Analysis of sex-specific risk factors and clinical outcomes in COVID-19

Background Sex has consistently been shown to affect COVID-19 mortality, but it remains unclear how each sex’s clinical outcome may be distinctively shaped by risk factors. Methods We studied a primary cohort of 4930 patients hospitalized with COVID-19 in a single healthcare system in New York City from the start of the pandemic till August 5, 2020, and a validation cohort of 1645 patients hospitalized with COVID-19 in the same healthcare system from August 5, 2020, to January 13, 2021. Results Here we show that male sex was independently associated with in-hospital mortality, intubation, and ICU care after adjusting for demographics and comorbidities. Using interaction analysis and sex-stratified models, we found that hypoxia interacted with sex to preferentially increase women’s mortality risk while obesity interacted with sex to preferentially increase women’s risk of intubation and intensive care in our primary cohort. In the validation cohort, we observed that male sex remained an independent risk factor for mortality, but sex-specific interactions were not replicated. Conclusions We conducted a comprehensive sex-stratified analysis of a large cohort of hospitalized COVID-19 patients, highlighting clinical factors that may contribute to sex differences in the outcome of COVID-19.

R eports of the COVID-19 pandemic from around the world have described more severe disease and worse outcomes among men [1][2][3][4][5][6][7] . Men with COVID-19 appear to preferentially require hospitalization and intensive care 2,8,9 . For example, one large Italian case series reported that 82% of COVID-19 patients requiring intensive care were men 10 . Case fatality rates are also higher among men; a nationwide analysis from China reported a case fatality rate of 2.8% for men, compared to 1.7% for women 4 . Male sex has been identified as a risk factor for mortality in several studies of hospitalized COVID-19 patients [11][12][13][14][15] .
These observations have been variously attributed to underlying comorbidities among men, hormonal factors, or immune differences between men and women [16][17][18][19] . Some comorbidities associated with worse COVID-19 outcomes may be more common among men, though published studies have not provided sex-disaggregated data 2,20,21 . An Italian study observed that among prostate cancer patients, those on androgen deprivation therapy had better outcomes than those who were not, suggesting a sex hormonal contribution to COVID-19 mortality 22 . In addition, there are sex-specific differences in expression of the ACE2 and TMPRSS2 proteins, which facilitate the entry of the SARS-CoV-2 virion into cells 23,24 . Understanding how risk factors of severe disease differ between sexes can improve clinical risk assessment and shed new biological insights into disease etiology.
Given the sex difference in COVID-19 outcomes, we hypothesized that clinical risk factors for mortality might show sexspecific effects, which remain largely uncharacterized. Although multiple prior studies have included multivariable regression models adjusting for sex, they have not explored the possibility of interactions between sex and other predictors, nor have they examined differences in clinical course between men and women admitted with COVID- 19. To fill this knowledge gap, we conduct a sex-specific analysis of clinical data from 4930 COVID-19 patients hospitalized in New York City from the start of the pandemic to August 5, 2020. We confirm the association between male sex and adverse COVID-19 outcomes in this primary cohort and in a validation cohort of 1645 COVID-19 patients hospitalized from August 5, 2020, to January 13, 2021. We also explore sex-specific risk factors of inhospital mortality, intubation, and intensive care.

Methods
Study setting. Mount Sinai Health System is a large hospital system in the New York City area, comprising 8 hospitals and more than 410 ambulatory practice locations. Our analysis focused on patients who were admitted to five hospitals:  Sinai facility with any of the  following: a COVID-19 related encounter diagnosis, a COVID-19  related visit type, a SARS-CoV-2 lab order, a SARS-CoV-2 lab  result, or a SARS-CoV-2 lab test result from the New York State  Department of Health's Wadsworth laboratory up to August 5,  2020. Of these, 75,996 patients had at least one COVID-19 test and 12,347 tested positive. COVID-19 positivity was defined as a positive or presumptive positive result from a nucleic acid-based test to detect SARS-CoV-2 in nasopharyngeal or oropharyngeal swab specimens.
We limited our analysis to 4930 adult patients (at least 18 years old) who were admitted to the hospital and had a positive or presumptive positive SARS-CoV-2 test within 2 days of admission. Race and ethnicity were self-reported. Comorbidity status was assessed cross-sectionally using each patient's problem list in the electronic medical record as of their hospital encounter. The problem list consists of diagnoses linked to International Classification of Disease codes; it is updated by medical providers during clinical encounters (inpatient and outpatient) and reflects ongoing medical issues. Obesity was based on body mass index (BMI) ≥ 30 kg/m 2 or a pre-existing diagnosis in the medical record if BMI was not available.
