Sex-bias in COVID-19: a meta-analysis and review of sex differences in disease and immunity

A striking anecdotal feature of the Coronavirus disease 2019 (COVID-19) outbreak is the difference in morbidity and mortality between the sexes. Here, we present a meta-analysis of 206, 128 reported cases to demonstrate that whilst there is no difference in the proportion of males and females with conrmed COVID-19, male patients have more than double the odds of requiring intensive treatment unit admission (OR 2.5) and higher odds of death (OR 1.60) when compared to females. We review data revealing how previous Coronavirus outbreaks have demonstrated a similar pattern. Important differences in the immune response to infection exist between sexes, which are likely to contribute to this observation. In this review, we discuss these differences highlighting that females have a more robust innate antiviral response and a better adaptive immune response to infection. An appreciation of how sex is inuencing COVID-19 outcomes will have important implications for clinical management and mitigation strategies for this disease.


Introduction
Anecdotal bias towards male sex has been observed globally in Coronavirus disease 2019 (COVID-19) 1 .
Sex is a traditionally under-appreciated biological variable in the immune response to infection, with emerging data highlighting its importance. To address whether the anecdotal bias reported is validated by statistical analysis, we have collected currently available case data and present a meta-analysis to investigate sex as a risk factor for COVID-19 infection, morbidity and mortality. To place this data in context, we have also provided a brief review of the role of sex in previous Coronavirus outbreaks, and more generally in other infections. We also provide a brief up to date overview of the literature that describes the difference in the male and female immune systems, which lead to females mounting a more robust innate and adaptive immune response to viral antigens.

Methods
Search strategy and selection criteria A google search was performed by multiple researchers working remotely for reports on COVID-19 cases that included sex as a variable (Search terms: COVID-19/case/ sex/ country/ data/death/ICU/ITU).
Reports were translated using Google translate if they were not in English. Speci cally, reports were included if they contained sex as a variable in data describing case number, intensive treatment unit (ITU) admission or mortality. Data were available at the level of summary data for each report but not at the level of covariates for all individuals within a study. Consequently, covariates such as lifestyle and comorbidities could not be controlled for. Data are available in Supplementary Table 1.

Data analysis
Meta-analysis was performed to estimate an overall proportion of male infected cases with 95% con dence intervals (CI). For this analysis, the classic inverse variance method for estimation of single proportions and standard errors was used. For comparison, estimated proportions were identical when the generalised linear mixed model (GLMM) method was used, indicating these different methods give the same overall result for these data. This meta-analysis tested the null hypothesis that the proportion of male infected cases was 0.5. Meta-analyses were also performed to estimate odds ratios (ORs) with 95% CI associated with male sex for ITU admission and death based on pooled average effect measures that were weighted according to the size and precision of each report. Both of these meta-analyses tested the null hypothesis that the estimated OR was 1. Fixed and random effects models were estimated. For all analyses, where parameters and p-values differ between xed and random effects models, the estimates from the random effects model are preferred as these account for variance between reports. Reports that did not contain the data required to calculate ORs were automatically excluded from meta-analyses.
Meta-analyses were performed using R version 3.6.1 and the "meta" package version 4.11-0.

Role of the funding source
The funders of this study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had nal responsibility for the decision to submit for publication.

Results
Forty two reports were found from across the world, from 01.01.2020 up until 30.3.2020 (Fig. 1). Reports were excluded if they did not report the total number of infections by sex (1 ITU case series and 1 mortality case series). One report was excluded as it contained less than 5 cases. Thirty nine reports remained. There were multiple reports originating from China which were carefully examined for duplication. For the analysis of case numbers by sex, 12 reports were excluded due to possible duplication. Of the 39, 6 reports included ITU admission by sex, 1 of these was excluded for possible duplication. Of the 39, 15 reports included mortality by sex, and 3 of these were excluded for possible duplication. Therefore, altogether 29 reports were included representing 206,128 cases 2-30 . The proportion of male cases with COVID-19 in these reports was only slightly over half at 0.52 (95% CI = 0.52,0.53, p = 2.3e-97 for xed effects model; 95% CI = 0.50,0.53, p = 0.12 for random effects model) demonstrating that males and females have similar numbers of infections (Fig. 2). Male sex associated with an increased risk of ITU admission (OR = 2.50; 95% CI = 2.25, 2.78; p = 3.8e-64 and 7.3e-64 for xed and random effects models, respectively; n = 43,075) (Fig. 3). Male sex also associated with an increased risk of mortality (OR = 1.62, 95% CI = 1.54, 1.71, p = 5.5e-77 for xed effect model; OR = 1.60, 95% CI = 1.41, 1.82, p = 7.4e-13 for random effects model; n = 170,983) (Fig. 4).

