Article Figures and data Abstract Editor's evaluation Introduction Methods Results Discussion Appendix 1 Data availability References Decision letter Author response Article and author information Metrics Abstract Background: Whilst timely clinical characterisation of infections caused by novel SARS-CoV-2 variants is necessary for evidence-based policy response, individual-level data on infecting variants are typically only available for a minority of patients and settings. Methods: Here, we propose an innovative approach to study changes in COVID-19 hospital presentation and outcomes after the Omicron variant emergence using publicly available population-level data on variant relative frequency to infer SARS-CoV-2 variants likely responsible for clinical cases. We apply this method to data collected by a large international clinical consortium before and after the emergence of the Omicron variant in different countries. Results: Our analysis, that includes more than 100,000 patients from 28 countries, suggests that in many settings patients hospitalised with Omicron variant infection less often presented with commonly reported symptoms compared to patients infected with pre-Omicron variants. Patients with COVID-19 admitted to hospital after Omicron variant emergence had lower mortality compared to patients admitted during the period when Omicron variant was responsible for only a minority of infections (odds ratio in a mixed-effects logistic regression adjusted for likely confounders, 0.67 [95% confidence interval 0.61–0.75]). Qualitatively similar findings were observed in sensitivity analyses with different assumptions on population-level Omicron variant relative frequencies, and in analyses using available individual-level data on infecting variant for a subset of the study population. Conclusions: Although clinical studies with matching viral genomic information should remain a priority, our approach combining publicly available data on variant frequency and a multi-country clinical characterisation dataset with more than 100,000 records allowed analysis of data from a wide range of settings and novel insights on real-world heterogeneity of COVID-19 presentation and clinical outcome. Funding: Bronner P. Gonçalves, Peter Horby, Gail Carson, Piero L. Olliaro, Valeria Balan, Barbara Wanjiru Citarella, and research costs were supported by the UK Foreign, Commonwealth and Development Office (FCDO) and Wellcome [215091/Z/18/Z, 222410/Z/21/Z, 225288/Z/22/Z]; and Janice Caoili and Madiha Hashmi were supported by the UK FCDO and Wellcome [222048/Z/20/Z]. Peter Horby, Gail Carson, Piero L. Olliaro, Kalynn Kennon and Joaquin Baruch were supported by the Bill & Melinda Gates Foundation [OPP1209135]; Laura Merson was supported by University of Oxford’s COVID-19 Research Response Fund - with thanks to its donors for their philanthropic support. Matthew Hall was supported by a Li Ka Shing Foundation award to Christophe Fraser. Moritz U.G. Kraemer was supported by the Branco Weiss Fellowship, Google.org, the Oxford Martin School, the Rockefeller Foundation, and the European Union Horizon 2020 project MOOD (#874850). The contents of this publication are the sole responsibility of the authors and do not necessarily reflect the views of the European Commission. Contributions from Srinivas Murthy, Asgar Rishu, Rob Fowler, James Joshua Douglas, François Martin Carrier were supported by CIHR Coronavirus Rapid Research Funding Opportunity OV2170359 and coordinated out of Sunnybrook Research Institute. Contributions from Evert-Jan Wils and David S.Y. Ong were supported by a grant from foundation Bevordering Onderzoek Franciscus; and Andrea Angheben by the Italian Ministry of Health “Fondi Ricerca corrente–L1P6” to IRCCS Ospedale Sacro Cuore–Don Calabria. The data contributions of J.Kenneth Baillie, Malcolm G. Semple, and Ewen M. Harrison were supported by grants from the National Institute for Health Research (NIHR; award CO-CIN-01), the Medical Research Council (MRC; grant MC_PC_19059), and by the NIHR Health Protection Research Unit (HPRU) in Emerging and Zoonotic Infections at University of Liverpool in partnership with Public Health England (PHE) (award 200907), NIHR HPRU in Respiratory Infections at Imperial College London with PHE (award 200927), Liverpool Experimental Cancer Medicine Centre (grant C18616/A25153), NIHR Biomedical Research Centre at Imperial College London (award IS-BRC-1215-20013), and NIHR Clinical Research Network providing infrastructure support. All funders of the ISARIC Clinical Characterisation Group are listed in the appendix. Editor's evaluation This manuscript compares COVID-19 mortality during the pre-Omicron and Omicron emergence periods in several countries. It finds evidence suggesting the Omicron variant was associated with lower mortality than previous dominant variants in multiple countries, though other factors than changing variant virulence