Methods Study design and patientsWe did a multicentre prospective observational cohort study at 247 acute general hospitals in England, Scotland, and Wales. We included adults (aged ≥18 years) from the ISARIC WHO CCP-UK cohort admitted to hospital during the first wave (between March 9 and Aug 2, 2020) with clinical signs and symptoms of COVID-19 or confirmed COVID-19 (by RT-PCR test) from assumed community acquired infection (appendix pp 2–3). The national strategy changed from containment to admission based on clinical need on March 12, 2020.14Department of Health and Social Care
COVID-19: government announces moving out contain phase into delay. Further information about the COVID-19 care model in the first wave in UK can be found in the appendix (pp 1–2). Community hospitals providing long-term treatment and residential mental health hospitals were excluded since patient populations were very different and were therefore not comparable. We excluded patients with nosocomial COVID-19 infection, who were defined as patients with onset of COVID-19 symptoms more than 5 days after they were admitted to hospital for a separate condition.Ethical approval for data collection and analysis by the ISARIC4C was given by the South Central-Oxford C Research Ethics Committee in England (reference 13/SC/0149) and by the Scotland A Research Ethics Committee (reference 20/SS/0028). Under the Control of Patient Information notice 2020 for urgent public health research, processing of demographic and routine clinical data from medical records for research does not require patient consent in England and Wales.15NHS Health Research Authority
Extension of COVID-19 COPI notice. In Scotland, a waiver for consent was obtained from the Public Benefit and Privacy Panel for Health and Social Care. The ISARIC WHO CCP-UK study is registered with the ISRCTN Registry, ISRCTN66726260, and has been designated as an Urgent Public Health Research Study by the National Institute for Health Research (NIHR). ProceduresData were extracted (cutoff date Sept 17, 2020) from routine health-care records and recorded in case report forms on the REDCap database. The ISARIC 4C Investigators collected information on key variables including patient characteristics, illness severity, level of respiratory support, COVID-19-specific treatments, and in-hospital mortality. Patient characteristics included age group (16Harrison EM Annemarie AB Barr B et al.Ethnicity and outcomes from COVID-19: the ISARIC CCP-UK prospective observational cohort study of hospitalised patients. index of multiple deprivation (derived from individual patient postal codes); health worker status (yes or no); and number of comorbidities (none, one, or two or more), measured by use of a modified Charlson Comorbidity Index (see appendix [p 3] for further details).Severity of illness at hospital admission (or within 24 h of admission) was assessed by use of the following physiological components of the ISARIC4C Mortality Score:13Knight SR Ho A Buchan I et al.Risk stratification of patients admitted to hospital with covid-19 using the ISARIC WHO Clinical Characterisation Protocol: development and validation of the 4C Mortality Score. respiratory rate (breaths per min); peripheral oxygen saturation on room air (SpO2); Glasgow coma scale score; serum urea concentration (mmol/L); and C-reactive protein concentration (mg/L). To capture patterns in access to hospital treatment, we calculated the number of days from symptom onset to hospital admission.

Patients were categorised as managed on the ward only or in critical care (ie, in the ICU or high dependency unit [HDU]) at any time (see appendix [pp 1–4] for more information on level of respiratory support). Maximum level of respiratory support received was classified as no respiratory support, oxygen (face mask, nasal cannulae, or high-flow nasal oxygen), non-invasive ventilation, and invasive mechanical ventilation.

For COVID-19-specific treatments, we only recorded whether patients had received corticosteroids (dexamethasone, hydrocortisone, methylprednisolone, or prednisolone), as these were the only treatments with recognised mortality benefit for COVID-19 in randomised controlled trials.9The RECOVERY Collaborative Group
Dexamethasone in hospitalized patients with Covid-19—preliminary report., 17The WHO Rapid Evidence Appraisal for COVID-19 Therapies (REACT) Working Group
Association between administration of systemic corticosteroids and mortality among critically ill patients with COVID-19. A meta-analysis.

The main exposure of interest was the week of admission to hospital, defined as the International Organization for Standardization date week (ie, the ordinal week of the year). To facilitate comparison across time periods, week of admission to hospital was also categorised into three equal time periods (time period 1 included weeks 11–17 [from March 9 to April 26, 2020]; time period 2 included weeks 18–24 [from April 27 to June 14, 2020]; and time period 3 included weeks 25–31 [from June 15 to Aug 2, 2020]).

 OutcomesThe primary outcome was the weekly in-hospital mortality at 28 days, defined as the proportion of patients who had died within 28 days of admission of all patients admitted in the observed week. Mortality was defined as an outcome of death or discharge to palliative care. The 28-day threshold aligns with the Public Health England definition of death due to COVID-19.18Covid-19: England comes into line with rest of UK on recording deaths. We included all patients who were admitted to hospital at least 6 weeks before data analysis to allow for 28 days follow-up.

