A cohort study of patients hospitalised with SARS-CoV-2 infection in Ontario: patient characteristics and outcomes by wave

. SARS-CoV-2 was the most responsible diagnosis for 70% of admissions during waves 1–2 and 42% in wave 5. The proportion of admitted patients who were long-term care residents was 18% (n = 1111) during wave 1, decreasing to 10% (n = 1468) in wave 2 and <5% in subsequent waves. During waves 1–3, 46% of all admitted patients resided in a neighbourhood assigned to the highest ethnic diversity quintile, which declined to 27% during waves 4–5. Compared to wave 1, 90-day mortality was similar during wave 2 (adjusted risk ratio [aRR]: 1.00 [95% CI: 0.95–1.04]), but lower during wave 3 (aRR: 0.89 [0.85–0.94]), wave 4 (aRR: 0.85 [0.79–0.91]) and wave 5 (aRR: 0.83 [0.80–0.88]). Improvements in survival over waves were observed among elderly patients (p-interaction <0.0001). Critical care admission was significantly less likely during wave 5 than previous waves (adjusted odds ratio: 0.50 [0.47–0.54]). The length of stay was a median of 8.5 (3.6–23.8) days during wave 1 and 5.3 (2.2–12.6) during wave 5. After adjustment, the mean length of stay was on average –10.4 (–11.1 to –9.8) days, i.e. shorter, in wave 5 vs wave 1.


Introduction
Since the start of the COVID-19 pandemic in the province of Ontario, Canada, the social and epidemiological circumstances surrounding infection and its consequences have changed with each wave of infection.There have been varying degrees of social behavioural adaptations (e.g.wearing a mask, working from home, avoiding large crowds) and social regulations (e.g.restrictions and closures); differences in the predominant SARS-CoV-2 (Severe Acute Respiratory Syndrome Coronavirus 2) variants circulating (wild-type, Alpha, Delta, Omicron) associated with observed differences in virulence and transmissibility; and changes in population-level immunity acquired through vaccination or prior infections [1][2][3][4].
Although Omicron was more transmissible than previous variants, some studies from the United States did not report substantial differences in mortality between Omicron and Delta, even after accounting for vaccination status [5].Conversely, other studies from England and Denmark found a lower 30-day mortality rate with Omicron after similar adjustment [6,7].One study from the Northeastern United States demonstrated better survival with the Omicron variant, but this finding was observed only for the BA.2 Omicron subvariant, with similar mortality between Delta and the original Omicron subvariant B.1.1.529[8].Generally, Canadian studies agreed that the severity of infection was lower with the Omicron variant, but more information is needed to reflect changes in severity and Steven Habbous, PhD Ontario Health 525 University Avenue Toronto CA-ON M5G 2L3 Steven.habbous[at]ontariohealth.ca patient characteristics over time, and more follow-up is needed to reliably ascertain differences in mortality [9][10][11].
Considering that testing and surveillance measures have changed throughout the pandemic, the population susceptible to recorded SARS-CoV-2 infection in the province may have shifted across waves.In March 2020, non-essential workplaces, schools and the Canada-United States border closed.Simultaneously, physical distancing and infection prevention and control requirements were implemented [1].Following the first wave, restrictions were gradually removed during the summer of 2020; however they were reinstated in the second wave of autumn 2020 and the beginning of winter 2021.Initial vaccination rollout targeted higher-risk individuals (e.g. the elderly and those with chronic disease), "hot spot" neighbourhoods (e.g. more marginalised communities that were exhibiting relatively high rates of SARS-CoV-2 positivity) and settings (e.g.people residing in long-term care facilities; essential workers) [12,13].
In the present study, we add to the literature by examining the wave-over-wave characteristics and outcomes of patients admitted with SARS-CoV-2 at the population level.The primary outcome examined was 90-day mortality.Secondary outcomes were use of critical care and total hospital length of stay.Patient characteristics examined included both clinical and sociodemographic attributes.