might explain these observations, as discussed by the authors. This paper will be of interest to infectious disease scientists both for its content and its methods, as it validates that population-level variant frequency can be a good proxy for individual-level variant data to derive insights on variant biology with population data. https://doi.org/10.7554/eLife.80556.sa0 Decision letter Reviews on Sciety eLife's review process Introduction The emergence of novel SARS-CoV-2 variants represents a threat to the long-term control of COVID-19 (Fontanet et al., 2021). Whilst efforts to develop vaccines that protect against severe disease have been successful (Polack et al., 2020; Voysey et al., 2021; Baden et al., 2021), mutations in the viral genome that lead to ability to escape immunity, and increased transmissibility and/or clinical severity, either via intrinsic virulence or reduced vaccine effectiveness (Lopez Bernal et al., 2021), have the potential to cause substantial disease burden despite high vaccine coverage in many countries (Our World in Data, 2021). These concerns motivated the prompt reporting, initially from South Africa (Wolter et al., 2021; World Health Organization, 2022), of clinical characteristics of infection with the Omicron variant only weeks after its emergence (Wolter et al., 2022; Ulloa et al., 2022; Veneti et al., 2022), which provided key information for risk assessment and health policies worldwide. Early data from South Africa showed reduced severity of Omicron lineage BA.1 and similar results were reported in the United Kingdom and the United States (Wolter et al., 2022; Lewnard et al., 2022; Nyberg et al., 2022). However, the impact, in terms of clinical consequences (i.e. disease severity), of new variants has been shown to be context-specific, due to regional differences in disease epidemiology, including local circulation of previous variants and their cumulative incidences, variable vaccination coverages, and heterogeneity in population-level frequencies of risk factors (e.g. frequency of comorbidities) for severe disease and mortality. For this reason, international studies with standardised protocols are necessary to allow comparative assessments across different countries and epidemiological contexts. To understand the impact of the emergence of the Omicron variant of SARS-CoV-2 on the clinical epidemiology of COVID-19 at the global level, in this study, we report multi-country data, from all six World Health Organization regions, on clinical characteristics and outcomes of Omicron variant infections in hospitalised patients and compare with infections in patients admitted with other SARS-CoV-2 variants. For that, we use publicly available population-level data on relative frequencies of the Omicron variant to determine periods when infections were likely to be caused by Omicron versus other variants/lineages and compare infections descriptively and using multivariable statistical models. In addition, we present an analysis that only includes patients with individual-level data on the infecting variant and paired clinical information. Methods ISARIC clinical characterisation protocol Analyses presented in this manuscript use the ISARIC (International Severe Acute Respiratory and Emerging Infections Consortium) COVID-19 database, which includes prospectively collected data from countries where ISARIC partner institutions are located (see a global map of all ISARIC partner institutions here https://isaric.org/about-us/membership/). A full description of the data collection protocol and database can be found here https://isaric.org/research/covid-19-clinical-research-resources/. In short, data collection for this initiative was standardised, using the ISARIC case report forms, and pivoted into pandemic mode in January 2020 to enable rapid characterisation of the clinical presentation and severity of COVID-19. After the emergence of the Omicron variant, first reported in November 2021 (Viana et al., 2022), a call was launched to encourage international investigators partnering with ISARIC to rapidly share data on patients with confirmed or suspected COVID-19 to describe the clinical characteristics of Omicron variant infection in different settings; recommendations on possible hospitalised population sampling approaches were shared. Patients admitted to hospital from 1st October 2021 to 28th February 2022 were included in this analysis. More information on ISARIC can be found in ISARIC Clinical Characterisation Group, 2021; Hall et al., 2021; Reyes et al., 2022. Population-level SARS-CoV-2 variant data Two statistical analysis plans (SAPs) were developed in December 2021 with approaches to be used in the characterisation of Omicron variant infection. Analyses described in the first SAP required individual-level data on the clinical presentation and paired data on the variant causing the infection. In the second SAP, we used