Secondary outcomes were changes in patient demographics and illness severity in patients managed on the ward and in critical care units. Within critical care, we looked separately at changes in the proportion of patients receiving oxygen only, non-invasive ventilation, and invasive mechanical ventilation. Within ward care, we looked at changes in the proportion of patients receiving no respiratory support, oxygen only, and non-invasive ventilation. We performed two sensitivity analyses: a complete case analysis in which only patients with outcomes (survived, died, or ongoing care) were included and an analysis in which patients with missing outcomes were assumed to be survivors.

 Statistical analysis

Continuous data are presented as means (SDs) or medians (IQRs) depending on the distribution. Categorical data are presented as percentage frequencies. For univariable comparisons, we used Welch's t test, ANOVA, Mann-Whitney U test, or Kruskal-Wallis tests, according to data distribution. Categorical data were compared by use of χ2 tests. Counts and proportions for each of the exposure variables were calculated across the three time periods. The proportion of patients admitted each week and weekly in-hospital mortality at 28 days were stratified by each explanatory variable of interest. 95% CIs for these proportions were calculated by use of the exact method.

There were missing data because of the challenges of real-time data collection during a pandemic. Missing data are reported in the results section and in appendix (pp 37–39), and patterns of missing data were explored. Missing data for number of comorbidities were classified as none, for health-care worker status were classified as no, and for respiratory support received (ie, oxygen, invasive mechanical ventilation, and non-invasive ventilation) were classified as none. For the primary analysis, multiple imputation with chained equations was done for missing markers of illness severity. Ten sets, each with ten iterations, were imputed by use of 35 explanatory variables including outcomes. We did graphical checks of convergence. All analyses were done by use of imputed datasets (see appendix [p 3] for further details).For the primary outcome analysis, we excluded patients without an outcome date (classified as “survivors” in the sensitivity analysis results in the appendix [p 31). For the primary outcome, our modelling strategy was informed by a putative causal model (see the proposed directed acyclic graph in the appendix [p 26]). Using logistic regression, we specified three models exploring the association between admission week (as a continuous variable) and in-hospital mortality. A baseline model included adjustment for age, sex, and admitting hospital as a random effect. A second model accounted for known baseline confounders, including variables previously shown to be associated with in-hospital mortality (age, sex, number of comorbidities, index of multiple deprivation score, and severity of illness [respiratory rate, SpO2, Glasgow coma scale score, serum urea concentration, and C-reactive protein concentration]). In a third model, potential mediators were added to explore the effect of treatment (steroids and respiratory support, thus considering accrued clinical knowledge of respiratory support) on the association between week of admission and mortality (controlled direct-effect models). We extended this same model in a potential outcomes framework to perform a three-way decomposition mediation analysis using natural effects models.19Vansteelandt A Bakaert M Lange T Imputation strategies for the estimation of natural direct and indirect effects. We sought to control confounding between exposure and outcome, exposure and mediator, and mediator and outcome, and we carefully considered potential mediator-outcome confounders influenced by the exposure. Using standard frequentist approaches, we imputed unobserved nested counterfactuals with an outcome model to accommodate our nominal mediator (respiratory support). Exposure-mediator interactions were explored and a joint model was used to incorporate steroid use. Robust SEs (based on a Sandwich estimator) were generated, and the results were presented as a proportion mediated on the risk difference scale.

For the secondary outcomes, to better understand patterns of mortality for different levels of respiratory support, time-series data were modelled with Bayesian generalised additive models to allow for easy incorporation of multiply imputed datasets (see appendix [p 3] for more details).

All statistical analyses were done in R (version 3.6.3) and Stan (rstan 2.21.2 and brms 2.14.4) with the tidyverse, finalfit, brms, mgcv, mice, medflex, gridExtra and cowplot packages. This study is registered with the ISRCTN Registry, ISRCTN66726260.

 Role of the funding source

The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.

ResultsBetween March 9, and Aug 2, 2020, we recruited 80 713 patients from 247 acute hospitals in England, Scotland, and Wales, of whom 63 972 were eligible and included in the final cohort (table; appendix pp 5–7), representing approximately 48% of all hospital admissions in the UK during this time period.1Public Health England
Coronavirus (COVID-19) in the UK. Admissions peaked in late March and in early April for all age groups, and steadily decreased until the end of the study period (figure 1A). A total of 15 864 (29·0%) of 54 632 patients managed on wards died within 28 days of hospital admission compared with 3317 (35·5%) of 9340 patients in critical care (appendix p 24). Of all 63 972 patients, 40 449 (63·2%) had a documented positive COVID-19 RT-PCR test result (appendix pp 6–7).

TableBaseline characteristics of adult patients admitted to hospital with COVID-19, stratified by time (N=63 972)

Data are n (%), median (IQR), or mean (SD). Time period 1 was from March 9 to April 26, 2020; time period 2 was from April 27 to June 14, 2020; and time period 3 was from June 15 to Aug 2, 2020.

