Evolution of illness severity in hospital admissions due to COVID-19

Abstract

Background: The coronavirus disease 2019 (COVID-19) severity is influenced by multiple factors, such as age, underlying medical conditions, individual immunity, infecting variant, and clinical practice. The highly transmissible Omicron variants resulted in decreased COVID-19 screening capacity, which limited disease severity surveillance.

Objective: To report on the temporal evolution of disease severity among patients admitted to Québec hospitals due to COVID-19 between January 2, 2022, and April 23, 2022, which corresponded to the peak period of hospitalizations due to Omicron.

Methods: Retrospective population-based cohort study of all hospital admissions due to COVID-19 in Québec, between January 2, 2022, and April 23, 2022. Study period was divided into four-week periods, corresponding roughly to January, February, March and April. Regression using Cox and Poisson generalized estimating equations (GEEs) was used to quantify temporal variations in length of stay and risk of complications (intensive care admission or in-hospital death) through time, using the Omicron peak (January 2022) as reference. Measures were adjusted for age, sex, vaccination status, presence of chronic diseases, and clustering by hospital.

Results: During the study period, 9,178 of all 18,272 (50.2%) patients hospitalized with a COVID-19 diagnosis were admitted due to COVID-19. Of these, 1,026 (11.2%) were admitted to intensive care and 1,523 (16.6%) died. Compared to January, the risk of intensive care admission was 25% and 31% lower in March and April respectively, while in-hospital fatality continuously decreased by 45% lower in April. The average length of stay was temporarily lower in March (9%).

Conclusion: Severity of admissions due to COVID-19 decreased in the first months of 2022, when predominant circulating variants were considered to be of similar severity. Monitoring hospital admissions due to COVID-19 can contribute to disease severity surveillance.

Introduction

When a new severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) variant or sublineage appears, efforts are made to rapidly characterize its transmissibility and severity compared to previous variants. The Omicron BA.1 variant was first detected in Québec on December 8, 2021, during the Delta wave of the pandemic, and became predominant by December 12, 2021. Subsequently, the Omicron BA.2 variant appeared on January 1, 2022, and was predominant by March 27, 2022. A peak in hospital admissions due to SARS-CoV-2 (the coronavirus that causes COVID-19) was recorded on January 18, 2022 ^{Footnote 1}. Overall, the Omicron variant had higher transmissibility but lower severity compared to the Delta variant ^{Footnote 2Footnote 3Footnote 4Footnote 5Footnote 6}, while the Omicron BA.1 and BA.2 sublineages had comparable severity ^{Footnote 5Footnote 7Footnote 8Footnote 9}. Such information is essential for public health teams to help them anticipate the evolution of the epidemic, including the new variant’s impact on healthcare resources. In Canada, where the number of hospital beds per inhabitant is low and the workforce has been affected by the COVID-19 pandemic, information on severity will help to determine whether or not public health measures should be applied or maintained ^{Footnote 10Footnote 11}.

Severity of COVID-19 cases depends on factors beyond the characteristics of the virus. In times of high incidence, more hospitalizations will occur and the threshold for hospital admission/discharge might change, regardless of virulence ^{Footnote 12}. Natural, vaccine-induced, and hybrid immunity have increased in the population since the beginning of the COVID-19 pandemic but will vary according to time since infection or vaccination ^{Footnote 13Footnote 14Footnote 15}. Clinical care has also evolved with increasing knowledge and experience in treatment, as well as with the arrival of antiviral treatments ^{Footnote 16Footnote 17}. Finally, with the explosion of cases following the emergence of Omicron variants and the availability of rapid tests, accurate estimates of the total number of cases and, consequently, the proportion of severe disease in the general population, were no longer possible. In contrast, all patients admitted to hospital in the province of Québec receive a PCR test for COVID-19, a practice that was consistent throughout the pandemic ^{Footnote 18}. Propensity of hospitalization given a certain level of severity in Québec also was not impacted by the adoption of rapid tests. Thus, tracking the evolution of the severity of cases among those admitted to hospital due to COVID-19 represents a potentially interesting alternative for disease severity monitoring.

We aimed to describe the severity of hospital admissions due to COVID-19 in Québec between January 2022 and April 2022, which corresponded to the Omicron BA.1 and BA.2 waves. We measured length of stay, risk of intensive care admission, and risk of in-hospital death, and quantified temporal variations of these measures.

Methods

Study design and population

A retrospective population-based cohort was built using linked data to study all Québec hospitalizations for which COVID-19 led to hospital admission between January 2, 2022, and April 23, 2022 (Centers for Disease Control and Prevention, weeks 1–16). Patients were followed from admission until discharge, death, or final date of data extraction (May 25, 2022).