population-level frequencies of SARS-CoV-2 lineages to infer individual infecting variant during different time periods as Omicron or non-Omicron variants (Figure 1). Since individual-level data on the infecting variant were limited to a few countries, these data are presented for comparison with the analysis performed using population-level variant data. Figure 1 Download asset Open asset Population-level relative frequency of Omicron variant infections by country and time. Here, data aggregated by epidemiological week and country were used to calculate the proportions of infections caused by the Omicron variant. For analyses reported in the Results section, two epidemiological periods were defined: the first corresponds to the two months before the Omicron variant reaches a threshold frequency of 10% (blue area in each panel; the pre-Omicron period); the second period corresponds to the two months after Omicron variant frequency reaches 90% (red area in each panel; the Omicron period). Sensitivity analyses, using other relative frequencies for defining periods, are presented in the Appendix 1. Each panel presents data for a country (ISO3 code as title) contributing clinical data for this analysis; y-axes represent proportions in each epidemiological week (x-axes). Data for Laos are not shown as, at the time of the analysis, samples were not included in the database that informed population-level frequency of Omicron variant during the study period. In Pakistan, due to fluctuations in Omicron variant frequency in the dataset, study periods were not defined. More information on the spread of the Omicron variant in Laos and analysis of the clinical data from Pakistan are presented in the Appendix 1. For the analysis that required information on population-level variant frequency, for countries contributing clinical data to this analysis, data from the Global Initiative on Sharing All Influenza Data (GISAID) on each of the main SARS-CoV-2 variants were collated. These data were aggregated by sample collection date and variant using a computational pipeline available here: https://github.com/globaldothealth/covid19-variants-summary, (Dasgupta and Kraemer, 2022, copy archived at swh:1:rev:8adf2f756b182711ad1d0b8707c44d3703786d23). The GISAID data were downloaded on 11 April 2022; Pango lineage designation v1.2.133 was used (pango-designation, 2022). We used these data to define calendar time periods when the Omicron variant represented the majority of infections in each country, and also periods during which the Omicron variant represented only a small (<10%) fraction of infections. For each country, the period during which infections were assumed to be caused by other variants ended in the epidemiological week before the Omicron variant relative frequency crossed a low threshold percentage (e.g. 10%) (see Figure 1). The first epidemiological week when Omicron variant frequency, as a proportion of all circulating variants, was higher than a given threshold percentage (90% in analyses presented in the Results section and 80% in sensitivity analyses) was used as the start date of the period during which all admissions were considered to be caused by the Omicron variant. Note (i) that amongst different countries these two study periods started in different calendar weeks, depending on when the Omicron variant was introduced to the location and on the rate of its local spread, and (ii) that in this analysis all Omicron sub-lineages are included (e.g. BA.1.1, BA.2). Only patients admitted to hospital in the two months before country-level Omicron variant frequency reached the lower threshold and those admitted in the first two months after Omicron variant relative frequency reached 90% were included in the primary analysis; the reason for restricting the study population to those admitted during these time windows was to reduce confounding by unmeasured factors whose frequencies in the hospitalised population also changed over time and which might be associated with clinical outcomes. Statistical analysis We report the frequencies of symptoms, comorbidities and vaccination status stratified by country and time periods (before and after Omicron emergence). We also assessed the case fatality risk and the frequency of a composite outcome that combined death and invasive mechanical ventilation use during the two study periods; in this analysis, patients who were discharged from hospital before the end of the follow-up period used in the definition of the outcome (14 or 28 days) were assumed to have been alive at the end of that period. When estimating risk of death by day 14 after admission or onset of symptoms, whichever happened later, numerators were numbers of patients who died before or on day 14 after admission; denominators in this calculation included those who died by day 14, those discharged at any time during follow-up, and those who were followed at least for 2 