The dataset captured inpatient medication exposures for a preselected list of drugs, including COVID-19 therapeutics such as remdesivir, steroids, and dexamethasone. Of note, medications administered in the context of clinical trials were not captured in this dataset.
Initial vital signs were the first vital signs documented for the encounter. Fever was defined as temperature ≥100.4°F; tachycardia was defined as heart rate >100 beats per minute; tachypnea was defined as respiratory rate >25 breaths per minute; hypotension was defined as systolic blood pressure <90 mmHg or mean arterial pressure <65 mmHg; hypoxia was defined as oxygen saturation <92%.
We defined initial labs as the first lab value within 24 h of the start of the encounter. A subset of patients had baseline serum interleukin-1-beta (IL-1B), interleukin-6 (IL-6), interleukin-8 (IL-8), and tumor necrosis factor-alpha (TNF-a) values obtained as part of a study, which enrolled COVID-19 patients hospitalized between March 21 and April 28, 2020 25,26 .
SARS-CoV-2 viral loads from nasopharyngeal swabs were measured in a randomly selected cohort of 1146 MSHS patients hospitalized between March 13 and May 4, 2020, using quantitative RT-PCR 27 . Direct overlap with our de-identified dataset was not feasible, but this random sample was drawn from the same population as our cohort and is likely representative.
Outcomes. The primary outcome was death from any cause during admission. The primary outcome was known for 99.5% of patients; only 23 patients had missing data. Secondary outcomes were mechanical ventilation during admission and intensive care unit (ICU)-level care (defined as admission to an ICU or mechanical ventilation) during admission.
Logistic regression analysis. Univariable and multivariable logistic regression were used to identify factors associated with each outcome. To identify sex-specific effects of risk factors, subgroup analysis was performed using a multivariable model including an interaction term between sex and the subgroup variable. This was performed for each subgroup variable to identify significant sex interactions. We further used sex-stratified multivariable models to estimate the magnitude of effect of each covariate within each sex cohort.
Predictors analyzed included demographic factors, comorbidity status, initial vital signs, treatment facility location (Manhattan vs. Brooklyn/Queens), and time period (prior to April 13, April 14 to June 2, June 2 to August 5). There was minimal clustering of the primary outcome by treatment site (ICC (ρ) = 0.021). Covariates were chosen a priori based on prior reports and low missingness. The covariates included in the final models were age (as a categorical variable), sex, race/ethnicity, treatment location, treatment time period, baseline hypoxia, and the following comorbidities: hypertension, diabetes, coronary artery disease, heart failure, atrial fibrillation, chronic kidney disease, chronic obstructive pulmonary disease (COPD) or asthma, obesity, cancer, chronic liver disease. We report the odds ratios derived from the coefficients of each model, along with the Wald-type confidence interval and p values.
Statistics. Patient characteristics and baseline vitals and labs were described using medians and ranges for continuous variables, and proportions for categorical variables. Continuous variables were compared using the Wilcoxon rank-sum test, and categorical variables were compared using Fisher's exact test or the chisquared test, as appropriate. All statistical analyses and data visualizations were carried out using R 4.0.0 (The R foundation, Vienna, Austria), along with the tidyverse, ggpubr, forestplot, and Hmisc packages. Statistical significance was defined as a twotailed p < 0.05 unless otherwise specified.
Study approval. The MSDW project (GCO# 12-0361) was reviewed and approved by the Institutional Review Board (IRB) of the Mount Sinai School of Medicine. The MSDW serves as a repository for data from the electronic medical record which can be used for research. The IRB has determined that the MSDW research involves no greater than minimal risk and approved the waiver for informed consent. A de-identified COVID-19 dataset was generated by the MSDW team for use by the Mount Sinai research community. The present study used this de-identified dataset for the rest of the analyses.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Results
The study cohort comprised 4930 COVID-19 patients hospitalized within the Mount Sinai Health System in New York City, of whom 2757 (55.9%) were men and 2173 (44.1%) were women. The majority (65.4%) of patients were treated in Manhattan facilities, which had larger bed capacities than facilities in Brooklyn or Queens. Most of the patients in the cohort (65.3%) were admitted before April 13, 2020, which corresponded to the peak of COVID-19 admissions within the hospital system.