Data sources and variables

The COVID-19 hospitalizations were identified using the provincial hospital admissions database, which is a real-time version of the provincial hospital discharge database (MED-ECHO) routinely available before the pandemic. For this real-time database, hospital medical archivists reported any presence of COVID-19 during a hospital stay, regardless of other health conditions. Since December 30, 2021, archivists also provided admission diagnosis for all patients with a COVID-19 diagnosis during their hospital stay. Admission and hospital stay diagnoses are recorded according to the International Classification of Diseases 10^th Revision (ICD-10). Among all patients with a COVID-19 diagnosis during their hospital stay, those with an admission code related to COVID-19 were identified as admissions due to COVID-19. The list of COVID-19-related diagnostic codes used in provincial surveillance is provided in Appendix, Table A1. In addition to admission diagnosis, admission and discharge dates, age, sex, intensive care admissions, and death while hospitalized are also recorded in this database. The study period was divided into four four-week periods that corresponded to the peak (January) and the tail (February) of the BA.1 wave, the transition towards BA.2 (March), and the beginning and peak of the BA.2 wave (April) (Figure 1). Using a unique identifier, the hospitalization database was linked to:

• The Québec Integrated Chronic Diseases Surveillance System to identify patients with at least one of 31 comorbidities ^{Footnote 19}
• The provincial laboratory database to identify patients who had a positive SARS-CoV-2 test more than 90 days before the current admission (interpreted as a reinfection)
• The provincial immunization registry for information on COVID-19 vaccination status (individuals with at least two doses were considered adequately vaccinated)

Figure 1 - Text description

Time trends in hospital admissions due to the coronavirus disease 2019 (COVID-19), median length of stay and frequency of complications, January 2, 2022, and April 23, 2022, Québec. The study period was divided into four four-week periods that corresponded to the peak (January) and the tail (February) of the BA.1 wave, the transition towards BA.2 (March), and the beginning and peak of the BA.2 wave (April). Globally, patients were more frequently admitted to intensive care units in January and February, while length of stay remained relatively stable over time. Patients also died more frequently during the January peak of hospital admissions, with a gradual decrease throughout the following weeks.

Analyses

The proportion of admissions due to COVID-19 with an intensive care admission, as well as the proportion of patient deaths, were computed for each time period. The 25^th, 50^th and 75^th percentiles of length of stay were additionally computed, once again for each time period. Although length of stay was censored after 28 days, the estimated percentiles were always less than 28 days and so were unaffected. These proportions and duration were also stratified by age group (0–45, 46–55, 56–65, 66–75 and over 75 years old), sex, vaccination status, history of COVID-19 infection, and presence or absence of comorbidities, respectively.

Regression analyses were used to quantify the association between both length of stay and risk of complications (intensive care admission or death) vs. time period, using the Omicron peak period (January 2022) as reference. Cox proportional hazards regression was used to analyze length of stay, using random effects to model the possible clustering effect of hospitals. General estimating equations using a Poisson distribution and exchangeable correlation matrix were used to analyze risk of intensive care admission and death, accounting for the possible clustering effect of hospitals. For the above regressions, unadjusted and fully adjusted associations with time period are presented. Fully adjusted models used age group, sex, vaccination status, and chronic disease as covariates; no adjustment was made for history of COVID-19 infection because patients hospitalized for a reinfection were too rare (less than 3% of patients hospitalized for COVID-19). To isolate changes in disease severity from patient immunity to the disease, additional models were produced only for patients who had no known history of COVID-19 infection and who were not vaccinated. In subgroup analyses, separate models were also produced for each age group. Finally, after learning that three hospitals had largely underestimated intensive care admissions in early 2022, the fully adjusted regression was done, excluding these three hospitals, in a post hoc sensitivity analysis. All analyses were done using R 4.0.2; mixed effects Cox regression was done using the coxme package ^{Footnote 20}, while Poisson GEE was done using the geepack package ^{Footnote 21}.

Results

Between January 2, 2022, and April 23, 2022, 9,178 (50.2%) of all 18,272 patients who were hospitalized with a COVID-19 diagnosis were admitted due to COVID-19. Of these, 1,026 (11.2%) were admitted to an intensive care unit and 1,523 (16.6%) died while hospitalized (these outcomes were not mutually exclusive). Slightly over half of patients admitted due to COVID-19 were male (52.8%), and a majority of patients were over 65 years old (72.2%), adequately vaccinated (72.1%), experiencing their first known SARS-CoV-2 infection (98.1%) and had at least one comorbidity (83.7%). These statistics are described per four-week period in Table 1 (see Table A2 and Table A3 for a description of intensive care unit admissions and deaths per four-week period). Patient characteristics were relatively stable over time, except for a higher proportion of older patients and a lower proportion of inadequately vaccinated patients in March and April. In any given group, patients still hospitalized after 28 days represented less than 10% of inpatients.