Setting
Ontario is Canada's most populous province (15 million people).Healthcare is provisioned under a single-payer universal healthcare system.During the study period, Ontario experienced five distinct waves of COVID-19 where infection was predominantly caused by a specific variant.Waves 1 and 2 corresponded to the wild-type variant, wave 3 Alpha, wave 4 Delta and wave 5 Omicron.Approximate start dates for each wave correspond to an inflection point in the number of cases in the population (figure 1), with the end date of each wave defined as the start of the next wave (31 March 2022 for wave 5) [14].

Inclusion criteria
Hospital admissions were captured from the Discharge Abstract Database (DAD), which comprehensively captures all hospitalisations in Ontario (figure S1 in the Appendix).To avoid duplicate counting of admissions due to hospital transfers, admission records were resolved into episodes using information about the time of admission since the previous discharge, evidence of inter-or intra-hospital transfers and planned versus unplanned (re)admissions [15].We started with the definition reported by the Canadian Institute for Health Information: an inpatient episode of care included admissions occurring within 6 hours of previous discharge or admissions occurring within 12 hours of previous discharge but with evidence of a transfer (e.g."transfer to" from the prior admission or a "transfer from" for the current admission) [16,17].Admissions known to be planned (readmit code 1) were classified as admissions that belonged to the previous episode until 1 week after the previous discharge.
Data were extracted on 15 August 2022.Episodes starting between 26 February 2020 (the start of the COVID-19 pandemic in Ontario) and 31 March 2022 were included.All admission records had a valid health card number, a unique encoded identifier used to link between datasets.

Definition of SARS-CoV-2 admission episode
To determine whether an admission episode was related to SARS-CoV-2, we searched all diagnostic codes (up to 25) during the entire episode for the ICD-10 code U071 (COVID-19 infection, known) or U072 (COVID-19 infection, suspected).We also searched for U071 or U072 from outpatient hospital encounters (any of the 10 diagnostic codes in NACRS, the National Ambulatory Care Reporting System) during and within 2 weeks prior to the start of each admission episode [18].Lastly, we also captured positive SARS-CoV-2 polymerase chain reaction or rapid anti-

Original article
Swiss Med Wkly.2024;154:3636 gen tests from the Ontario Laboratories Information System (OLIS).OLIS includes data from approximately 90% of labs in Ontario conducting SARS-CoV-2 tests since the start of the pandemic [19].We considered an admission episode to be related to SARS-CoV-2 if any source indicated SARS-CoV-2 positivity during (DAD, NACRS, OLIS) or within 2 weeks prior (NACRS, OLIS) to the admission episode (Technical Appendix).Since our index date is the first admission related to SARS-CoV-2, we did not apply a hierarchy to these sources of evidence.The first ever admission episode was retained per patient.

Definitions and covariates
The index date was defined as the start of the first hospital admission episode related to SARS-CoV-2.Patients were omitted if they could not be linked to the Registered Persons Database or their death date preceded the admission date.
Comorbidity was derived using ICD-10 codes for the year prior using DAD or NACRS.Residence in a long-term care (LTC) facility or similar was defined using the Ontario Health Insurance Plan (any billing code starting with "W") or DAD/NACRS with a transfer code to/from a LTC facility.Sociodemographic characteristics included urban/rural residence, neighbourhood-level material deprivation quintiles and ethnic diversity quintiles from the Ontario Marginalization Index and transfer from supportive/group housing (defined using transfer codes to/from such housing).Material deprivation, residential instability, dependency and ethnic diversity were neighbourhood-level indices derived from the 2016 Census [20].SARS-CoV-2 was considered the most responsible diagnosis (MRDx) if the admission record that started the episode indicated either U071 or U072 as such.

