Skip to main content

Our take —

In a matched pair case-control study, available as a preprint and thus not yet peer-reviewed, comparing breakthrough cases in 247 partially and 149 fully immunized adults with BNT162b2 vaccine to unvaccinated controls in Israel, the odds of SARS-CoV-2 infection by the B.1.1.7 variant versus the wild-type were twice as high among partially immunized adults compared to unvaccinated controls but there was no difference between fully vaccinated cases and unvaccinated controls. All 8 B.1.351 variant infections in the fully immunized group were detected prior to the 14-day threshold for optimal immune protection by BNT162b2 vaccination. Observed associations are not measures of vaccine effectiveness, as study inclusion was restricted to individuals with confirmed SARS-CoV-2 infections.

Study design

Case-Control

Study population and setting

Between January 3 and March 7, 2021, individuals with SARS-CoV-2 infections, confirmed by RT-PCR and sequenced (for B.1.1.7, B.1.351, or wildtype lineages), were identified in Israel and assigned to one of three study groups: 1) infections detected from two weeks after the first dose to seven days after the 2nd dose of receiving BNT162b2 mRNA SARS-CoV-2 vaccine (“partial effectiveness” cases), 2) infections detected seven days or more following the 2nd BNT162b2 vaccine (“full effectiveness” cases), and 3) infections with no vaccination history (“unvaccinated controls”). Controls were matched to partial and full effectiveness cases, respectively, based on age (+/- 10 years), sex, district of residence, religion, and date of PCR sampling. A conditional logistic regression model was used to determine whether partially or fully vaccinated cases had higher odds of infection with a SARS-CoV-2 variant of concern relative to unvaccinated controls.

Summary of Main Findings

Of 3,491 infections detected among partially or fully vaccinated cases in the 4.7 million patients investigated, only 433 (12.4%) were able to be matched with a control and sequenced. Among the partial effectiveness cases (n = 247), 92% of SARS-CoV-2 infections were detected within 28 days of the 1st vaccine dose; among the full effectiveness cases (n = 149), one-third (31%) were detected 21+ days after the 2nd vaccine dose. The B.1.1.7 variant was responsible for ~90% of SARS-CoV-2 infections in vaccinated cases and unvaccinated controls; only 11 infections of the B.1.351 variant were detected in case-control pairs. No significant difference in B.1.1.7 infections was observed when comparing full effectiveness cases to unvaccinated controls, but for partial effectiveness cases the odds of B.1.1.7 infection was twice as high. The odds of B.1.351 infection was significantly higher among the full effectiveness cases relative to unvaccinated controls, but there was no increased odds in the partial effectiveness group. Importantly, all eight matched B.1.351 cases identified among the full effectiveness cases were detected within 7-13 days of the 2nd vaccine dose and none were detected after that.

Study Strengths

Investigators matched vaccinated individuals to unvaccinated individuals to control for factors (i.e., age, date of PCR test, district of residence) that could bias observed associations between the lineage of SARS-CoV-2 infections and immunization status.

Limitations

Because this study compares the distributions of strains between vaccinated and unvaccinated individuals, not rates of infection, effect estimates obtained from this study should not be interpreted as measures of vaccine effectiveness; this study cannot directly assess the vaccine effectiveness for any strain but can only assess the relative vaccine effectiveness of one strain versus another (i.e., an odds ratio of 2.0 could equate to vaccine efficacy of 90% relative to 95%). The low number of total B.1.351 and wildtype infections in both the full effectiveness analysis (n=9 and 14, respectively) and partial effectiveness analysis (n=2 and 64, respectively) may have underpowered analyses. SARS-CoV-2 vaccinations in Israel were offered first to older adults, which resulted in age-imbalanced vaccinated case group and unvaccinated control group; SARS-CoV-2 breakthrough cases in the full effectiveness group, therefore, could be partially attributed to older age, which could be associated with an attenuated immune response to vaccination.

Value added

This is among the first studies to assess the relative propensity for breakthrough infections between SARS-CoV-2 lineages among BNT162b2 vaccine recipients in a non-controlled, observational setting.

Our take —

In a case-control study, available as a preprint and thus not yet peer-reviewed, of 528 matched pairs of healthcare workers in Manaus, Brazil, the estimated effectiveness of at least one dose of CoronaVac vaccine (Sinovac Life Sciences) against any SARS-CoV-2 infection (regardless of symptoms) was 35.1%, and was 49.6% against symptomatic infection. While the study suggests at least one dose of CoronaVac vaccine is not significantly protective against SARS-CoV-2 infection, these findings are subject to numerous limitations, including inattention to differences between healthcare workers who received one or two doses of CoronaVac.

Study design

Case-Control

Study population and setting

Between January 19 and March 25, 2021, investigators identified healthcare workers in Manaus, Brazil, with RT-PCR-confirmed SARS-CoV-2 infections (both with and without symptoms), who were matched to non-infected healthcare workers on age group, neighborhood of residence, and specimen collection dates. Manaus experienced two distinct waves of SARS-CoV-2 transmission in March and November 2020, respectively, with the latter wave attributed to the emergence of the P.1 variant. The authors conducted a matched analysis to identify the odds of SARS-CoV-2 vaccination by CoronaVac, comparing infected healthcare workers to non-infected controls.