weeks, regardless of final outcome, including those who died after 14 days. The same approach was used to analyse the 28 day fatality risk. Note that for 35.5% of patients admitted to hospital during the two study periods defined by Omicron variant frequency, date of onset of symptoms was missing; for these patients we assumed onset of clinical disease happened before admission – that is that these were not hospital acquired infections. Furthermore, for 7.2% of patients, outcome date (date of death or discharge or latest date with follow-up information) was missing and 0.4% had an outcome date that was earlier than date of admission or of symptoms onset; except for those who were discharged and had missing outcome date, these two groups of patients were not included in analyses on the frequencies of clinical outcomes but were included in analyses describing distributions of symptoms and comorbidities. As described in the Results section, some patients included in this study were admitted for treatment of a medical condition other than COVID-19 but tested positive incidentally during hospitalisation. We used mixed-effects logistic regression models to assess the association between study period, that is periods defined by the Omicron variant frequency at the population level, and 14-day death risk, adjusting for age, sex, and vaccination status. Age was included with the following categories: patients younger than 18 years, aged between 18 and 60 years, and older than 60 years. Random intercepts were used to account for potential variation in the risk of death between study sites in different countries. We also present models that adjust for the most commonly reported comorbidities; for each comorbidity included in the analysis, a binary variable was used to indicate presence or absence of the condition. Cox proportional hazards models on time to death, adjusted for age and sex and stratified by country and previous vaccination, were also fit; results of survival analyses are shown in the Appendix 1. Note that vaccination status was used as a binary variable in these models, without dose counts or timing of vaccination, and due to limited information on dates of doses we did not adjust for time since the most recent vaccination. R and Python were used for data processing and descriptive analyses (R Development Core Team, 2022; The pandas development team, 2020). Code used for analyses and aggregated data used to generate figures are available (ISARIC Data Platform, 2022) (see also Data availability statement). Stata 17 was used to fit mixed-effects logistic models and perform survival analysis. Results Description of study population and study periods Overall, 129,196 records from patients admitted to hospital between 1st October 2021 and 28th February 2022 were included in this analysis. Clinical centres in 30 countries contributed data (median 53 observations per country, interquartile range [IQR] 18–162); 11 countries contributed data on more than 100 hospitalised clinical cases (Appendix 1—table 1). A total of 54.0% and 42.6% of records were from South Africa and the United Kingdom, respectively. Appendix 1—table 2 and Appendix 1—table 3 show information on missing data for both symptoms and comorbidities. In addition to the clinical data contributed by the collaborating centres, population-level variant frequency data were used to define time periods when most infections in a country were assumed to be caused by Omicron versus other lineages. As presented in Figure 1, different countries reached the threshold relative frequencies of 10 and 90% of infections being caused by the Omicron variant at different times. Similar plots are presented in Appendix 1—figure 1 for other threshold frequencies. In Appendix 1—table 4, we list limitations in the use of these data to define time periods when infections were more likely caused by Omicron versus previous variants. Using information presented in Figure 1, 103,061 patients, from 28/30 countries, were admitted either in the two months before the Omicron variant represented 10% of infections at the country-level (N=22,921; henceforth, the pre-Omicron period) or in the two months after (N=80,140) the Omicron variant was responsible for at least 90% of the infections; for ease of reference, the latter period will be referred to as the Omicron period. Note that 12,085 patients were admitted during weeks between the end of the pre-Omicron period and the start of the Omicron period and are not included in analyses presented in the following subsections (Figure 2); and 12,560 records of patients admitted two months after Omicron variant represented 90% of infections were not analysed. All patients from South Africa, the United Kingdom and Malaysia were assumed to be SARS-CoV-2 positive, as this is one criterion for inclusion in their databases. Of the 2296 records from other countries, information on SARS-CoV-2 diagnostic testing