Globally, patients were more frequently admitted to intensive care units in January and February, while length of stay remained relatively stable over time (Figure 1). Patients also died more frequently during the January peak of hospital admissions, with a gradual decrease throughout the following weeks (Figure 1). These time trends were also observed in regression analyses, after adjusting for age, sex, vaccination status, and presence of at least one comorbidity (Table 2). The proportions of patients admitted to intensive care were 25% and 31% lower in March and April (peak of BA.2; Table 1), respectively, as compared with the Omicron peak (January); this trend towards a risk reduction in time was observed in all age groups except for those 0–45 years old (Table 3). Results were similar when excluding the three hospitals that underestimated intensive care admissions (adjusted risk ratios of 0.92, 0.76 and 0.70 for February, March and April, respectively). The proportion of in-hospital deaths decreased continuously and was 45% lower in April, compared to January (Table 2); this trend was driven by patients over 75 years old, as 78% of deaths occurred in this age group (Table A3). In non-vaccinated patients admitted for a first COVID-19 episode, adjusted time trends in risk of intensive care admission and in-hospital death were very similar to those observed in the entire cohort (Table 2). Finally, the probability of remaining in hospital after any given number of days was 9% lower in March (transition towards BA.2) compared to January (BA.1 peak), but this was a temporary decrease (Table 2). No statistically significant change in length of stay was observed for hospitalizations of non-vaccinated patients. Cox regressions stratified by age group had extremely high statistical variability, indicating both increasing or decreasing lengths of stay (Table 3).

Discussion

This study showed a decreasing trend in the risks of intensive care admission and in-hospital death among patients admitted to hospital due to COVID-19 in Québec throughout the first 16 weeks of 2022. No clear trend emerged with respect to temporal variations in the length of hospital stay. Conclusions were similar in sensitivity analyses focusing on unvaccinated patients with no previous documented COVID-19 infection.

Many factors may have contributed to decreasing severity. Patient age, sex and comorbidities have been identified as risk factors for severe outcomes in the early stages of the pandemic ^{Footnote 12Footnote 22Footnote 23Footnote 24}, but analyses for these factors were adjusted and/or stratified, as well as controlled for vaccination status. Residual confounding may nevertheless remain. Xia et al. reported a positive association between in-hospital mortality (in all COVID-19-positive inpatients) and the proportion of available beds occupied by COVID-19-positive patients in Québec, during the first three waves of the pandemic ^{Footnote 12}. The arrival of the Omicron variant led to the highest number of patients hospitalized with a COVID-19 diagnosis since the beginning of the pandemic ^{Footnote 1}. This patient load may also have contributed to the trends observed in our study. However, the last four-week period included the peak of the BA.2 wave, and severity kept decreasing even though an increase would have been expected given the higher number of admissions. It is possible that this phenomenon may still have occurred but was not strong enough to reverse the overall trend. Clinical practices also keep evolving, with antiviral treatments becoming available at the beginning of the study period and with increasing accessibility over time ^{Footnote 17Footnote 25}. However, without access to patient load, healthcare worker absenteeism, or antiviral use data, these variables could not be accounted for. Finally, social determinants of health, which represent a well-known driver of inequalities in COVID-19 susceptibility and outcomes, were not accounted for in these analyses ^{Footnote 26}. However, the effect of social determinants is likely controlled for in the regression analyses, at least in part, through other covariates, such as comorbidities and vaccination status.

Variant composition also evolved during the study period and could have contributed to observed severity trends. Delta-infected patients were still being admitted to hospital in early January, which could explain a higher in-hospital mortality during the first four-week period, but not the decrease in severity observed for the last two periods ^{Footnote 27}. Estimates of the severity of the BA.1 and BA.2 sublineages have suggested a possible lower severity of BA.2 ^{Footnote 5Footnote 7Footnote 8}, though differences measured within each study were not statistically significant. Whole genome sequencing data were unavailable for hospitalized patients; therefore, an association between observed severity trends and variant composition could not be confirmed. Other possible factors are that patients from the more recent periods had shorter follow-up and thus less time to experience outcomes (discharge, intensive care unit admission or death), as not all inpatients had been discharged by the end of the study period. However, all patients were followed for at least 28 days, which should be sufficient to capture the majority of outcomes. The practice of PCR testing of all the patients admitted to hospital in Québec ^{Footnote 18} also rules out changes in testing practices as a factor in severity trends.