Summary of Main Findings

Of the 67,718 healthcare workers registered in Manaus, 393 and 135 case-control pairs of symptomatic and asymptomatic illness, respectively, were formed. Controlling for age, sex, race, healthcare cadre, frequency of healthcare interactions, and history of SARS-CoV-2 infection before January 19, 2021, the odds of symptomatic COVID-19 infection were 50% lower among healthcare workers who received at least one dose of the CoronaVac vaccine in the 14 days prior to specimen collection. Receiving at least one dose of CoronaVac at least 14 days before specimen collection did not significantly protect against symptomatic or asymptomatic SARS-CoV-2 infection. The estimated effectiveness of at least one dose against any SARS-CoV-2 infection (regardless of symptoms) was 35.1%, and was 49.6% against symptomatic infection.

Study Strengths

In the absence of longitudinal, prospective data on individuals, investigators created matched pairs of SARS-CoV-2 infections and non-infected controls, which can provide valid proxies for SARS-CoV-2 risk measures required for estimating vaccine effectiveness. To examine CoronaVac vaccine effectiveness in the context of P.1 transmission, investigators restricted the study period to 2021, when P.1 accounted for >75% of sequenced SARS-CoV-2 specimens in Manaus.

Limitations

Investigators did not enumerate or distinguish between individuals who had received one or two doses of CoronaVac, which could produce biased estimates of vaccine effectiveness. Additionally, healthcare workers were not matched on occupational cadre, nor was the setting in which participants worked (i.e., hospital intensive care unit, outpatient primary care center); because healthcare workers cadres may have differential SARS-CoV-2 exposure risks, inattention to occupational settings could further bias vaccine effectiveness estimates. Since recruitment into the study required availability of a SARS-CoV-2 test specimen, vaccine effectiveness measures could be inflated for healthcare workers with asymptomatic illness, as SARS-CoV-2 testing would likely be influenced by the presence of COVID-19 symptoms. Lastly, vaccine effectiveness was calculated over a two-month period (January to March, 2021), which limits inferences about the CoronaVac’s potential effectiveness over more prolonged periods (i.e., six months).

Value added

This is among the first studies to estimate the effectiveness of the CoronaVac vaccine in the context of P.1-dominant SARS-CoV-2 transmission.

Our take —

Obesity has been identified as a risk factor for severe COVID-19 since early in the pandemic, but associations in observational studies may be subject to confounding. This study tried to circumvent this issue by using Mendelian randomization, a study design that uses genetic variation to estimate causal associations between exposures and outcomes. The authors found no associations between 16 out of 17 cardiometabolic risk factors and the risks of severe COVID-19 using data from the COVID-19 Host Genetics Initiative, to which 22 cohorts had contributed. Only a higher body mass index (BMI) was associated with a greater risk of severe COVID-19, though this association was not present after adjusting for effects of obesity-related conditions. It is likely that the relationships among genetic variants, BMI, obesity-related conditions, and COVID-19 risk are sufficiently complex to warrant caution in interpreting these results.

Study design

Case-Control

Study population and setting

This study used Mendelian randomization to examine associations between 17 cardiometabolic traits (type 1 diabetes, type 2 diabetes, hemoglobin A1c, fasting glucose, fasting insulin, body mass index [BMI], waist–hip ratio, LDL cholesterol, HDL cholesterol, triglycerides, systolic blood pressure, diastolic blood pressure, creatinine-based estimated glomerular filtration rate [eGFR], chronic kidney disease, coronary artery disease, any stroke, and C-reactive protein) and two outcomes: susceptibility to COVID-19 and hospitalization for COVID-19. Mendelian randomization is an analytic technique designed to circumvent potential confounding, in which a genetic variant associated with an exposure is tested for association with an outcome. Genetic variants associated with the 17 cardiometabolic traits were taken from meta-analyses of genome-wide association studies (GWAS). Tests for association between genetic variants and outcomes were performed using data from the COVID-19 Host Genetics Initiative, to which 22 cohorts had contributed. Analyses were restricted to individuals of European ancestry. For COVID-19 susceptibility, those with a positive test for SARS-CoV-2 infection via PCR, a clinical diagnosis of COVID-19, or serological evidence for prior SARS-CoV-2 infection (n=14,134) were compared to population controls (n=1,284,876). For hospitalization, 6,406 hospitalized patients with laboratory-confirmed or clinically diagnosed COVID-19 were compared to 902,088 population controls. Controls were required to have no history of COVID-19 diagnosis or laboratory-confirmed SARS-CoV-2 infection. Two-sample Mendelian randomization was performed using inverse variance weighting with random effects, and odds ratios were estimated for each exposure-outcome relationship. To explore the degree to which the exclusion restriction may have been met (that the genetic variant can only affect the outcome through the exposure of interest), the authors performed sensitivity analyses using different methods (weighted median estimator, MR-Egger regression, mode-based estimation). To test for mediation and to account for possible pleiotropy (when one gene influences two or more unrelated phenotypic traits), the authors performed multivariable (pairwise) Mendelian randomization analyses of the BMI-outcome relationships by including genetic effects of coronary artery disease, stroke, chronic kidney disease, and type – diabetes one exposure variable at a time.