Discussion
These data show no difference in the proportion of COVID-19 infected cases between sexes, but highlight that male sex associates signi cantly with morbidity (ITU admission OR 2.50) and mortality (OR 1.60). Due to the nature of these rapidly acquired data, the age and comorbidity for individual cases are not available. This limits our ability to accurately predict the role of sex in disease without adjusting for these factors, as certain comorbidities might associate with sex. Once more data become available, future studies will be able to adjust for additional factors using techniques such as mediation analysis. In addition, all reports did not use the same COVID-19 tests (with differing sensitivity and speci city) and different population groups were included in each report (e.g. hospital admissions vs community cases).
Despite these limitations, these data highlight an important trend in the epidemiology of COVID-19, with male sex acting as a risk factor for severe disease.
Gendered differences in social activity and behaviours could contribute to the sex difference seen in COVID-19 disease severity. In China, 50.5% of men versus only 2.1% of women smoke 31 , although smoking has not yet emerged as a clear risk factor for severe disease 32 . Early data from the COVID-19 outbreak in China indicate that hypertension and diabetes are the most common comorbidities in hospitalised patients 33 . Globally from age 45-54, it is estimated that 33% of women compared to 36% of men are hypertensive, with this ratio reversing in the elderly (> 75), where it is estimated that 81% of women and 73% of men are hypertensive 34 . Similarly in diabetes, it is estimated that globally 9.0% of women and 9.6% of men are diabetic, and this ratio tends to increase and reverse in the elderly 35 . Men are less likely to wash their hands with soap after entering a restroom 36 and in many cultures men may be more likely to leave the house and enter crowded areas. Our data however shows no difference in the numbers of infected cases between the sexes, so sex difference in hygiene behaviour is unlikely to explain the sex difference in disease severity. There may be non-immunological biological differences between sexes that play a role. For example, COVID-19 most likely binds to angiotensin converting enzyme 2 (ACE2) receptors facilitating viral entry and human to human transmission 37 . Pre-print studies are con icting as to whether ACE2 expression in lung tissue is different between sexes 38,39 . Therefore, these factors are unlikely to fully explain the stark sex inequality in morbidity and mortality associated with COVID-19. Fundamental differences in the immune response to infection between sexes are more likely to contribute to this phenomenon. Below, we review literature pertaining to the sex differences seen in previous Coronavirus outbreaks, other infections and the known sex differences in the immune system.

SEX DIFFERENCE IN PREVIOUS CORONAVIRUS OUTBREAKS
-HUMANS: Severe Acute Respiratory Syndrome Coronavirus 1 (SARS-CoV-1) showed a similar sex discrepancy in the 2003 epidemic. In Hong Kong, the case fatality rate was 13.2 % for females (95 % CI = 11.1, 15.3) and 21.9 % for males (95% CI = 19.0, 24.8), with an age-adjusted relative mortality risk ratio of 1.62 (95 % CI = 1.21, 2.16) for males 40 . In a study in Singapore during the same outbreak, male sex associated with an odds ratio of 3.10 (95 % CI = 1.64, 5.87; p = <0.001) for ITU admission or death 41  -ANIMALS: This sex bias is also observed in animal models of SARS infections. In a mouse model of SARS-CoV-1 infection, female mice had a lower case fatality ratio, and less lung in ammation and oedema than males. Moreover, ovarectomy and treatment with the oestradiol antagonist, ICI 182, 780, diminished the sex advantage of female mice, implicating oestradiol in this sex difference 43 . Studies in animals that are known vectors for human disease have not shown the same sex bias that is observed in humans. Seropositivity in these reservoir animals, which act as hosts for viral pathogens but generally do not display symptoms, has been investigated with sampling studies. Studies in wild bats, known reservoirs for pathogenic coronaviruses, have shown either similar seropositivity in males and females 44 , or higher seropositivity in female animals 45 . Similar studies in dromedary camels, the main reservoir host for MERS-CoV, showed higher seropositivity in females 46,47 . A possible explanation for the higher rate of seropositivity in females is the higher risk of exposure: the close proximity of female dromedaries to their highly susceptible calves puts them at repeated risk of infection 47 , while pregnant bats collect to stay warm during roosting season 45 .