Outcomes
The primary outcome was 90-day all-cause mortality since the start of the admission episode.The date of death was obtained from the Registered Persons Database.Patients with no death date were assumed to have been alive at 90 days.For Kaplan-Meier plots, the follow-up time was calculated as the date of death minus the date the admission episode started.The date 31 July 2022 was used as the censor date (the most recent date mortality data were available).Secondary outcomes included use of critical care during the episode and total length of stay (LOS).Use of critical care involved any stay in either an adult or neonatal intensive care unit.The total length of stay was calculated, in hours, using the date-time from the admission that started the admission episode until the date-time of the discharge from the admission episode.

Statistics
Wave-over-wave patient and admission characteristics were presented using descriptive statistics.All-cause survival was presented using Kaplan-Meier plots.90-day mortality was compared using modified Poisson regression, reporting risk ratios (RR) with 95% confidence intervals (CI) [21].To examine whether 90-day mortality changed over wave for certain patients, we also introduced interaction terms between that patient's characteritic and wave.If the interaction term p value <0.05, then we reported the RR of wave on 90-day mortality stratified by that characteristic.
Use of critical care during the admission was assessed using logistic regression, reporting odds ratios (OR) with 95% CI.Linear regression was used to compare hospital LOS, reporting beta coefficient and 95% CI, representing the effect of a 1-unit change of the covariate on the total LOS (in days).Homoscedasticity was assessed via inspection of residual-versus-predictor plots and the modified Park test for linear-normal, log-gamma, and log-Poisson models, but heteroscedasticity could not be resolved.We instead used a heteroscedasticity-consistent standard error estimator of OLS parameter estimates (HC3), which does not require the assumption of homoscedasticity for valid inferences [22].
To estimate the effect of wave of COVID-19 on outcomes, unless otherwise stated, all effect measures were adjusted for age at admission, sex, SARS-CoV-2 as the most responsible diagnosis, rurality, neighbourhood marginalisation quintiles, long-term care residence, transfer from supportive housing residence, comorbidity (continuous), ambulance arrival, urgent admission, overnight admission and hospital transfer because these were all believed to be confounders for the effect of COVID-19 wave on the outcome.For 90-day mortality, we also adjusted for use of critical care since this may be a strong predictor of SARS-CoV-2-related severity.All analyses were performed at Ontario Health using Statistical Analysis Software version 9.4 (SAS Institute Inc., Cary, NC, USA).Only completecase analyses were employed.

Privacy
This study was compliant with section 45(1) of PHIPA (Ontario Health is a prescribed entity): ethics review and patient consent were not required.The health card number was used to link between data sources (unique identifier).A protocol was not prepared for this work.

Wave-over-wave characteristics
During waves 1-3, 96% of all hospital admissions were for patients who lived in an urban region, which dropped to 90% during waves 4-5 (compared with ~86% for the entire Ontario population).Using the marginalisation index quintiles, patients hospitalised with SARS-CoV-2 were more likely to reside in an area having the highest quintile of residential instability (31% over all waves) and were more likely to reside in a neighbourhood assigned the highest quintile of material deprivation (28% over all waves), with little change over time.In contrast, 47% of all admitted patients during waves 1-3 resided in a neighbourhood assigned the highest ethnic diversity quintile, which fell to 29% during waves 4-5.The proportion of admitted patients who were residents of a LTC facility was 18% (n = 1099) during wave 1, which decreased to 10% (n = 1455) in wave 2 and remained <5% thereafter.The proportion of admitted patients who were transferred from supportive housing followed a similar trend as LTC residence.The reduction in mean comorbidity score (lowest in wave 3) was mirrored by a rise in the composition of patients who were <60 years of age (highest in wave 3).
Over the study period, admitted patients were more likely than not to arrive by ambulance (n = 43,505 or 59%) and be admitted during daytime hours (n = 46,761 or 64%).Most admissions were urgent (>95% in waves 1-3; 89% in wave 5).The number of admission episodes with at least one hospital transfer increased from 7.9% in wave 1 to 21% in wave 3, but subsequently declined to 14% in wave 4 and 5.8% in wave 5.