Summary of Main Findings

In Mendelian randomization analyses of the 17 cardiometabolic exposures and 2 outcomes, only the BMI-hospitalization association was statistically significant after adjustment for multiple tests. Authors repeated the analysis in UK Biobank data to obtain an interpretable effect estimate, finding that each unit increase of BMI (kg/m^2) was associated with 1.14 times the odds of COVID-19 hospitalization (95% CI: 1.07 to 1.21). Estimates using the weighted median estimator, MR-Egger regression, and mode-based estimation yielded results that were heterogeneous but in the same direction. In the pairwise multivariable Mendelian randomization analyses, BMI had no statistically significant associations with either outcome after adjustment for the effects of (any one of) coronary artery disease, stroke, chronic kidney disease, or type 2 diabetes, implying that any effect of BMI on the outcome may be mediated by these conditions.

Study Strengths

The authors performed analyses using several methods to test the robustness of their findings to alternate assumptions. The study was able to draw from a large amount of genotype and phenotype data linked to COVID-19 outcomes.

Limitations

Control groups were population-based, and thus may have differed from the source population giving rise to COVID-19 cases; this may have biased effect estimates. Population-based controls were required to have no prior positive SARS-CoV-2 test, and presumably included individuals who would have been hospitalized with COVID-19 had they become infected; a more appropriate counterfactual proxy for hospitalized COVID-19 cases would be individuals who were infected but not hospitalized. The study population included only those of European ancestry, limiting generalizability. Other characteristics of the cases and controls (e.g., age, sex, clinical characteristics) were not described in detail. Causal hypotheses underlying the multivariable analyses and their interpretation were unclear (e.g., postulated causal relationships among genetic variants, BMI, and other cardiometabolic risk factors). The genetic instruments were weak (they explained a small proportion of variation in the exposures), and pleiotropy (pathways from variant to outcome that do not involve the exposure) was likely. In general, results from Mendelian randomization are valid causal effects only when largely untestable assumptions about causal dynamics are met. Finally, the effect of BMI on the log odds of hospitalization was modeled as linear, though associations between BMI and other outcomes (e.g., mortality) are known to be nonlinear.

Value added

This study adds to the evidence base implicating obesity (as measured by BMI) as a causal risk factor for severe COVID-19.

Our take —

The clinical trial of the Pfizer/BioNTech vaccine showed that the vaccine is almost 95% effective in preventing COVID-19 1-2 weeks following administration of the second dose of the vaccine. With vaccine shortages and high SARS-CoV-2 infection rates world-wide, many have proposed postponing the second dose of the vaccine to expand the number of people getting their first dose to curb viral spread. This study shows that the Pfizer/BioNTech vaccine may be 85% effective in preventing symptomatic cases of COVID-19 15-28 days after following administration of first dose. Further studies are needed to determine the long-term protection after the first dose of the vaccine.

Study design

Case-Control, Prospective Cohort

Study population and setting

The study analyzed the SARS-CoV-2 infection rate among health care workers in The Sheba Medical Center, the largest hospital in Israel, after vaccination with Pfizer/BioNTech BNT162b2 vaccine. By the end of the study, among 9109 health care workers eligible for the vaccine, 7214 (79%) had received at least one dose and 6037 (66%) had received both doses of the vaccine. Most (91%) of the fully vaccinated workers received their second dose 21 or 22 days following the first dose. SARS-CoV-2 infection was detected through daily required symptom reporting and/or contact tracing followed by PCR lab diagnosis.

Summary of Main Findings

The study found that the vaccine reduced the infection rate (both symptomatic and asymptomatic) from 7.4/10,000 person-days in the unvaccinated group to 5.5/10,000 person-days (30% reduction) and 3/10,000 (75% reduction) on days 1-14 and 15-28 after administration of the first dose of the vaccine, respectively. The symptomatic cases were also reduced by 47% (2.8/10,000 person-days) and 85% (1.2/10,000 person-days) on days 1-14 and 15-28 after administration of the first dose of the vaccine, respectively, compared to the unvaccinated group (5/10,000 person-days).

Study Strengths

The study included 9109 health care workers with daily reporting of SARS-CoV-2 symptoms and contact tracing followed by same-day molecular diagnosis of the infection. This allowed for relatively high quality data. The high rate of SARS-CoV-2 infection in Israel at the time in which this study was conducted permitted detection of differences in infection rates among different groups tested in this study.

Limitations

One of the main limitations of this study is that it included health care workers only. This group has a higher exposure rate relative to the general population. The study also analyzed Pfize/BioNTech vaccine protection for up to 28 days after the first dose. Longer follow up to determine the protection level for one dose relative to two dases of the vaccine. For the effectiveness estimate of the window 15-28 days after vaccination, person-time from vaccinated individuals who had not been infected through 14 days were compared to all person-time contributed by unvaccinated individuals. It is possible that this could induce a selection bias that could overestimate effectiveness during this time window.

