SARS-CoV-2 variants and considerations of inferring causality on disease severity

The SARS-CoV-2 B.1.1.7 variant that was first identified in Kent (UK) in December, 2020, has now spread to many countries and shown consistent fitness advantage over other variants in circulation at the time.¹ This means an increase in transmission potential, which alone can lead to increased rates of hospitalisations and deaths. In The Lancet Infectious Diseases, Peter Bager and colleagues report the risk of hospitalisation with B.1.1.7 variant using the impressive population-level sequencing data in Denmark that include cases detected from both community-based and hospital-based testing.² All SARS-CoV-2-positive cases confirmed by RT-PCR in Denmark, sampled between Jan 1 and March 24, 2021, with 14 days of follow-up for COVID-19 hospitalisation were assessed for this observational study. COVID-19 hospitalisation was defined as first admission lasting longer than 12 h within 14 days of a sample with a positive RT-PCR result, and the study population and main analysis were restricted to the proportion of cases with viral genome data. Bager and colleagues calculated crude and adjusted risk ratios (RRs) of hospital admission, with adjustments done for several important confounders such as sex, age, calendar time, region, and comorbidities.

The analysis included 30 572 individuals with genomic data (60·0% of 50 958 positive cases with 14 days follow-up), of whom 10 544 (34·5%) had been infected with B.1.1.7. Compared with other lineages, the authors found a seemingly protective effect of B.1.1.7 (RR 0·79, 95% CI 0·72–0·87; p<0·0001) in the crude analyses but, after adjustment, B.1.1.7 was associated with increased risk of hospitalisation (1·42, 1·25–1·60; p<0·0001). These findings are consistent with early reports and strengthen the association between B.1.1.7 and increased disease severity observed previously (table ).⁴ Particularly increased severity observed with B.1.1.7 appeared to be specific to adults older than 30 years,⁵ and pronounced among those older than 65 years.⁸

Table.

Summary of studies assessing the association between B.1.1.7 and disease severity

	Number of cases with B.1.1.7	Data source	Adjustments for			Risk ratios
			Comorbidities	Deprivation	Time period
Studies assessing the risk of hospitalisation among those who tested positive
Bager et al2	10 544	Hospital and community	Yes	Yes	Yes	1·42 (1·25–1·60)
Dabrera et al3	6038	Hospital and community	No	No	Yes	1·34 (1·07–1·66)
PHS4	NA	Hospital and community	Yes	Yes	Yes	1·63 (1·48, 1·80)
HOCI4	2386	Hospital (ICU)	No	No	No	1·15 (0·86–1·53)
Nyberg et al5	27 710	Community	No	Yes	Yes	1·52 (1·47–1·57)
Studies assessing the risk of death among those hospitalised
Patone et al6	3400	Hospital (ICU)	Yes	Yes	Yes	0·93 (0·76–1·15)
CO-CIN4	216	Hospital	No	Yes	Yes	0·63 (0·20–1·69)
CO-CIN4	404	Hospital	No	Yes	Yes	0·67 (0·32–1·40)
CO-CIN4	412	Hospital	No	Yes	Yes	0·81 (0·50–1·32)
Frampton et al7	289	Hospital	Yes	No	Yes	1·12 (0·71–1·78)
Studies assessing the risk of death among those who tested positive
Davies et al1	674 539	Community	No	Yes	Yes	1·55 (1·39–1·72)
Patone et al6	80 494	Hospital and community	Yes	Yes	Yes	1·59 (1·25–2·03)
Grint et al8	91 775	Primary care	Yes	Yes	Yes	1·67 (1·34–2·09)
Challen et al9	54 906	Community	No	Yes	Yes	1·64 (1·32–2·04)
Dabrera et al3	6038	Hospital and community	No	No	Yes	1·06 (0·82–1·38)
PHS4	NA	Hospital and community	Yes	Yes	Yes	1·37 (1·02–1·84)

CO-CIN=COVID-19 Clinical Information Network. HOCI=COVID-19 Genomics UK Consortium Hospital Onset COVID-19 Infection Study. ICU=intensive care unit. NA=not applicable. PHE=Public Health England. PHS=Public Health Scotland.

The study highlights three key considerations when trying to attribute an increase in disease severity to a variant of concern that also increases transmission risk, in the context of surveillance data. First, potential confounders in this context include factors that increase acquisition risk overlapping with factors known to increase severity, irrespective of lineages, such as age or comorbidities. For example, because B.1.1.7 is associated with a higher secondary attack rate, then the outbreak setting could introduce confounding if not accounted for, especially if the setting, such as a congregate living or workplaces, is more likely to include individuals with comorbidities. Increased transmission potential with B.1.1.7 means that it has reached and concentrated, like the early lineages, among economically marginalised communities who might also have higher rates of comorbidities.¹⁰ Therefore, the attributable effect of variants of concern on disease severity should account for confounders in the pathway of infection risk, such as social determinants and outbreak settings, while also addressing confounders in the pathway to severity risk in the case of infection (age, sex, and comorbidities).

Second, selection biases can play a major role in drawing inference on relative severity risks.¹¹ The key player here is the reason for testing, and thus the sample selected for discerning the relative severity of B.1.1.7. For example, if a study excludes cases detected and admitted to hospital at the time of testing,¹² then the study population might underestimate the severity of the lineage that causes more severe infection. Similarly, if the study is restricted to individuals admitted to hospital,⁷ we might not observe an increased risk of death among those hospitalised even if B.1.1.7 increases mortality risk among those diagnosed. Additionally, an important selection bias that is common across most studies with surveillance data is that only a subset of all cases will have information on lineages, and missing information on lineage might not be random.

Third, increased transmission potential means that the lineage can take over and thus, some studies might be limited in comparing B.1.1.7 cases with historical cases of other lineages. This difference in timing of comparator cases could lead to confounding if other factors, such as hospital pressure, influence mortality.¹³ Timing of cases and comparators might also affect how readily cases of B.1.1.7 can be compared with those in a similar transmission network (confounders on the pathway of infection risk) when increased transmission potential leads to variant replacement. Therefore, evaluating the attributable relative risks of severity is particularly challenging when the risk factor in question also increases transmission potential. Determining relative severity across emerging variants will become increasingly challenging with SARS-CoV-2 vaccination, in the context of variable effectiveness, by factors such as vaccine coverage by social determinants of acquisition and transmission risks and differential vaccine effectiveness by lineages.

The mechanisms by which increased transmission potential might challenge our ability to estimate attributable severity further emphasise the crucial role that increased transmissibility potential plays in hospitalisations and mortality—irrespective of any direct effect of the variant on severity. As Bager and colleagues show, careful consideration of all three potential challenges requires detailed data and a systematic approach to matching or adjusting for confounding. Not all surveillance systems can support this, and an investment in data platforms that enable rapid and robust analyses of relative transmission potential and relative severity remain crucial as new variants emerge.

MC is a member of the New & Emerging Respiratory Threats Advisory Group. SM declares no competing interests.