us suspect that the authors have failed to account for the time from cohort entry to start of target treatment/s therefore resulting in immortal time bias3. Well known in pharmaco-epidemiology, immortal time bias results in an artificial inflation of the denominator of person-time in the exposed group, as well as an artificial inflation of the numerator in the unexposed group due to the patients who experience the outcome (here death) before fulfilling the exposure definition, hence leading to a reduction in risk and rates of events in the treatment group/arm. Many authors have discussed different methods to minimize or avoid immortal time bias, including the use of target trial emulation4 . We believe the analyses of Zhang P et al. need further detail and, if needed, a re-analysis after accounting for this important issue. These issues can also be tested by the authors as we would expect other antihypertensive drugs, for example diuretics, to show similar protective associations when analyzed using the same design that detects exposure during hospitalization versus a non-exposed group.
Secondly, we worry that the comparison with a non-user control group as the primary analysis is problematic. Post-hospitalization events define non-users in this analysis and there is no account for medicine use in the pre-hospitalization setting. Therefore, it is possible that non-users of ACEi/ARB may have used these medicines prior to hospital admission. Further, non-users typically differ from medicine users in many ways, and it is unlikely that any of the methods used can control for such differences sufficiently. An active comparator group is a better choice, and it is clear from the results that many other anti-hypertensives find use in the study population, thereby enabling a standard active comparator cohort analysis. A sub-analysis using users of other anti-hypertensives was conducted by Zhang et al which still seemed subject to substantial immortal time bias as suggested by Figure 2b
These methodological issues need addressing before suggestions about a potential protective role, or more importantly as the authors suggest evidence of no harm, of ACEi/ARB can be made and we are in full agreement with the linked editorial that further research is needed.
Prof Daniel Prieto-Alhambra, CSM, NDORMS, University of Oxford, Oxford, UK
Prof Fredrik Nyberg, School of Public Health and Community Medicine, Institute of Medicine, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden
Dr Nicole L. Pratt, Quality Use of Medicines and Pharmacy Research Centre, University of South Australia, Adelaide, Australi
Dr Daniel R. Morales, Division of Population Health and Genomics, University of Dundee, UK
Dr Thamir M. Alshammari, Saudi Food and Drug Authority, and King Saud University, Riyadh, Saudi Arabia
Prof Patrick B. Ryan, Observational Health Data Sciences and Informatics, New York, Janssen Research & Development, and Department of Biomedical Informatics, Columbia University
Prof Marc A. Suchard, Observational Health Data Sciences and Informatics, New York, and Department of Biostatistics, UCLA Fielding School of Public Health, University of California, Los Angeles
- Zhang P, Zhu L, Cai J, Lei F, Qin JJ, Xie J, Liu YM, Zhao YC, Huang X, Lin L, Xia M, Chen MM, Cheng X, Zhang X, Guo D, Peng Y, Ji YX, Chen J, She ZG, Wang Y, Xu Q, Tan R, Wang H, Lin J, Luo P, Fu S, Cai H, Ye P, Xiao B, Mao W, Liu L, Yan Y, Liu M, Chen M, Zhang XJ, Wang X, Touyz RM, Xia J, Zhang BH, Huang X, Yuan Y, Rohit L, Liu PP and Li H. Association of Inpatient Use of Angiotensin Converting Enzyme Inhibitors and Angiotensin II Receptor Blockers with Mortality Among Patients With Hypertension Hospitalized With COVID-19. Circ Res. 2020.
- Shah R, Murthy VL and Koupenova M. ACEing COVID-19: A Role For Angiotensin Axis Inhibition in SARS-CoV-2 infection? Circ Res. 2020.
- Suissa S. Immortal time bias in pharmaco-epidemiology. Am J Epidemiol. 2008;167:492-9.
- Hernan MA, Sauer BC, Hernandez-Diaz S, Platt R and Shrier I. Specifying a target trial prevents immortal time bias and other self-inflicted injuries in observational analyses. J Clin Epidemiol. 2016;79:70-75.