was available for 1,999 observations; whilst patients with negative PCR test result (N=10) were excluded from the rest of the analysis, those with missing PCR data (N=297) were assumed positive (see Appendix 1—table 5 for distribution by country). Of note, clinical data from Laos were not included in comparative analyses as there was only limited evidence of increase in local Omicron variant relative frequency during the study period (additional information is provided in the Appendix 1). For Pakistan, population-level data available at the time of the analysis indicate increasing Omicron variant frequency during the study period, but the proportion of local infections caused by this variant fluctuated; analyses of clinical data from that country are described in the Appendix 1. Figure 2 Download asset Open asset Study flowchart. In this figure, we present the numbers of observations included in analyses in the different subsections of the Results section. The median (IQR) ages of patients during the pre-Omicron and Omicron periods were 62 (43 – 76) and 50 (30 – 72) years, respectively; however, country-specific medians suggest that the younger age of patients after Omicron variant emergence in the combined dataset is at least partially explained by an increase in the proportion of data contributed by South Africa, relative to the proportion of data contributed by other countries (Appendix 1—table 6). A total of 48.3% and 54.8% of patients admitted during these periods, respectively, were female. 5.2% and 9.1% of patients in the pre-Omicron and Omicron periods, respectively, had the date of disease onset after admission date. In some countries, information on whether COVID-19 was the main reason for hospitalisation was also collected: 70.1% (N=2248) and 69.0% (N=27,804) of patients during the pre-Omicron and Omicron periods respectively were admitted to hospital due to COVID-19; patients for whom this information was available were primarily from South Africa (94.9%). There was no consistent pattern of within-country changes related to this variable (Appendix 1—table 7). Of note, 465/36,761 (1.3%) individuals reported a history of previous SARS-CoV-2 infection before the acute episode leading to hospitalisation included in this analysis (128/15,563 [0.8%] and 337/21,198 [1.6%] in the pre-Omicron and Omicron periods, respectively). Temporal changes in frequencies of symptoms and comorbidities Figure 3 shows age distributions of hospitalised patients before versus after Omicron variant emergence; only countries with at least 50 observations in each period are included. Despite similar medians of age in the two periods within countries, in some, but not all, country-specific datasets, an increase in the proportion of the study population from younger ages was observed, although the number of patients in some age categories is small. Furthermore, there were differences between countries with regard to age distribution of cases, which could reflect either epidemiological differences between settings or else differences in recruitment of patients for this analysis. Figure 3 Download asset Open asset Age distributions by study period and country. Age distributions (x-axes show proportions; y-axes, age groups) when Omicron variant relative frequency was below 10% (blue bars) and when the frequency was 90% or higher (red bars). Data from different countries are shown in different panels; only countries with 50 or more records in each period are presented. Numbers of observations with age information are shown for each study period next to country names. The frequencies of the five most commonly reported symptoms and comorbidities in the combined (all countries) dataset during the two study periods are presented in Figure 4A and B, by country and study period. When analysing the combined dataset, there was a decrease in the percentage of patients with at least one of the comorbidities listed in Appendix 1—table 3 before versus during Omicron variant dominance (78.9% [N=15,574] and 59.6% [N=60,625], respectively); however, country-specific data show variable patterns (Appendix 1—table 8). With a total of 14 comorbidities being considered, median (IQR) numbers of comorbidity variables with non-missing information in the pre-Omicron and Omicron periods were 11 (0–12) and 9 (1 – 11), respectively. Whilst the directions of changes (increase or decrease) in frequencies of comorbidities were not consistent across countries, for many symptoms frequencies were lower during the Omicron period versus the pre-Omicron period. As can be seen in Appendix 1—figure 2, this pattern was consistent after stratifying frequencies of symptoms by age groups. The percentage of patients during the pre-Omicron period with at least one of the symptoms in Appendix 1—table 2 was 96.6% (N=11,683); this percentage was 88.6% (N=17,859) during the Omicron period (see Appendix 1—table 9 for country-specific numbers). These numbers refer to records from countries other than South Africa, where data on symptoms were not systematically available. The median (IQR) numbers of variables with non-missing data on symptoms were 14 (0–19) and 17 (0–19) for the pre-Omicron and Omicron periods, respectively. Figure 4 Download asset Open asset Frequencies of symptoms and comorbidities by study period and country. Frequencies of the five most common symptoms (A) and comorbidities (B) during the pre-Omicron (blue bars) and Omicron (red bars) periods. 