Critical care
Use of critical care was least likely during wave 5 (OR: 0.50 [0.47-0.54])and more likely if SARS-CoV-2 was the most responsible diagnosis (p<0.0001).Use of critical care was also more likely among males (OR: 1. 49  Stratified by the diagnosis type, use of critical care decreased wave-over-wave more substantially when SARS-CoV-2 was the most responsible diagnosis (OR: 0.39 [0.36-0.43]for wave 5 versus 1), although there was little effect for patients where SARS-CoV-2 was not the most responsible diagnosis (OR: 0.81 [0.69-0.96]for wave 5 vs wave 1) or if diagnosed by Ontario Laboratories Information System alone (OR: 0.92 [0.71-1.18]for wave 5 vs wave 1) (table 5).

Length of stay
The total crude hospital length of stay was shortest during wave 5 (mean ± SD: 10.5 ± 14.3 days).After adjustment ), but remained somewhat steady across all waves when SARS-CoV-2 was the MRDx (range was 5-7 days shorter between waves 2-5 versus wave 1) (table 5).

Discussion
We found that the wave-over-wave characteristics of patients hospitalised with COVID-19 changed markedly throughout the pandemic.Many of the sociodemographic disparities observed during the first two waves of the pandemic (e.g. higher rate of admission among the elderly, people from ethnically diverse neighbourhoods, residents of long-term care or supportive housing) were reduced by the third (Alpha-dominant) or fourth (Delta-dominant) waves as vaccination efforts beginning in the middle of wave 2 prioritised high-risk populations (figure 1).These efforts may have translated into an improved 90-day mortality, specifically among elderly patients.
Patients admitted with SARS-CoV-2 in the first wave were predominantly urban dwellers (96% compared with 85% in the rural areas) and were more likely to live in neighbourhoods in the quintile with the highest material depri-vation (29%), the highest residential instability (32%) and the highest ethnic diversity (46%) [23].By waves 4 (Deltadominant) and 5 (Omicron-dominant), however, 89% of admitted patients resided in an urban area and 28% of admitted patients resided in neighbourhoods having the highest ethnic diversity.These findings are consistent with other studies demonstrating that more-marginalised communities were disproportionately affected by COVID-19 [12,24,25].Active health system monitoring of SARS-CoV-2 positivity identified communities disproportionately affected by COVID-19, resulting in targeted vaccination efforts (figure 1) and prioritised modernisation of LTC facilities [1,13,[26][27][28].While wave-over-wave shift to rural residents may reflect the natural progression of viral transmission to less population-dense regions [29], reduced disparities based on ethnic diversity and LTC residency suggests that these efforts were successful in reducing (but not eliminating) the disproportionate burden of COVID-19 in marginalised neighbourhoods.
There were some indications that the severity of SARS-CoV-2 admissions attenuated over time, particularly with the Omicron variant (wave 5 in Ontario) [30][31][32].We observed a small reduction in wave 5 in the proportion of admissions deemed urgent, patients arriving by ambulance, the proportion of admissions requiring critical care and the mean total length of stay.During waves 4-5, SARS-CoV-2 was the most responsible diagnosis in only 46% of patients admitted (compared with 73% in the first 3 waves).This observation may be driven by increases in incidental diagnoses, protection from prior immune responses and higher transmission despite lower case fatality associated with the Omicron variant [33][34][35].Even after adjusting for important sociodemographic and clinical characteristics, we observed a reduction in 90-day mortality beginning in wave 3 for most groups of patients.Changes in survival were not observed among patients aged 40-69 years, which was unexpected since vaccination eligibility was expanded to 18+ four months before the start of the fourth wave (a wavelet attributed to the Delta variant) and 8 months before the start of the fifth wave (driven by the Omicron variant).Although survival was generally higher among younger pa-tients, the lack of wave-over-wave improvement in this group requires further investigation and is likely multifactorial (e.g.vaccine hesitancy among young adults, small room for improvement or changing case mix) [14,36].

