Predictors of all‐cause mortality among patients hospitalized with influenza, respiratory syncytial virus, or SARS‐CoV‐2

Abstract Background Shared and divergent predictors of clinical severity across respiratory viruses may support clinical and community responses in the context of a novel respiratory pathogen. Methods We conducted a retrospective cohort study to identify predictors of 30‐day all‐cause mortality following hospitalization with influenza (N = 45,749; 2010‐09 to 2019‐05), respiratory syncytial virus (RSV; N = 24 345; 2010‐09 to 2019‐04), or severe acute respiratory syndrome coronavirus 2 (SARS‐CoV‐2; N = 8988; 2020‐03 to 2020‐12; pre‐vaccine) using population‐based health administrative data from Ontario, Canada. Multivariable modified Poisson regression was used to assess associations between potential predictors and mortality. We compared the direction, magnitude, and confidence intervals of risk ratios to identify shared and divergent predictors of mortality. Results A total of 3186 (7.0%), 697 (2.9%), and 1880 (20.9%) patients died within 30 days of hospital admission with influenza, RSV, and SARS‐CoV‐2, respectively. Shared predictors of increased mortality included older age, male sex, residence in a long‐term care home, and chronic kidney disease. Positive associations between age and mortality were largest for patients with SARS‐CoV‐2. Few comorbidities were associated with mortality among patients with SARS‐CoV‐2 as compared with those with influenza or RSV. Conclusions Our findings may help identify patients at greatest risk of illness secondary to a respiratory virus, anticipate hospital resource needs, and prioritize local prevention and therapeutic strategies to communities with higher prevalence of risk factors.


| INTRODUCTION
The COVID-19 pandemic has put tremendous strain on hospital systems, and exposed long-standing issues in healthcare capacity. 1 Knowing who is at highest risk of severe disease from respiratory viruses may support proactive clinical decision-making and help distribute resources to healthcare settings with high prevalence of risk factors. 2 This is particularly useful in the context of a new and emerging respiratory virus where information and resources are scarce. 2,3 Several studies have compared shared and divergent predictors of severe disease among patients with influenza and respiratory syncytial virus (RSV), [4][5][6][7][8][9] two respiratory viruses with high seasonal prevalence prior to the emergence of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). However, few papers have compared predictors of severity across influenza, RSV, and SARS-CoV-2.
Communities are returning to pre-pandemic contact and exposure patterns, which may increase the risk of all respiratory infections. At the same time, laboratory diagnostic testing is transitioning to pre-pandemic approaches, where only a subset of hospitalized patients with viral respiratory or influenza-like illness receive laboratory-confirmed diagnoses. 10 Thus, during periods of respiratory viral epidemics (particularly with novel emerging pathogens), shared predictors of severity across the clinically important respiratory viruses may: (1) reduce morbidity and mortality by prioritizing preventions (e.g., vaccinations), testing, and access to therapeutics (e.g., antivirals); and (2) prepare healthcare settings that will require greater resources based on the prevalence of the underlying predictors.
We conducted an observational study using extensive health administrative data from Ontario, Canada, to identify the direction and magnitude of shared and divergent predictors of 30-day all-cause mortality following hospitalization with influenza, RSV, or SARS-CoV-2 (prior to vaccine availability or variant emergence).

| Study setting and design
We conducted a retrospective cohort study of patients hospitalized with influenza, RSV, or SARS-CoV-2 using population-based laboratory and health administrative data from Ontario, Canada (population 14.7 million). 11 Ontario's healthcare system provides publicly funded physician services, laboratory testing, and hospital care for all residents with a provincial health card. Datasets used in this study were linked using unique encoded identifiers and analyzed at ICES. 12 2.2 | Case definitions and outcomes

| Hospitalizations
We generated three study cohorts to assess predictors of severe outcomes among patients hospitalized with influenza, RSV, and SARS-CoV-2, respectively. Patients with influenza and RSV were identified using hospitalization data from the Canadian Institute for 91%; RSV: 91%). 13 We used DAD, the Ontario Laboratories Information System (1) they were documented as hospitalized in DAD and had a positive polymerase chain reaction test for SARS-CoV-2 within 14 days before or 3 days after hospital admission; or (2) they were documented as hospitalized in CCM.

| Mortality
Our primary outcome of interest was 30-day all-cause mortality following hospital admission with influenza, RSV, or SARS-CoV-2.
We used the Registered Persons Database (RPDB) and CCM to identify patients who died within 30 days of hospital admission.
RPDB contains basic demographic information including age, sex, postal code, and date of death among all residents with an Ontario health card.

