Risk of mortality in patients infected with SARS-CoV-2 variant of concern 202012/1: matched cohort study

To the editors,
With great esteem, we have read the original article written by Challen et al. [1] The authors aim to establish if the VOC-202012/1 infection, a new variant of SARS-CoV-2, compared with other circulating variants has increased the mortality in SARS-CoV-2 infection. The study demonstrates higher mortality in patients with VOC-202012/1 infection (4.1 per 1000 patients) than infections caused by the other variants (2.5 per 1000 patients).
In the study, the authors used rigorous methodology to reduce most of the biases to which the cohort studies are susceptible, such as using the matched cohort approach and sensitivity analysis. The researchers focused their investigation on patients diagnosed in the community with a positive test between October 1, 2020, and January 29, 2021. They divided them into two groups: S gene positive, which included patients affected by the other variants, and S gene negative, including those affected by VOC-202012/1.
The existence of variants, already proven by several studies, is undeniable. Nevertheless, it is hard to measure the real burden of these new variants on social health. Even though the study of Challen et al. [1] follows strategies to minimize the risk of bias, the presence of certain biases is unavoidable, and the medical community should be aware of these risks.
During the study period, there was an increasing number of detection of the new variant, concomitantly with a new wave of transmission that led to the highest infection incidence in the United Kingdom. In this way, it is evident that the health care system was under one of the highest levels of demand, which may lead to an increase in mortality. Also, the health system overwhelming during the wave peak may increase the probability of underdiagnosed patients. These factors are important sources of uncertainty.
Even though the study followed the patients for a restricted period of months, the SARS-CoV-2 virus incidence varies a lot in a few amount of time. In this way, each variant may follow its own transmission wave, following each specific growth phase. [2] If a particular variant reaches its peak exactly when the health system is overwhelmed, its mortality may be worsened. [3] Comparing the variants in their different growth phase may elicit the time period bias, underestimating or overestimating the effect size. [4-6]
Sampling bias consists of improper conduct of selection. [6, 7] The study, even with a matching strategy, is subject to this bias. It occurs because there is no possibility of controlling all the prognostic variables of the SARS-CoV-2 infection, such as obesity and systemic hypertension, among others that were not matched.
The study stopped the follow-up of the patients on February 12, 2021, which may lead to the risk for attrition bias (15% of patients did not complete the 28 days follow-up). Several deaths of patients infected during the study follow-up did occur only after February 12, 2021. Considering that VOC-202012/1 infection was in frank ascension in the last months, maybe several deaths related to the variant were omitted. In this sense, the VOC-202012/1 infection-related death could be even higher.
Detection bias arises when the participants have not a uniform chance for certain diagnostic tests or treatments. [8, 9] In the present study, the risk for detection bias may be seen in certain situations. The first is the tendency of patients with mild symptoms not to look for medical care. Consequently, these patients may be underdiagnosed. Maybe patients who had previously been infected by the nonvariant form during the first SARS-CoV-2 wave are more susceptible to reinfection with milder symptoms [10, 11], raising the risk for underdiagnosis in this nonvariant group. If it were a correct statement, the denominator of the mortality rate would increase in this group. Therefore, by reversing this bias, there could be an increase in the mortality rate by the VOC-202012/1 when compared to other variants. Besides, there were patients in whom the searching for the sequencing of the new variant was not performed, and the criteria used for assessing for S gene status was not uniform. Authors focused their attention in 49.2% of the infections of the studied period that represents the community based covid-19 diagnoses (pillar 2), which are generally a younger population with less severe disease than hospital based covid-19 diagnoses (pillar 1). If there is a chance that the incidence of the VOC-202012/1 is different in pillar 1 (higher probability for more severe disease) and pillar 2 (higher probability for less severe disease), it could have a major impact on the final mortalities rates of both variant and nonvariant groups.
Challen et al. [1] performed a well-done study, considering their robust methodology. However, nulling all the bias by managing data and information available in a cohort study is challenging. In this way, it is essential that the medical community is aware of biases risks to not take hasty conclusions about variants.