Value added

This study showed that Pfizer/BioNTech (BNT162b2) vaccine may provide up to 85% protection against symptomatic infection of COVID19 15-28 days after the first dose of the vaccine.

Our take —

This case series and case-control study set in Brazil examined risk factors for COVID-19 recurrence among 33 cases and 62 controls who were diagnosed and treated at the study hospital. The study found that the vast majority (>90%) of recurrent cases were healthcare workers. While symptom rates did not differ between first and second episode, cases were more likely to be hospitalized during their second episode; other studies have observed mixed results related to infection severity among reinfected cases, though often reinfections have been milder in other reports. The interval range between first and second episodes was 8 to 130 days (mean: 41 days). The major limitation was the challenge in determining if recurrence was actually a second new infection, or just symptoms of the same first infection. The study partially addressed this using next generation sequencing, providing evidence of new infection. It remains to be seen whether these secondary infections reflect new variants and potentially increased severity of disease associated with them.

Study design

Case Series, Case-Control

Study population and setting

The study objective was to determine risk factors for recurrent COVID-19. At the Federal University of Sergipe in Brazil, daily phone monitoring of confirmed COVID-19 cases from the Centro de Doencas Respiratorias (Center for Respiratory Diseases) was conducted. From this patient population, the study recruited 33 cases of recurrent, symptomatic, reverse transcription polymerase chain reaction (RT-PCR) positive infection with SARS-CoV-2. They defined recurrence as symptom recurrence in an individual having developed symptomatic COVID-19, medically isolated for 14 days, and then clinically recovering with at least 7 days without symptoms. They required cases to test positive by RT-PCR in both the first and second episode of COVID-19. They collected sociodemographic and clinical variables from the original 33 cases, and then randomly selected a control group of 62 people with only a single episode of the database. SARS-CoV-2 antibody testing was performed for a subset (51.5%, N=17) of recurrent cases, and the controls (50%, N=31). Next generation sequencing of viral genomes from 2 recurrent cases were also conducted.

Summary of Main Findings

The study found 33 recurrent COVID-19 cases, the majority of whom were female (78.8%, N=26) and were more likely to be healthcare workers (OR: 36.4, 95% CI: 9.7 – 137.2). Healthcare workers in the recurrent group were likely exposed in their work environment (97% reported possible work exposure), 48.5% of whom worked specifically in COVID-19 units. The average interval between recovery and second onset of symptoms was 41 days, with a range of 8 to 130 days. Blood type A+ had the highest prevalence (42%, N=14 cases), followed by O+ (30%, N=10). 45.5% (N=15) had comorbidities which placed them at higher risk for severe COVID-19, with obesity (N=10, 30.3%) being the most prevalent. For clinical symptoms, patients did not have any significant differences between number of symptoms in the first and later episode. Hospitalization was not required for any individuals in the first episode, but 12.1% of patients were hospitalized in the second episode, 2 required ICU admission, and 1 died from COVID-19 in their second episode. Genomic sequencing provided further evidence that these were true reinfections of different viral variants.

Study Strengths

The study had significant case data available for the recurrent cases, which are often understudied. Using next-generation sequencing, they were able to show that a likely second infection did in fact occur among the two participants tested. They also used a case definition that required clinically confirmed recovery before being eligible to be considered a recurrent case, which further added to the robust case definition and likely avoided some misclassification from cases who had a single, long infection as having recurrent infection.

Limitations

The major limitation was that they were not able to sequence all of the study participants to determine if each had a new infection. There may have been some misclassification of cases who had a single, long infection where symptoms may have abated for a short period, before returning. These individuals would therefore not reflect an actual second infection. Additionally, the study found that individuals were less likely to be hospitalized in the first episode compared to the second episode, however there may have been some survivorship bias due to individuals who were hospitalized and passed from COVID-19 no longer being at risk of second infection, and had they not died they may have gone on to become reinfected. This different survivorship biases their study towards more mild symptoms in the first episode. Finally, there were constraints on the power for the study, given the number of recurrent cases was only 33 and the confidence intervals were very large; adjusting for key confounders thus is difficult and controls were not matched to cases based on underlying characteristics.

Value added

This study has one of the largest sample sizes of recurrent infections included in research. Prior reinfection studies have suffered from very low sample sizes, given that reinfection is unlikely in the majority of COVID-19 patients.

Our take —

In this study during the early pandemic in the UK (prior to June 2020), researchers described 148 patients who had been discharged alive after hospitalization with severe COVID-19 and high troponin levels, at a median of 68 days post-discharge. Of these, 27% demonstrated evidence of myocarditis-pattern by late gadolinium enhancement (LGE) on cardiovascular magnetic resonance (CMR) imaging; a third of those with myocarditis-pattern also met the criteria for active myocardial inflammation, with no functional consequences. These prevalences are lower than in prior reports of patients with mild COVID illness. In 28% of the 148 patients, there were findings of ischemic heart disease (infarct or ischemia), among whom 66% had no prior history. This latter finding could suggest ischemic injury precipitated by the hemodynamic, prothrombotic, and proinflammatory stressors of critical COVID illness. Overall, the study findings highlight myocarditis and ischemic injury (but not diffuse fibrosis or global edema) in a significant proportion of patients with baseline high prevalence of cardiovascular comorbidities and recovered, severe COVID illness. However, there were no significant adverse functional consequences of the CMR injury patterns when compared to 40 controls and 40 other healthy volunteers. These results do not imply a substantially different prognosis to that of cardiac injury of non-COVID etiology nor do they confirm myocardial injury by or specific to COVID vs. other viral illnesses. The findings should also be interpreted with caution as the study had a relatively small sample size and was limited to contactable, non-contraindicated patients who consented to a post-discharge CMR scan.