| Inclusion and exclusion criteria
Hospitalized patients were excluded if: they were not eligible for the Ontario Health Insurance Plan; their birthdate, sex, or postal code was missing from RPDB; their residential postal code was outside of Ontario; they were older than 105 years according to their birthdate in RPDB; or their recorded death date predated hospital admission ( Figure 1). Only one hospitalization per patient was included (per season for influenza and RSV, and overall for SARS-CoV-2). Among patients hospitalized with influenza or RSV, we included the first hospital admission of the season. Among patients hospitalized with SARS-CoV-2, we included any hospitalization that resulted in death within 30 days of admission, or the first admission if no other admission was associated with 30-day mortality. Variation in inclusion criteria were due to suspected differences in hospital admission and discharge behavior across virus cohorts. For example, early in the pandemic, evidence suggested that patients hospitalized with SARS-CoV-2 had a relatively high likelihood of readmission within 60 days of discharge. 14 Patients hospitalized with influenza or RSV were excluded if they were hospitalized outside of the respective respiratory virus season. Respiratory virus seasonality was defined as November to May for influenza, and November to April for RSV to align with case definitions from Hamilton et al., 13 and to create the most inclusive time frame to capture seasonal virus activity in Ontario. 15

| Predictors of 30-day all-cause mortality
We selected potential predictors of 30-day all-cause mortality a priori.
Variables were considered if they had documented or suspected associations with respiratory virus acquisition or severity, or healthcare access in peer-reviewed, published literature.

| Demographic characteristics
We used data from RPDB to describe pertinent individual-level demographic characteristics including age, sex, and residence in rural neighborhoods. 16 Rural neighborhoods were defined as those outside commuting zones of population centers (i.e., centers with more than 10,000 residents).
We used aggregated 2016 Canadian census data to describe neighborhood-level social determinants of health associated with risk of respiratory virus acquisition, [17][18][19] access to care, and discrimination within health care settings 20,21 including income, 22 household size, 23 and "ethnic concentration" 24 (herein referred to as percent racialized).
Neighborhood-level variables were categorized into quintiles (i.e., 1 = 20% of neighborhoods with lowest values; 5 = 20% of neighborhoods with highest values). Patients were assigned a quintile according to their residential postal code. We describe derivation of the neighborhood-level determinants of health in depth in the supporting information.

| Underlying health conditions
Pertinent underlying health conditions included: asthma, chronic obstructive pulmonary disease (COPD), hypertension, cardiac ischemic disease, congestive heart failure, stroke, dementia or frailty, chronic kidney disease, advanced liver disease, and immunosuppression (i.e., patients with a cancer diagnosis in the past 5 years, human immunodeficiency virus, solid organ or bone marrow transplant, or another immunodeficiency condition). 25,26 We used validated case definitions and health administrative data to classify each individuallevel health condition. Case definitions and validity are described in detail in Table S1.

| Other covariates
Other predictors of severe outcomes included residence in a longterm care home (LTCH), 27,28 and seasonal immunization against influenza. 29 Table S2.    Figure 2).

| Statistical analyses
In adjusted models, shared predictors of mortality included: older age, male sex, residence in a LTCH, and chronic kidney disease ( Figure 2,

| DISCUSSION
We identified shared and divergent predictors of mortality among patients hospitalized with influenza, RSV, or SARS-CoV-2 using population-based health administrative data from Ontario, Canada. In multivariable models, common predictors of 30-day all-cause mortality following hospitalization included older age, male sex, residence in a LTCH, and chronic kidney disease.
Older age and male sex were predictive of increased mortality across all respiratory virus cohorts, which aligns with numerous studies from high-income countries 4-9,33-35 and confirms the need to consider age and sex in clinical practice. The magnitude of association between older age and mortality was largest among patients with SARS-CoV-2, confirming robust evidence that age is an important predictor of severity among COVID-19 patients, and should be used to guide targeted COVID-19 preventions and therapeutics. 35,36 Residence in a LTCH was also a common predictor of 30-day allcause mortality; however, associations were weaker among patients hospitalized with SARS-CoV-2. Differences in magnitudes of association may be due to greater selection bias of LTCH residents hospitalized with SARS-CoV-2 in comparison to those with influenza or RSV.
For example, in Ontario, less than one quarter of COVID-19-positive LTCH residents were hospitalized prior to death, compared with nearly 80% of COVID-19-positive community residents during the first wave of the pandemic. 37 It has been suggested that LTCH residents with SARS-CoV-2 may have been less likely to be hospitalized due to advanced care directives and/or informal policies that discouraged transfers of critically ill residents. [37][38][39] The difference in hospitalizations prior to death narrowed in the pre-vaccination second wave of the pandemic in Ontario, 37 suggesting that the selection biases may have been specific to wave 1 and may not be reflective of past influenza or RSV seasonal epidemics.
Similar to previous studies, 40   This study is also limited by a lack of data on other important predictors of respiratory infection severity such as pregnancy, 44,45 obesity, 46 and individual-level social determinants (e.g., economic marginalization and racialization), which are known to mediate quality of hospitalized care and rates of respiratory virus infection. [17][18][19][20][21] When using our results to inform prioritization of services, or to develop clinical prediction tools, we must consider these limitations so that other at-risk patients do not fall through the cracks.
Finally, to provide insights on shared predictors of mortality in the context of a novel, emerging pathogen, we purposefully restricted the study period of SARS-CoV-2 to exclude hospitalizations of patients supervision.

PEER REVIEW
The peer review history for this article is available at https://publons. com/publon/10.1111/irv.13004.

SUPPORTING INFORMATION
Additional supporting information may be found in the online version of the article at the publisher's website.