A subset of patients had IL-1B, IL-6, IL-8, and TNF-a measured from serum at baseline. Of these cytokines, only IL-6 was significantly different between men and women (78.7 vs. 57.9 pg/mL, Wilcoxon p = 7.2 × 10 −5 ). Additional laboratory values are presented in Supplemental Table 1.
There were 1198 in-hospital deaths (24.3%). The rate of inhospital mortality was 25.2% in men and 23.1% in women (Fisher's exact p = 0.09). Among the patients who died, the median time from admission to death was 7.1 days (IQR 3.2-13). There was no significant difference in the median length of time from admission to death between men (7.3 days, IQR 3.3-12.4) and women (6.2 days, IQR 2.9-13.2; Wilcoxon p = 0.24). Among the patients who did not die, the median length of admission was 6 days (IQR 3-11) and men spent longer in the hospital (median 6 There was no significant difference in the proportion of men and women developing acute kidney injury (7% vs 6%, Fisher's exact p = 0.2) or venous thromboembolism (1% vs. 0.7%, Fisher's exact p = 0.3) during their hospitalizations.
Sex-specific predictors of hospitalization outcomes. In multivariable logistic regression models adjusting for demographics, comorbidities, admission time period, and baseline oxygen saturation, male sex was an independent predictor of in-hospital mortality ( Fig. 2).
To compare predictors of mortality, intubation, and ICU care between men and women, we first performed subgroup analysis to identify significant interactions between sex and other covariates. Then, we used sex-stratified regression models to estimate the effect sizes of the individual predictors in each sex.
Subgroup analysis for predictors of mortality identified a significant interaction between sex and baseline hypoxia (interaction p = 0.02) ( Fig. 1 and Supplemental Table 5). Among predictors of intubation, there were significant sex interactions for age (interaction p = 0.01), obesity (interaction p = 0.03), and chronic liver disease (interaction p = 0.02). Among predictors of ICU-level care, there were significant sex interactions for age (interaction p < 0.001) and obesity (interaction p = 0.02); there were also suggestive but non-significant interactions with hypertension (interaction p = 0.09) and hypoxia (interaction p = 0.08). In sex-stratified models, we observed that while hypoxia was associated with increased mortality in both sexes, the effect size was larger among women (OR 3.67, 95% CI 2.85-4.73) than men (OR 2.44, 95% CI 1.98-3.01) ( Fig. 2 Table 7).
One notable exception was obesity, which was more common (36.2% vs 32.9%, Fisher's exact p = 0.02) in the validation cohort.

Discussion
Sex differences in prevalence, pathogenesis, and outcomes have been observed across a variety of infectious diseases 27 . In general, men are more susceptible to infection while women mount more vigorous immune responses 28,29 . However, this does not necessarily translate into worse outcomes for men; a 2010 WHO study concluded that influenza outcomes are often worse for women, even though they consistently produce higher antibody titers in response to vaccination 30,31 . Sex differences in immunity may be related to the "dose" of immune-related genes located on the X chromosome, or sex hormonal effects on immune cells 32 . Observational cohort studies have consistently identified a difference in mortality outcomes between men and women infected with SARS-CoV-2 1-6 . An emerging body of literature has characterized differences in cytokines between men and women with COVID-19 26,[32][33][34] . However, these findings have yet to be translated into effective therapies, indicating that our mechanistic understanding of COVID-19 pathogenesis and sex differences remains limited. To enhance our understanding of the effect of sex on COVID-19 outcomes, we characterized the COVID-19 disease course of hospitalized men and women in a large healthsystem cohort and used stratified and interaction analyses to identify interactions between sex and other COVID-19 risk factors.
Men in our primary cohort were younger than the women and were less likely to have common comorbidities such as obesity and hypertension, which have been linked to adverse COVID-19 outcomes 20,35 . Despite this apparently favorable clinical profile, men were more likely to present with signs of severe disease, such as hypoxia and lymphopenia 10,36 , and were more likely to be intubated (16.4% vs. 13.2%, Fisher's exact p = 0.002) or receive intensive care (26.7% vs. 20.3%, Fisher's exact p = 1.8 × 10 −7 ). After adjusting for demographics and comorbidities, male sex was an independent risk factor for mortality, intubation, and intensive care, in accordance with findings of studies from the USA 2,5 , the UK 6 , Italy 1 , and China 11 . Male sex remained an independent risk factor for adverse outcomes in a validation cohort comprised of patients admitted to our health system at a later phase of the pandemic.