Study design

Case-Control; Retrospective Cohort

Study population and setting

This case-control study assessed patterns of myocardial injury using multiparametric cardiovascular magnetic resonance (CMR) imaging in 148 patients who recovered from severe COVID-19 (e.g. all requiring hospital admission) and had troponin elevation during their hospitalization. Patients were recruited across six hospitals within the UK’s National Health Service network; CMR was performed at 3 sites. Case selection criteria included patients with a diagnosis of COVID-19 made by a positive reverse-transcriptase-polymerase chain reaction test for SARS-CoV-2, or positive for a triad of viral symptoms characteristic of COVID-19 illness, typical blood biomarkers, and findings of probable COVID-19 on chest radiography or by computed tomography. Eligible cases had to have an elevated high-sensitivity troponin level during admission, and were referred clinically for CMR because of the biomarker elevation. Discharges before June 20, 2020 were included. Case CMR scans were compared to research scans from a group of healthy volunteers (n=40) obtained prior to January 1, 2020 as well as a historical control group (n=40) obtained from stable outpatients prior to January 1, 2020 and matched for age, gender, diabetes, and hypertension but without suspicion for myocardial injury. Primary outcomes included presence, patterns, and extent of myocardial injury in recovered COVID patients compared to a control group and healthy volunteers.

Summary of Main Findings

Among case patients recovered from severe COVID-19 illness with elevated troponin, 30% (n=44) were female, average age was 64 years (+/- 12), and self-reported ethnicity was 50% white (n=75), 18% Afro-Caribbean (n=26), and 15% Asian (n=22). Case patients were imaged a median of 68 days following confirmed diagnosis (56 days from discharge). Cardiovascular risk factors were frequent among the cases (prior myocardial infarction in 7%, prior coronary revascularization in 12%, hypertension in 57%, diabetes in 34%, hypercholesterolemia in 46%, and smoking history in 24%). The control group was relatively well-matched, except for ethnicity and prior revascularization. Left ventricular (LV) function was normal in 89% with average case left ventricular ejection fraction (LVEF) at 67% +/- 11%, which was no different from matched controls (LVEF 67% +/- 9%, P=0.99) or healthy volunteers (LVEF 66% +/- 5%, P=0.55). In 54% (80/148) of case patients, there was evidence of late gadolinium enhancement (LGE), both ischemic and non-ischemic and/or ischemia on CMR (adenosine stress performed in a subset). Observed patterns of myocardial injury consisted of non-infarct myocarditis-like scar (26% [39/148]), ischemic heart related (infarction by LGE and/or ischemia) (22% [32/148]), non-ischemic non-specific scar (5% [7/148]), and dual ischemic and non-ischemic pathology (6% [9/148]). Myocardial infarction was found in 19% (28/148) of case patients, and inducible ischemia was found in 26% (20/76) of case patients undergoing stress perfusion. 66% (27/41) of patients with ischemic injury had no history of coronary artery disease. Among patients with non-ischemic, myocarditis-like injury, 30% (12/40) had active myocarditis with elevations in both T1 and T2, and 68% (27/40) had evidence of healed myocarditis, with only elevated native T1 values. The T1/T2 indices were not different in the remote regions between cases and matched controls, suggesting the absence of diffuse fibrosis and edema. For the LGE findings, there was a higher prevalence of the subepicardial LGE pattern among cases compared to controls, but no other differences in patterns with similar overall prevalence of any LGE (49% vs. 45% compared to 0% in the healthy volunteers). Peak troponin level was not predictive of myocarditis diagnosis. Extracardiac findings (pleural and pericardial effusions) were relatively infrequent (9% and 5%, respectively). Among those with a normal CMR scan, 51% underwent computed tomography pulmonary angiography during admission, of which 29% were diagnosed with pulmonary embolism, which could have contributed to the troponin elevation. However, among those with an abnormal CMR, the prevalence of pulmonary embolism was also high at 43% and hence, pulmonary embolus should not be assumed to be the only cause of troponin elevation. Right ventricular ejection fraction was lower and RV volumes higher among cases compared to controls (though still in normal range), perhaps attributable to the high prevalence of pulmonary emboli as well as the likely severe pulmonary involvement in these severely ill COVID patients.

Study Strengths

This multicenter case-control study examined a high-risk cohort with elevated troponin (most at risk for cardiac injury) for post-COVD sequelae. Patients were enrolled across 6 hospitals in the UK and included an ethnically diverse cohort of subjects. There were comparison CMR scans from COVID-negative participants, including those from matched controls and normal volunteers, the latter to establish normal values of T1 and T2 indices as well as phantom standardization of the mapping sequences.