Literature review
Studies in the literature reported similar shifts in sociodemographic characteristics, with reductions in age and comorbidities across waves [37][38][39][40][41][42].In particular, several studies reported lower age over time, including studies conducted in the United Kingdom [37], United States [38, # Odds ratio (OR) with 95% confidence intervals (CI) from logistic regression predicting use of critical care anytime during the hospital stay.
## Beta coefficient interpretable as the number of days longer (positive) or shorter (negative) than the reference category, a 10-year increase in age or a 1-unit increase in comorbidity score.Beta coefficients correspond to ordinary least square regression (identity link, normal distribution) with standard errors and p values estimated using heteroscedasticity-consistent standard error estimators.39], Europe [40,42], South America [41] and Canada [11].

Original article
One study found that individuals with Omicron tended to be 20-40 years old, whereas Delta infections were more often found in children 12 years or younger or in adults 60 ** p = 0.054 for interaction between wave and long-term care residence.years or older [9].In one study, there was a higher number of black patients hospitalised in the third wave compared to the first two waves [38].Additionally, the Delta variant had a more uniform geographical spread, whereas Omicron was more concentrated within urban geographical regions in British Columbia [9].Similar to our study, admitted patients were less likely to be LTC residents by wave 3 [43].
Owing to a multitude of factors (e.g.decreasing age and comorbidity, vaccination efforts, change in treatment), several studies internationally reported reductions in total admission length of stay or use of critical care over time [11,[38][39][40][41][42][44][45][46].One Canadian study found no difference in mortality among older adults (≥65 years) after adjusting for age, sex, number of comorbidities, ICU admission, frailty and delirium between waves 1-3.[43].This aligns with our study, where we observed improvement in survival among patients ≥70 years by the Delta and Omicron waves.In one English study that stratified the results by ethnic group, a lower mortality was reported over time during the first two waves of the pandemic, but the decline in mortality was more pronounced among the white group [47].One Canadian study from Alberta and Ontario observed a 2-fold increase in mortality in the Delta wave compared to previous waves [11].A more recent Canadian study demonstrated reduced mortality with Omicron compared with wild-type and Alpha, but no difference with Delta, but did not assess changing use of critical care and was not population-based (comprised a subcohort of an angiotensin receptor blocker observational cohort) [48].
A study from Copenhagen observed a 40% reduction in 6-day mortality after admission with Omicron versus Delta [7].

Strengths
One of the strengths of this study is the examination of patient characteristics and outcomes over time and stratified by important sociodemographic characteristics.Our findings are consistent with reports from the early pandemic for 90-day mortality (24.4% in Brazil, 20.3% from the United States) and total length of stay, in addition to the sociodemographic characteristics associated with hospitalisations [49,50].Thus, we expect our results to be generalisable to other jurisdictions.Moreover, this is the first population-based study in Ontario (Canada's most populous province) that includes the Omicron wave.
The hypothesis-generating nature of this study is another strength of this work.The most common post-admission diagnosis was SARS-CoV-2, which may represent a subset of patients who contracted SARS-CoV-2 during their hospital admission or were diagnosed incidentally during their admission [51].The next most common post-admission diagnoses included urinary tract infection and acute renal failure [51].Acute kidney injury has been documented as a common and serious adverse event associated with SARS-CoV-2 infection.Acute kidney injury is a risk factor for end-stage renal disease [52], but whether this relationship holds when SARS-CoV-2 is the aetiological agent remains to be established.These represent important avenues for future work.