95% confidence intervals are shown. Note that South Africa is included in panel B but not panel A. For panel (A), only data from the pre-Omicron period were used to identify the most frequent symptoms; for panel (B), as data on comorbidities were available in the two countries contributing most records, the United Kingdom and South Africa, and since their relative contributions to the study population changed in the two study periods, the dataset including both the pre-Omicron and Omicron periods was used to identify most common comorbidities. Only countries with at least 50 observations during each study period are included. For each symptom or comorbidity, whenever fewer than five observations without missing data were available, bars were not shown and the text ‘NS’ (not shown) was included. Vaccination history in hospitalised patients Data on vaccination status were available for 42,850/103,051 hospitalised patients (8,952 during the pre-Omicron period and 33,898 during the Omicron period). In Table 1, we present vaccination status for study participants in each of the two periods by country. As expected, there is considerable inter-country variation in the frequency of vaccination. Age-stratified vaccination frequencies are shown in Appendix 1—figure 3 and suggest increases in frequency of previous vaccination during the period after Omicron variant emergence. However, as shown in Appendix 1—figure 4, with population-level vaccination coverage from before Omicron variant emergence up to the end of February 2022, in many countries contributing data to this study there was an increase in vaccination coverage over time, including in the periods during and after the emergence of the Omicron variant. Note that 55.8% of vaccinated patients received two or more doses before hospital admission. Table 1 Vaccination status by country and study period. Data for period-country combinations with less than 10 observations are not presented. Data on vaccination status were not available for patients from Saudi Arabia. pre-Omicron periodOmicron periodCountry% VaccinatedTotal N% VaccinatedTotal NBrazil84.61387.933Canada32.25957.3686Colombia42.119-<10Estonia-<10Germany-<10-<10India34.82384.833Malaysia79.32980.010Nepal25.319039.3183Netherlands60.06051.051New Zealand5.934-<10Norway-<1082.245Philippines78.614-<10Portugal-<10-<10Romania-<1078.698South Africa15.1160527.924752Spain45.02070.955United Kingdom65.4686570.37846United States of America-<10-<10Argentina-<10Australia-<10Indonesia-<10Israel54.511Kuwait66.718Turkey74.127 Clinical outcomes Overall, 11,314 patients admitted during the two study periods died during hospitalisation: 8517/94,524 by day 14 after hospital admission or disease onset, whichever occurred latest, and 10,530/94,461 by day 28; 738 patients died after day 28 and 46 patients who died did not have an outcome date recorded. As explained in the Methods section, denominators for fatality risks included patients who were discharged or still in hospital by day 14 or 28. Median (IQR) times to death were 10 (5 – 17) and 6 (3 – 13) days for the periods before and after Omicron emergence, respectively; similar information, on time from admission or symptoms onset to death, stratified by country is shown in Appendix 1—table 10. In some countries (see Figure 5 for comparisons on 14-day fatality risk, and Appendix 1—figure 5 for comparisons using the 28-day period), during the Omicron period, a lower proportion of patients died during hospitalisation, compared to the period before Omicron emergence; in India, the opposite pattern was observed although numbers for that country were limited. Figure 5 Download asset Open asset Risk of death (y-axes) in the first 14 days after hospital admission or disease onset, whichever occurred latest, during the pre-Omicron and Omicron periods. In each panel, the x-axis shows countries (ISO3 codes are presented), with different periods represented by circles with different colours (blue circles for the pre-Omicron period; red circles, for the Omicron period). 95% confidence intervals are also presented. The top panel shows data for individuals of all ages; the bottom panels, data for patients aged less than 18 years, between 18 and 60 years, and older than 60 years. Only countries with at least 50 observations in both study periods are included in the figure; for panels presenting age-specific estima