Limitations

The total number of discharged patients with elevated troponins was not stated and hence, there could be selection bias in those electing to follow-up with post-discharge CMR. Inclusion of only discharged patients introduces survivorship bias; by only studying those with troponin elevation, this likely overestimates the prevalence of myocardial injury among the broader group of COVID patients. Troponin-negative COVID-19 patients were not included as a comparison group, reducing usefulness of control groups. There were no pre-hospitalization studies in the cases so it was unclear the extent to which the CMR findings are new. Post-contrast extracellular volume fraction was not assessed, which is generally more sensitive than native T1 for diffuse fibrosis. This study largely included the early pandemic period in the UK, a time with limited diagnostic resources and evolving therapies, which may have resulted in selection bias toward patients with the most severe and classic disease presentations. The extended period between discharge and CMR imaging may have contributed to underestimation of some disease presentations. As an observational study, confounding is likely to exist. There were no follow-up CMR scans to track temporal evolution of the CMR patterns.

Value added

This study is the largest reported CMR effort to date to comprehensively investigate myocardial injury in patients who have recovered from severe COVID-19 disease. It included appropriate COVID-negative controls and a quality assurance standardization protocol for the T1 and T2 mapping sequences.

Our take —

This genome-wide association study among people primarily of European ancestry compared over two thousand critically ill COVID-19 patients to controls from several population-based cohorts. The authors replicated an association between severe illness and a genetic variant on Chromosome 3 (at 3p21.31) previously observed in other studies, and identified novel associations between several other genes and severe COVID-19. A Mendelian randomization study provided some evidence that severe COVID-19 is linked with lower IFNAR2 expression (involved in the antiviral defense mechanism of type I interferon signaling, the suppression of which has been previously implicated in severe COVID-19) and higher TYK2 expression (which may promote inflammatory lung injury). The study used population-based controls with no prior positive SARS-CoV-2 test, which makes unmeasured confounding likely. The results, if replicated elsewhere using non-critically ill COVID-19 patients as controls, may lead to development of therapeutic approaches to stimulate interferon activity and suppress harmful inflammation.

Study design

Case-Control, Other

Study population and setting

This was a genome-wide association study (GWAS) comparing genetic variants from 2,244 critically ill COVID-19 patients from 208 UK intensive care units to variants from ancestry-matched controls (5:1) in UK Biobank, a large population-based cohort. Potential controls were excluded if they had any record of a positive test for SARS-CoV-2. Ancestry was inferred using principal components analysis with a population reference from the 1000 Genomes Project. GWAS was performed separately by ancestry group. Tests for association with COVID-19 status were performed via multivariable logistic regression with the covariates sex, age, residential deprivation score, and the first 10 genomic principal components. Replication was attempted using a meta-analysis of data from the COVID-19 Host Genetics Initiative (2,415 hospitalized COVID-19 cases; 477,741 population-based controls). Two-sample Mendelian randomization was employed to assess the causal effects of RNA expression of several genes on the risk of severe COVID-19. The authors additionally performed a transcriptome-wide association study (TWAS) testing for links between GWAS results and tissue-specific (lung and whole blood) gene expression.

Summary of Main Findings

There were 15 independent associations between genetic variations and COVID-19 case status that had genome-wide significance (p < 5 x 10−8); of these, 8 were validated in GWAS using additional sources of population-based controls and thus included in replication analysis. Five of the 8 associations were replicated with statistical significance; these were found in loci on chr3 at 3p21.31 (odds ratio (OR): 2.14, p= 4.77 x 10-30), chr12 near the OAS gene cluster (OR: 1.59, p 1.65 x 10^−8), near TYK2 on chr19 (OR: 1.40, p: 2.3 10^−8), in DPP9 on chr19 (OR: 1.36, p 3.98 x 10^−8), and on chr21 containing the gene IFNAR2 (OR: 1.28, p 4.99 x 10^−8). A list of target genes was assessed in Mendelian randomization for possible effects on severe COVID-19; low expression of IFNAR2 showed statistically significant association with severe disease. Transcriptome-wide Mendelian randomization on unselected genes yielded no significant associations after adjustment for multiple tests; the smallest p-value (0.00049) was observed for an association between higher expression of TYK2 and severe disease. Both the IFNAR2 and TYK2 associations were replicated with statistical significance in an external dataset. In TWAS results, five genes had genome-wide differences in expected expression between cases and controls: CCR2, CCR3, CXCR6, and MTA2B from lung tissue; and OAS3 from whole blood.

Study Strengths

The COVID-19 cases were drawn from an existing network of ICUs in the UK and thus represented a well-defined group of critically ill patients, which better represents the outcome of interest (severe disease) than other methods of case selection (e.g., all hospitalized cases). The authors attempted to replicate their results using multiple independent control groups.