REFERENCES:
[1]. Challen R, Brooks-Pollock E, Read JM, Dyson L, Tsaneva-Atanasova K, Danon L. Risk of mortality in patients infected with SARS-CoV-2 variant of concern 202012/1: matched cohort study. BMJ. 2021 Mar 9;372:n579. doi: 10.1136/bmj.n579. PMID: 33687922; PMCID: PMC7941603.
[2]. Viboud C, Simonsen L, Chowell G. A generalized-growth model to characterize the early ascending phase of infectious disease outbreaks. Epidemics. 2016 Jun 1;15:27-37
[3]. Davies NG, Abbott S, Barnard RC, Jarvis CI, Kucharski AJ, Munday JD, Pearson CAB, Russell TW, Tully DC, Washburne AD, Wenseleers T, Gimma A, Waites W, Wong KLM, van Zandvoort K, Silverman JD; CMMID COVID-19 Working Group; COVID-19 Genomics UK (COG-UK) Consortium, Diaz-Ordaz K, Keogh R, Eggo RM, Funk S, Jit M, Atkins KE, Edmunds WJ. Estimated transmissibility and impact of SARS-CoV-2 lineage B.1.1.7 in England. Science. 2021 Mar 3:eabg3055. doi: 10.1126/science.abg3055. Epub ahead of print. PMID: 33658326.
[4]. Delgado-Rodríguez M, Llorca J. Bias. J Epidemiol Community Health. 2004 Aug;58(8):635-41. doi: 10.1136/jech.2003.008466. PMID: 15252064; PMCID: PMC1732856.
[5]. Sjölander A, Humphreys K, Palmgren J. On informative detection bias in screening studies. Stat Med. 2008 Jun 30;27(14):2635-50. doi: 10.1002/sim.3091. PMID: 17918781.
[6]. K. Choi BC, P. Pak AW. Bias, overview. Wiley StatsRef: Statistics Reference Online. 2014 Apr 14.
[7]. Biele G, Gustavson K, Czajkowski NO, Nilsen RM, Reichborn-Kjennerud T, Magnus PM, Stoltenberg C, Aase H. Bias from self selection and loss to follow-up in prospective cohort studies. Eur J Epidemiol. 2019 Oct;34(10):927-938. doi: 10.1007/s10654-019-00550-1. Epub 2019 Aug 26. PMID: 31451995.
[8]. Horwitz RI, Feinstein AR. Alternative analytic methods for casecontrol studies of estrogens and endometrial cancer. N Engl J Med 1978;299:1089e94.
[9]. Arfè A, Corrao G. Tutorial: strategies addressing detection bias were reviewed and implemented for investigating the statins-diabetes association. J Clin Epidemiol. 2015 May;68(5):480-8. doi: 10.1016/j.jclinepi.2014.12.001. Epub 2014 Dec 4. PMID: 25554519
[10]. Boyton RJ, Altmann DM. Risk of SARS-CoV-2 reinfection after natural infection. Lancet. 2021 Mar 17:S0140-6736(21)00662-0. doi: 10.1016/S0140-6736(21)00662-0. Epub ahead of print. PMID: 33743219; PMCID: PMC7969128.
[11]. Lumley SF, O'Donnell D, Stoesser NE, Matthews PC, Howarth A, Hatch SB, Marsden BD, Cox S, James T, Warren F, Peck LJ, Ritter TG, de Toledo Z, Warren L, Axten D, Cornall RJ, Jones EY, Stuart DI, Screaton G, Ebner D, Hoosdally S, Chand M, Crook DW, O'Donnell AM, Conlon CP, Pouwels KB, Walker AS, Peto TEA, Hopkins S, Walker TM, Jeffery K, Eyre DW; Oxford University Hospitals Staff Testing Group. Antibody Status and Incidence of SARS-CoV-2 Infection in Health Care Workers. N Engl J Med. 2021 Feb 11;384(6):533-540. doi: 10.1056/NEJMoa2034545. Epub 2020 Dec 23. PMID: 33369366; PMCID: PMC7781098.