SEX DIFFERENCES IN OTHER INFECTIONS
Sex differences in response to infection occur at all ages with a generally higher burden of bacterial, viral, fungal and parasitic infections in human males. New-born males are more likely to die from infection than females 48 and male children have higher rates of parasitic infections 49 . Tuberculosis and hepatitis B infection are more common in males, who are also more likely to die of sepsis than females 50,51 . An exception to this male predominance in infectious disease is urinary tract infections, where adult females are 40 times more likely than males to develop disease 52 . Females with human immunode ciency virus (HIV) infection have less circulating viral RNA, but are 1.6 times more likely to progress to advanced immune de ciency syndrome (AIDS) at the same viral load as men 53 . Women are more susceptible to infections of the upper respiratory tract such as tonsillitis and sinusitis while men are more susceptible to lower respiratory tract infections i.e. community-acquired pneumonia 54 . Notably, the female sex bias in infection is mainly observed after puberty and before menopause, suggesting that sex hormones play an important role in this phenomenon 52 .

SEX DIFFERENCES IN INFLUENZA
Unlike COVID-19, mortality rates during previous in uenza pandemics have typically been higher in females, despite a higher prevalence of infection in men. Studies from Japan during the most recent H1N1 pandemic found that during reproductive years morbidity rates were higher in females, but outcomes were worse for males under 20 and over 80 55 . Pregnancy is an established risk factor for in uenza morbidity and mortality: in US data from the 2009 In uenza A (H1N1) pandemic, pregnant women represented 5% of all deaths despite comprising less than 1% of the population 56 . From the limited data about COVID-19 in pregnancy to date, there does not seem to be the same association with morbidity that is seen in in uenza, but this may change as more information becomes available 57 . Data from in uenza-infected animal models demonstrate no difference in viral load between males and females, but greater pro-in ammatory cytokine and chemokine production in females, suggesting that host-mediated in ammatory responses contribute to the disparity in morbidity between the sexes 58 .
Analysis of the effect of sex hormones versus the sex chromosome complementon the response to in uenza infection demonstrates that sex steroids are likely to be driving the differences observed 58 .
The stronger female response to in uenza antigens that may underlie more severe disease is mirrored in the female response to vaccination. Women consistently report more severe local and systemic side effects and produce higher antibody titres in response to seasonal u vaccinations 59,60 than men. Females achieved equivalent protective antibody titres to males at half the dose of inactivated in uenza vaccine 61 . After in uenza vaccination, female B cells produced more antigen-speci c IgG, mediated by sex-based differences in gene expression within B cells 62 .