Limitations

Control groups were drawn from population-based cohorts that differed from COVID-19 cases with respect to ancestry, demographic variables, and comorbidities; this may have introduced confounding of the variant-outcome associations. For example, if controls were less likely to have been exposed to SARS-CoV-2 than cases due to occupational or sociodemographic differences, effect estimates would be biased. This confounding is likely given the observed genetic correlations with educational attainment. Moreover, potential controls with any prior positive SARS-CoV-2 test were excluded, further confusing the contrast between cases and controls (i.e., cases had critical illness, whereas controls may not have ever been exposed to SARS-CoV-2). The study population of critically ill COVID-19 cases was not well described and its characteristics were not reported in detail, which limits interpretation of the observed associations. One of the assumptions of Mendelian randomization (that the genetic variant can only influence the outcome through the exposure of interest) appears to have been violated in the results for IFNAR2, which may have biased the effect estimate in an unknown direction. Finally, the study population was predominantly of European ancestry, limiting generalizability of results.

Value added

This GWAS confirms prior evidence for associations between genetic variants and severe disease, and adds novel evidence for associations that are amenable to replication in future analyses and may provide the basis for therapeutic interventions.

Our take —

In this study of 154 patients with asymptomatic or severe COVID-19 in India, individuals with severe COVID-19 (defined as ICU admission) had lower vitamin D levels and were more likely to have vitamin D deficiency. Those with vitamin D deficiency were more likely to die than those without deficiency. However, caution is warranted as these findings do not provide evidence that vitamin D supplementation can prevent severe COVID-19, results were not adjusted for any potential confounders (including age), and vitamin D levels were measured after illness onset. Further research regarding this relationship is warranted.

Study design

Case-Control, Prospective Cohort

Study population and setting

This observational study at the M.L.B. Medical College in Jhansi, India was designed to evaluate the association between vitamin D and COVID-19 disease severity. Individuals aged 30-60 years with PCR-confirmed COVID-19 starting were consecutively enrolled for 6 weeks starting on June 5, 2020. Patients were included if they were asymptomatic at admission and remained so until their discharge (Day 12) or if they required ICU admission due to severe COVID-19 disease, defined as clinical pneumonia with respiratory rate >30 or oxygen levels <90%, signs of multi-organ impairment, or laboratory evidence of coagulation abnormalities. Vitamin D levels were estimated from Serum 25 (OH)D, and levels <20 ng/mL were considered as vitamin D deficiency. Participants were followed until death or discharge alive from the hospital.

Summary of Main Findings

In total, 154 patients were included: 91 were asymptomatic (53% male; mean age: 42 years) and 63 had severe COVID-19 (67% male, mean age: 51 years). The mean concentration of 25 (OH)D was higher among asymptomatic cases (mean: 27.9 ng/mL, 32% with vitamin D deficiency), compared to those with severe disease (mean: 14.4 ng/mL, 97% with vitamin D deficiency). Serum inflammatory markers, such as IL-6, ferritin, and TNF-alpha, were also associated with vitamin D deficiency. Regardless of symptom severity at enrollment, those with vitamin D deficiency were more likely to die (19/90 [21%] vs. 2/64 [3%]).

Study Strengths

This was a prospective cohort study with completed follow-up of included participants.

Limitations

No adjusted analyses were presented, and confounding is likely due to differences in age and comorbidities between asymptomatic and severely ill COVID-19 patients (older age is a known risk factor for vitamin D deficiency). The timing of vitamin D measurement relative to SARS-CoV-2 infection is not reported. Temporality between vitamin D levels and COVID-19 severity cannot be established; it is unknown whether vitamin D deficiency is a cause or consequence of severe disease. The sample size was relatively small and did not include patients with moderate disease (symptoms but not requiring ICU admission), which limits generalizability and inhibits inferences about any potential dose-response relationship between vitamin D levels and COVID-19 severity.

Value added

This is one of the first studies examining the relationship between Vitamin D and COVID-19 disease severity.

Our take —

This preliminary study suggests that using artificial intelligence on cough recordings can be a low-cost, rapid method for COVID-19 screening, especially in resource-limited settings and as a supplement to other screening techniques such as symptom questionnaires and temperature checks. Further work needs to be done to validate the accuracy of the diagnostic algorithm for COVID-19 identification in the general population and whether it has value over other established low-cost screening methods.

Study design

Case-Control; Other

Study population and setting

This study included 5,320 participants from across the world who submitted forced-cough data between April to May 2020 on a website and agreed to participate in the study to train a machine learning model to detect COVID-19 through cough recordings. Using online submission of cough recordings from various browsers and devices, all COVID-19 positive cases collected (n = 2,660, determined by either testing, or physician or self-assessment) and the same number of randomly selected negative cases (determined negative by the same methods) were included. 80% of the total cohort were used in a training set, and the remaining 20% for validation. Sensitivity, specificity, and diagnostic accuracy were estimated.

Summary of Main Findings

The artificial intelligence-based COVID-19 screening tool achieved a reported sensitivity of 98.5% (eg. the model correctly identified 98.5% of positive cases as positive) and specificity of 94.2% (eg. correctly identified 94.2% of negative cases as negative), achieving an overall accuracy of 98.5%. In asymptomatic subjects, it achieved a reported sensitivity of 100% and specificity of 83.2%.