Limitations
One limitation is that SARS-CoV-2 vaccination data were unavailable for analysis.Vaccination efforts prioritised high-risk populations based on existing morbidity (e.g.chronic kidney disease), setting (e.g.long-term care facility) and geography (e.g.neighbourhoods with high marginalisation indices based on postcode).The drastic changes in patient sociodemographics since the early pandemic suggests that these efforts were effective, but the high prevalence of acute kidney failure among patients hospitalised with SARS-CoV-2 requires further examination.Second, without individual-level sociodemographic information, we instead relied on neighbourhood-level characteristics.While these were informative and valid for this work (e.g.since public health measures were often made at the neighbourhood-level), neighbourhood-level features are not a substitute for individual-level data such as race and ethnicity [53].Third, there is no gold-standard definition for SARS-CoV-2-related admission.Although pressure on laboratories and the resulting backlog is unlikely to affect hospitalised patients who may be prioritised over community testing, validation is needed.Our survival analysis suggests that using DAD, even in the absence of evidence from Ontario Laboratories Information System, identified many patients hospitalised for SARS-CoV-2 that may not be captured by OLIS.The 90% coverage by OLIS that we observed is expected, since not all laboratories report to OLIS.Fourth, the characteristics of patients admitted for SARS-CoV-2 infection may differ from those admitted with SARS-CoV-2 infection.We used COVID-19 as the most responsible diagnosis as a surrogate for this, but validation is needed.Fifth, we only retained the firstever admission episode per patient related to SARS-CoV-2 infection.This was done in order to reduce bias when comparing outcomes between different waves, since the likelihood of previous admissions increases over time and may influence the outcomes for subsequent admissions.

Conclusion
Over the course of the pandemic, the sociodemographic characteristics of patients hospitalised with SARS-CoV-2 changed significantly, but SARS-CoV-2 hospitalisations still disproportionately affected more-marginalised regions.Improved survival and reduced use of critical care during the Omicron wave are reassuring.
However, the analyses, conclusions, opinions and statements expressed herein are those of the authors, and not necessarily those of CIHI or MOH.
We acknowledge the support of the Ministry of Health in this report.All views expressed are those of the authors of this report and do not necessarily reflect those of Ontario Health or the Ministry.

Financial disclosure
This research received no specific grant from any funding agency.

Potential competing interests
All authors have completed and submitted the International Committee of Medical Journal Editors form for disclosure of potential conflicts of interest.No potential conflict of interest related to the content of this manuscript was disclosed.

Figure 1 :
Figure 1: Timeline of important events as it relates to COVID-19 waves, vaccination roll-out and vaccination eligibility.

Figure S4 :Figure S5 :
Figure S4: Overlap between sources of evidence for SARS-CoV-2-related hospitalisations.(A)Evidence of SARS-CoV-2 in NACRS and Ontario Laboratories Information System (OLIS) during the admission episode.(B) Evidence of SARS-CoV-2 in NACRS and OLIS during or 2 weeks before the admission episode.

Table 1 :
Characteristics of admissions by COVID-19 wave.forpatient and admission characteristics, the total admission length was a mean 7.8 (95% CI: 7.1-8.5)days and 10.5 (95% CI: 9.9-11.2) days shorter for patients admitted in wave 2 and 5, respectively, vs wave 1.The total length Not mutually exclusive, any diagnostic position and type (ICD-10 diagnostic code starts with the indicated letter). *
* Odds ratios (OR) with 95% confidence intervals (CI) were derived from a logistic regression model.**Adjusted for all covariates presented in this table.Swiss Medical Weekly • www.smw.ch• published under the copyright license Attribution 4.0 International (CC BY 4.0)
* Risk ratios (RR) with 95% confidence intervals (CI) were derived from a modified Poisson regression model predicting 90-day mortality since the start of the admission episode.** Adjusted for all covariates presented in this table.*** p value from interaction term with COVID-19 wave.

Table 4 :
90-day mortality stratified by covariates with a significant statistical interaction with wave (see table3).Risk ratios (RR) with 95% confidence intervals (CI) were derived from a modified Poisson regression with wave 1 serving as the reference category.RRs are adjusted for all covariates presented in table 2.
* Adjusted for all variables in table3.Swiss Medical Weekly • www.smw.ch• published under the copyright license Attribution 4.0 International (CC BY 4.0)

Table S1 :
Agreement between DAD, NACRS and OLIS.n = 2,059,517 admission episodes; Restricted to admission episodes starting between 1 April 2020 (when ICD-10 codes became available for use) and 31 March 2022 (most recent data available).