Dear Editor
Consistent with past findings, mortality rates in the data reported for this study differ widely across age groups (e.g., 0.11% in 30-59 year olds to 12.64% in over 80s). However, the data reported does not include CFR by age group for S gene positive vs. S gene negative. There are two reasons this information is important. First, it is practically important to know whether the variant changes the differences in CFR across age. Second, if the variant changes the CFR more in some age groups than others (i.e., there is an interaction between S gene positive/negative and age group), there are implications for the confidence intervals reported. If, for example, the increase in the overall CFR within the S gene negative group was caused by larger CFRs in older groups, the confidence intervals around the overall CFR would necessarily be wider due to the smaller sample size of older groups. A test of the interaction between S gene positivity/negativity and age would provide initial evidence about whether this is likely to affect conclusions.

Dear Editor,

The study by Challen et al.[1] indicates that the mortality in subjects infected by SARS-CoV-2 B.1.1.7 variant is higher than that observed in a control population whose infection did not include this variant. Challen and coworkers studied 54,906 matched pairs of participants who tested positive for SARS-CoV-2 between 1 October 2020 and 29 January 2021, followed-up until 12 February 2021. The outcome measure was death within 28 days of the first positive SARS-CoV-2 test result. Results showed that “mortality hazard ratio associated with infection with B.1.1.7 variant was 1.64 compared with infection with previously circulating variants (95% confidence interval 1.32 to 2.04); this represents an increase in deaths from 2.5 to 4.1 per 1000 detected cases.” However, this result might be influenced by the low representativeness of the enrolled population of 54,906 matched subjects because their mortality rate (0.25 to 0.41 per 100 detected case) was approximately 10-fold lower that the mortality rate generally found in UK and other Western countries (e.g. Germany, France, Italy); this is between 2.1 and 3.5 per 100 detected cases in 2021 according to the website “Our World in Data” based at the University of Oxford [2].

Andrea Messori, ESTAR Toscana, Regional Heath Service, Firenze (Italy)

References

1. Challen R, Brooks-Pollock E, Read JM, Dyson L, Tsaneva-Atanasova K, Danon L. Risk of mortality in patients infected with SARS-CoV-2 variant of concern 202012/1: matched cohort study. BMJ. 2021 Mar 9;372:n579. doi: 10.1136/bmj.n579. PMID: 33687922; PMCID: PMC7941603.
2. University of Oxford: “Our World in Data” website, page https://ourworldindata.org/mortality-risk-covid?country=ITA~GBR~DEU~FRA , accessed 14 March 2021

Dear Editor

Is it possible that the increased mortality observed by Challen et al, might be the result of confounding by viral load, a variable which does not appear to be controlled for in this study? Higher viral load may be linked to worse Covid outcomes (Burgess et al BMJ 2020;371:m4763).

The cohort infected with the more transmissible variant can likely be divided into those who had a community exposure to the virus which would not have resulted in an infection had it been to the original strain, and those cases where it would have caused infection anyway, even if it had been to the original variant. Is it not probable that the later group are likely to have received a higher inoculum than the cohort who had a comparable environmental exposure to the original variant?

If so, would this not call into question the conclusion that the new variant, in addition to being more transmissible, seems to be more lethal? If viral load were a confounder, then is it not possible that the observed increased mortality might be accounted for by the increased transmissibility alone.

Another questionable conclusion is that the data lend weight to the argument for further stringent measures. Is it not possible that the global curtailment of human interactions over the past year has given a selection advantage to more transmissible viral strains? If so, the somewhat unsettling implications of this are that whereas social restrictions may prove very effective as emergency measures locally, extensive prolonged use risks increasing the likelihood of the virus adapting to the lockdown environment. This creates a dilemma for policy makers who tend to think and work locally but, in this instance at least, may need to think globally.

Kind Regards

Vincent