Given this persistent male-female disparity, we reasoned that there might be clinical factors, which preferentially affected men's or women's outcomes. By testing interaction terms and using sexstratified models, we identified hypoxia and obesity as risk factors, which preferentially increased women's risk of mortality and intubation/intensive care, respectively, in the primary cohort. However, in the validation cohort, the sex specificity of hypoxia did not replicate, while the sex specificity of obesity was reversed.
In the primary cohort, men were more likely to present with hypoxia (24.3% vs. 20.2%, Fisher's exact p = 8.6 × 10 −4 ), but the presence of hypoxia only increased men's risk of death by 2.44 times, compared to an increase of 3.67 times for women (interaction p = 0.02). In the validation cohort, hypoxia remained an independent predictor of mortality (OR 3.9, 95% CI 2.49-6.1), but there was no significant interaction between sex and hypoxia (interaction p = 0.94).
In a recent study of 6916 patients in Southern California, Tartof et al. reported an interaction between BMI and sex; increasing BMI had a greater impact on men's mortality risk than women's 35 . By contrast, we initially found that obesity preferentially increased women's risk of intubation and intensive care while there was no interaction between sex and obesity with regards to in-hospital mortality. However, the validation cohort  showed a significant interaction between sex and obesity (interaction p = 0.03) with higher mortality for obese men, in line with the findings of Tartof et al.
There were also significant interactions in the primary analysis between chronic liver disease and sex with regards to intubation, and between age and sex for both intubation and intensive care. Liver disease had a greater impact on women's risk of intubation than men; however, the number of patients with chronic liver disease in the primary cohort was small (N = 164), and this finding requires further evaluation. The interaction analysis for sex and age revealed that men in the youngest and oldest age groups were more likely than women to be intubated or receive intensive care (Fig. 1). These interactions were not significant in the validation cohort, but power was limited due to the reduced rate of intensive care intubation during the validation period.
Differences in the conditions of the pandemic and patient population may account for the differences in sex-specific risk factors between our primary and validation cohorts. The primary cohort consisted of patients admitted during the initial pandemic surge; there were no proven COVID-19 therapies at that time, and hospital capacity was severely taxed. The validation cohort reflects a later period of the pandemic, after the establishment of evidence-based treatment protocols and without the same strains on bed capacity [37][38][39][40] . The use of treatments such as remdesivir and dexamethasone was significantly more common in the validation cohort, and outcomes were correspondingly better. The comorbidity profile of patients in the validation cohort was also different. Consequently, it is challenging to compare results from these two cohorts. Nevertheless, the persistent association of male sex with worse outcomes is notable. Sex-specific interactions noted in either cohort should be taken in context.
The strengths of our study include its size, the diversity of the patient population, and two cohorts encompassing the evolution of the pandemic in New York City from initial surge to implementation of evidence-based treatments. Although larger, national-level patient cohorts have been published 6,41 , this Mount Sinai Health System cohort is on par with other health-system and hospital-network-based cohorts from the US and Italy 1-3,5,8 . The detailed clinical information in this cohort empowers us to explicitly address sex differences in mortality risk factors using a rigorous combination of stratification and interaction models. To our knowledge, we are the first to report comprehensive sexstratified data on clinical features, risk factors, and their association with outcomes.
This study has several limitations. Our dataset did not capture socioeconomic factors nor pre-admission clinical factors such as duration of symptoms; we cannot rule out the possibility that sex differences in seeking or accessing care may have confounded the outcomes. Comorbidity data were extracted from the electronic medical record and were not manually reviewed; their completeness and accuracy depend on data entered by medical providers during routine clinical care. In addition, due to limited exchange of medical data between institutions, comorbidity data may be incomplete for some patients who receive care from multiple health systems. Our results should be considered hypothesis-generating and require confirmation in other large cohorts as well as causal studies to establish mechanisms.
In conclusion, this study provides a detailed characterization of the clinical features and outcomes of men and women hospitalized with COVID-19. It highlights interactions between sex, obesity, and hypoxia with regards to mortality, intubation, and intensive care. These interactions nominate patient subgroups for further study and provide insights that may help explain the sex differences in outcomes of this disease.

Data availability
Data underlying Figs. 1 and 2 are available in Supplementary Data 1 and 2. Other deidentified data is available in accordance with Mount Sinai policy. Requests will be assessed by the Icahn School of Medicine Institutional Data Access Committee (contact via Joy.Dicker@mssm.edu).