In the time preceding the study, PCR testing was done in the general population; nevertheless, not all cases, especially if mild or asymptomatic, were necessarily detected. Therefore, reinfection or the presence of previous COVID-19 infection could have gone undetected in some patients. However, this would only affect severity trends if the proportion of undetected reinfections varied over time. Overall, COVID-19 testing quality and coverage in Québec were high before December 2021 and the advent of Omicron. It is possible, however, that the proportion of hospital patients with unmeasured previous COVID infection acquired during or after December 2021 could have contributed to the observed decreasing severity for the month of April, since a previous infection is defined as one that occurs at least three months before the testing date. Finally, it is possible that the “adequate vaccination” criterion used in the regression analyses does not account for the effect of waning vaccine efficacy, which could result in misclassification of patients that were thought to be protected due to vaccine immunity. However, this effect is likely minimal, given that the majority (84%) of adequately vaccinated patients in this study received their last dose within seven months of hospital admission. This seven-month threshold is based on vaccine effectiveness studies ^{Footnote 28}. Sensitivity analyses (not shown), where patients receiving their last dose more than seven months after admission to hospital were classified as inadequately vaccinated, showed negligible difference in estimated severity trends.

When Omicron hit the province of Québec in December 2021, screening clinics and laboratories were quickly overloaded. January 2022 marked the end of two years of universal screening. At this time, a new screening strategy was adopted that targeted only certain subpopulations, mostly consisting of the elderly, especially in long-term care facilities, healthcare workers, and patients admitted to hospital ^{Footnote 29}. Surveillance of disease severity by following up on COVID-19 cases until hospital admission or death would therefore have been biased given the reasons behind the selection of these groups (e.g., increased vulnerability, higher exposure to disease and to vaccines, and the healthy worker effect). Monitoring severity among inpatients represented an alternative because all inpatients were still tested. Our previous work on disease severity comparing Omicron and Delta variants among inpatients suggested a lower severity of Omicron hospitalizations, concordant with other studies comparing these two variants with different methodologies ^{Footnote 3Footnote 4Footnote 5Footnote 6Footnote 30}. Wolter et al. reached convergent conclusions regarding the relative severity of BA.1 and BA.2 sublineages by measuring and comparing the difference in both risk of hospital admission among cases and risk of severe outcomes among inpatients ^{Footnote 8}.

The restriction of analyses only to patients admitted due to COVID-19 is an important strength of this study, as about half of all COVID-19-positive inpatients were admitted for other illnesses that can differ in severity from COVID-19. As well, healthcare-associated cases of COVID-19, which are more frequent in periods of high viral circulation, have been related to more severe outcomes ^{Footnote 31Footnote 32}. Unfortunately, admission diagnosis was only available from December 30, 2021, which prevented a comparison of Omicron waves with earlier waves. Before January 2022, all COVID-19-positive patients were analyzed, with the finding that median length of stay, proportion admitted to intensive care, and proportion of in-hospital deaths all varied in a similar manner over time, suggesting that length of stay could be used to inform disease severity ^{Footnote 30}. This correspondence was not observed in the present analysis, however. Length of stay may be influenced by patient load during peaks and its utility for surveillance of severity is therefore unclear. Also, the results of this study do not provide information on the effect of interventions that aim to prevent hospitalizations. For instance, compared to the general population, hospitalized cases over-represent individuals where vaccines and antivirals have not been successful. Finally, as was previously pointed out by Twohig et al., this surveillance informs the evolution of severity with a delay, as admissions follow case onset by a few days and as a majority of patients have to be discharged before intensive care admissions, in-hospital deaths, and length of stay can be assessed ^{Footnote 22}.

Conclusion

Throughout the first months of 2022, the risks of in-hospital death or intensive care admission decreased in individuals admitted due to COVID-19. Many factors, including changing immunity, reinfection prevalence, antiviral usage, and patient load may have contributed to this trend, which occurred during a time when virulence of predominant circulating variants were not excessively different. Hospital admissions due to COVID-19 represent an opportunity for monitoring trends in disease severity.

Authors' statement

• EL — Conceptualization, data analysis, interpretation, writing–original draft, writing–review & editing
• EF — Conceptualization, data analysis, interpretation, writing–original draft, writing–review & editing
• P-LT — Data analysis
• RG — Interpretation, writing–review & editing
• RT — Interpretation, writing–review & editing
• HG — Interpretation, writing–review & editing
• ZZ — Interpretation, writing–review & editing

The contents of this article and the opinions expressed therein are those of the authors and do not necessarily reflect those of the Government of Canada.

Competing interests

The authors have no competing interests to declare.

Funding

RG received funding from Québec Ministry of Health for a hospital surveillance network of respiratory hospitalizations, not related to the present study.

Appendix