Study Strengths

Cough recordings were collected from various devices and platforms, across a variety of symptomatic groups, to generate the artificial intelligence-based model for identifying COVID-19. Advanced, robust convolutional neural networks featuring modules previously shown to be successful in diagnosing Alzheimer’s disease from audio voice recordings were used to train the model which may yield to stronger diagnostic ability.

Limitations

Participant recruitment was based on self-selection and online-based volunteering which is prone to selection bias and could potentially limit the value of this artificial intelligence tool in the general population. Additionally, this study did not report the performance of the AI to detect COVID-19 within specific subject (e.g. racial, gender, age) and recording device subgroups, which could lead to variability in the value of this tool in different populations. Lastly, the bulk of COVID-19 “cases” used to train the model and evaluate its accuracy were from self-assessment (59%) or doctor assessment (28%), versus results from (undefined) “official” tests (13%), so it is unknown what proportion of the cohort’s COVID-19 status were correctly assigned. Validation of this artificial intelligence-based COVID-19 detection tool in the general population and on verified COVID-19 cases and non-cases is necessary to further gauge its value.

Value added

This was a preliminary study that developed and evaluated the performance of an artificial intelligence-based model using recorded coughs to identify COVID-19 cases. The reported diagnostic accuracy was high (98.5%) with an 100% asymptomatic detection rate. This study proposes an alternative low-cost and rapid screening method for evaluating COVID-19 using artificial intelligence.

Our take —

Variations in immune response are suspected to underlie some COVID-19 disease outcomes. In this important study, about 10% of patients with severe COVID-19 pneumonia had neutralizing IgG auto-antibodies to type I interferons, which were not present in any patients with mild SARS-CoV-2 infection. Production of auto-antibodies inhibited the ability of type I interferons to fight SARS-CoV-2 infection in vitro, and appeared to precede disease in patients. Individuals producing auto-antibodies were nearly all (94%) male. These results suggest a mechanism contributing to severe disease in a sizable minority of patients, which may be X-chromosome-linked, and they may lead to screening and therapeutic implications if replicated elsewhere.

Study design

Case series, case-control

Study population and setting

This study included 987 patients hospitalized with severe “life-threatening COVID-19 pneumonia” (defined as pneumonia with critical disease involving any organ system, requiring ICU admission) from multiple cohorts in several countries, along with 663 patients with SARS-CoV-2 infection with mild or no symptoms, and 1,127 healthy controls. Plasma and serum samples were obtained from participants, and these samples were tested for the presence of IgG auto-antibodies (Ab) against the type I interferons (IFNs) IFN-ɑ2 and IFN-ω. The authors tested whether these auto-Ab were neutralizing in vitro using plasma from patients with auto-Ab and healthy controls. A subset of samples from patients with auto-Ab to IFN-ɑ2 (n=22) were tested for the presence of auto-Ab to all 15 type I IFNs. Another subset of samples from patients with auto-Ab to IFN-ɑ2 (n=8) were tested to determine if they impaired the ability of IFN-ɑ2 to block SARS-CoV-2 infection in vitro. Finally, plasma concentrations of all 15 type I IFNs were measured among patients who had auto-Ab to these type I IFNs (n=41).

Summary of Main Findings

Among the 987 patients with severe COVID-19 pneumonia, 135 (13.7%) had IgG auto-Ab to IFN-ɑ2 and/or IFN-ω; these were also present in the two available samples from patients prior to SARS-CoV-2 infection. The prevalence of these auto-Ab against type I IFNs was estimated to be 0.33% (95% CI: 0.015% to 0.67%) in the healthy controls. In vitro, these antibodies were neutralizing for at least one type I IFN in 101 patients (10.1%), 95 (94%) of whom were male. All of the tested patients with auto-Ab to IFN-ɑ2 also had auto-Ab to all 13 type I INFs. The 8 plasma samples tested from patients with auto-Ab to IFN-ɑ2 were able to block the ability of IFN-ɑ2 to protect against infection from SARS-CoV-2 in vitro. Among the 41 patients tested who had auto-Ab to all 13 type I IFNs, 40 had no detectable levels of any type I INFs, and one had low levels. Among individuals with neutralizing auto-Ab against type I INFs, there was a predominance of males (94%), which was higher than the proportion of men among severe COVID-19 patients without these Ab (75%) and among those with mild or no symptoms (28%). One of the 7 women with auto-Ab to type I INFs had a condition in which cells activate only one X chromosome, providing evidence for X chromosome linkage.

Study Strengths

This study considered evidence at multiple levels (prevalence in cases and controls, neutralizing effect in vitro, effect on ability of IFN-ɑ2 to block SARS-CoV-2 infection in vitro, plasma type I INF concentrations). The study population was large and included from diverse ethnicities and nationalities.

Limitations

Only two samples were available from patients before SARS-CoV-2 infection, which limits the ability to conclude with certainty that auto-Ab production was not a consequence of severe disease. Demographic and clinical characteristics of patients and controls were not well described. Methods for selecting asymptomatic or paucisymptomatic controls were unclear.

Value added

This, along with an accompanying study in the same journal (Zhang et al.), is an important and novel piece of evidence; it is among the first to identify underlying variations in immune response that may cause severe COVID-19.