SEX DIFFERENCES IN THE IMMUNE SYSTEM
Despite being a previously underappreciated biological variable, it is now well established that males and females mount different immune responses to infection. Whilst some differences are seen in the immune response from birth, certain differences are only observed after sexual maturity. This suggests that both sex chromosome complement as well as sex hormones in uence the immune response. In very general terms, females tend to be skewed towards a more robust immune response towards pathogens with relatively decreased self-tolerance, whereas males tend to have better self-tolerance, but a less robust response to pathogens. This contributes to the clinical phenotype of females being relatively protected against infection and malignancy when compared to males, but more prone to developing autoimmune diseases.
SEX CHROMOSOMES AND SEX HORMONES: There is an over-expression of genes with an immune function on the X chromosome 63 , as evidenced by the existence of many X-linked immunode ciency disorders 64 . There is emerging evidence that there may be variable inactivation and regulation of the inactive X chromosome in immune cells with subsequent bi-allelic expression of X-encoded immune genes in females 65 66 . Many immune cells express oestrogen receptors alpha and beta 67 . The effect of oestradiol seems to be dose-dependent, with low doses corresponding to a T helper 1 (Th1)-type response and cell-mediated immunity, and higher doses (such as in pregnancy) corresponding to T helper 2 (Th2)type responses and humoral immunity 50,68 . Oestradiol modulates CD4 T cells and CD8 T cells 69 , promotes T regulatory (Treg) cell expansion in vitro and in vivo 70 and decreases T helper 17 (Th17)/interleukin-17 (IL-17) production 71 . Oestradiol is associated with increased antibody production, somatic hyper-mutation and class switching 72 , abundance of neutrophils 73 , and monocyte/macrophage cytokine production 74 . In human T cells, almost half of the activated genes have an oestrogen response element in their promoter region 75 . Testosterone is generally thought to dampen the immune response; a lack of testosterone associates with increased in ammatory cytokines, antibody titres, CD4/CD8 ratios and natural killer cells along with a decrease in Treg cells 50,76 .
INNATE IMMUNITY: In terms of innate immunity, limited data suggest sex-based differences in the expression and response of pattern-recognition receptors on various immune cells. For example, both neutrophils from human males and peritoneal macrophages from male mice express higher levels of tolllike receptor 4 (TLR4) 77,78 and produce higher levels of tumour necrosis factor alpha (TNFα) after lipopolysaccharide (LPS) stimulation than females 79 .
There are important sex differences in the innate antiviral response that may be relevant to the sex discrepancy seen in COVID-19. There is a large body of evidence showing that females have a more robust production of type 1 interferon (IFN) upon sensing of viral RNA via toll-like receptor 7 (TLR7) [80][81][82][83][84][85] . When exposed to HIV-derived RNA, plasmacytoid dendritic cells from females produce more type 1 IFN than males 86 . This phenomenon has been shown to associate with sex hormone concentration but also the number of X chromosomes present 81 83 .
ADAPTIVE IMMUNITY: The adaptive immune system mirrors the trend of females being skewed towards a more robust immune response, but poorer self-tolerance, and males having improved tolerance, but a less robust response to infections. Autoimmune regulator (AIRE) gene expression is decreased in female thymic tissue compared to males and females have improved thymic function when compared to males throughout life [87][88][89] . At all ages, and even during HIV infection, females have more CD4+ T cells than males [90][91][92][93][94][95] . Female T cells have more robust cytotoxic activity and upregulation of in ammatory genes than males 75 , while males have more Tregs than females 96 . Females have more B cells and produce more immunoglobulin than males 90,97 .

SEX DIFFERENCE IN IMMUNE-AGING
There is a marked association between morbidity/mortality and advanced age in COVID-19. Although age-associated changes in immunity are beyond the scope of this review and are available elsewhere 98 , it is of note that age-associated changes in the immune system are also different between sexes. There is a male-speci c, age-associated decline in B cells and a trend towards accelerated immune ageing in males 99,100 which may further add to the sex difference in disease phenotype.

Conclusion
These data demonstrate that although there is no sex difference in the proportion of people infected with COVID-19, men are at a signi cantly higher risk of severe disease and death than women. Previous reports describe fundamental differences between sexes in the immune response to infection. These include a more robust antiviral innate interferon response and increased adaptive immunity towards viral antigens in females. In people infected with COVID-19, these differences are likely to lead to more effective viral control in females, which may contribute to the relatively lower risk of developing severe disease. Although further studies are needed, these data have implications for the clinical management of COVID-19 and highlight the importance of considering sex as an important variable in fundamental and clinical research. Figure 1 Study selection. 42 reports were found. 2 reports were excluded as they did not report the total number of infections by sex (1 ITU admission only case series and 1 mortality case series)and 1 report was excluded as it contained less than 5 cases. Of the 39 remaining reports; for the analysis of case numbers by sex, 12 reports were excluded due to possible duplication; of the 6 reports that reported ITU admission by sex, 1 was excluded for possible duplication; of the 15 reports that included mortality by sex, 3 were excluded for possible duplication. There is no obvious sex difference in the proportion of people infected with COVID-19. Table and forest plot shows number of infected males and total number of infected COVID-19 cases in 27 studies. The random effects model, which accounts for variance across reports, indicated no statistical difference (0.52 95% Con dence Interval (CI)=0.50,0.53, p=0.12).

Figure 3
Male sex is associated with a signi cantly increased risk for ITU admission within COVID-19 patients.  Male sex is associated with a signi cantly increased risk for mortality within COVID-19 patients. Table  and forest plot shows number of deaths and total number of infected COVID-19 cases for each sex. Odds Ratio (OR)=1.60; 95% Con dence Interval (CI)=1.41, 1.82; p=7.4e-13 for random effects model, n=12 reports with complete data on COVID-19 cases and mortality in males and females.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. FinalTableforSubmission06042020.docx