Figure 1Hospital admissions and in-hospital mortality between March 9 and Aug 2, 2020

Show full caption

(A) Daily adult COVID-19 admissions by age. Cases are stacked by age group. (B) Weekly unadjusted mortality in adult inpatients admitted to hospital with COVID-19. Error bars represent 95% CIs, calculated by use of an exact method. Dashed lines indicate three equal time periods (weeks 11–17, 18–24, and 25–31).

Most patients (55 562 [86·9%] of 63 972) admitted during the first wave were aged 50 years or older. There was an increase in the proportion of younger people (ie, those aged vs 486 [17·5%] of 2775 patients in time period 3; table, figure 2A). There were initially more men admitted than women (27 616 [58·2%] of 47 453 patients in time period 1 were men), but the proportions of men and women were similar from mid-April to Aug 2, 2020 (figure 2B). The cohort was multimorbid, with 25 275 (53·3%) of 47 453 patients in time period 1 having two or more comorbidities. The proportion of patients with two or more comorbidities increased over time (table, figure 2C; appendix p 7). Most patients were White, with an increasing proportion of south Asian patients and a decreasing proportion of patients from Black ethnic groups over time (figure 2D). In all time periods, the highest proportion of patients were in the most deprived quintile, and the proportion of patients in this deprivation quintile increased over time (figure 2E).

Figure 2Proportion of adults admitted to hospital with COVID-19 and inpatient mortality between March 9 and Aug 2, 2020, stratified by age (A), sex at birth (B), number of comorbidities (C), ethnic group (D), and deprivation quintile (E)

Show full caption

In A–E, all plots on the left show the proportion of adults admitted to hospital with COVID-19 by patient characteristics, and all plots on the right show unadjusted in-hospital mortality rates during the time period. In all plots on the left, the proportions of participants are stacked by characteristic. In E, the deprivation quintile ranges from 1 (most deprived) to 5 (least deprived). Missing data are excluded from the figure. Shaded areas represent 95% CIs.

Illness severity peaked from around March 30 to April 12, 2020 (weeks 14–16), when patients had faster respiratory rates, lower peripheral oxygen saturations on room air, and lower Glasgow coma scores, higher levels of acute kidney injury, and higher levels of inflammation at presentation to hospital than did patients admitted subsequently (figure 3). Patients presented later in their disease course at the beginning of the first wave compared with at the end of the first wave (median 4 days (IQR 0–7) in time period 1 vs 2 days (0–7) in time period 2 vs 3 days (0–7) in time period 3; table).

Figure 3Proportion of adults admitted to hospital with COVID-19 and inpatient mortality between March 9 and Aug 2, 2020, stratified by respiratory rate (A), peripheral oxygen saturation on room air (B), Glasgow coma score (C), urea concentration (D), and C-reactive protein concentration (E)

Show full caption

In A–E, all plots on the left show the proportion of adults admitted to hospital with COVID-19 by severity of illness at admission, and all plots on the right show the unadjusted in-hospital mortality rate each week during the time period. In all plots on the left, the proportions of participants are stacked by severity of illness at admission. Missing data are excluded from the figure.

At the peak of admissions in time period 1, 38 139 (80·4%) patients admitted to hospital received supplementary oxygen. The proportion of patients receiving supplementary oxygen reduced consistently over subsequent weeks to approximately 50% in patients admitted from July onwards (table).Most patients (54 632 [85·4%) of 63 972) admitted during the first wave were managed on the ward, with the proportion of patients admitted to critical care units peaking at the start of the first wave (7732 [16·3%] of 47 453 patients in time period 1; table; appendix pp 8,28). Compared with patients on wards, those in critical care units were younger (appendix pp 8–9) and more likely to be male (6433 [68·9%] of 9340 in critical care units vs 29 690 [54·3%] of 54  632 on wards; appendix pp 8–11). Patients with multiple comorbidities accounted for a substantial proportion of the total number of patients admitted (3733 [40·0%] of 9340 patients admitted to critical care units had two or more comorbidities vs 32 531 [59·5%] of 54 632 patients admitted to wards had two or more comorbidities; appendix pp 8–11). In critical care units, the proportions of younger patients and those with multiple comorbidities increased over time; a pattern that was also observed in patients admitted to wards (appendix pp 8–11).The level of respiratory support received reduced over time in patients admitted to critical care units and wards (appendix pp 8–11). In critical care units, the requirement for invasive mechanical ventilation declined over time (4958 [64·1%] of 7732 patients in time period 1 vs 98 [29·4%] of 333 patients in time period 3), the proportion of patients requiring non-invasive ventilation increased substantially from 1768 (22·9%) patients in time period 1 to 157 (47·1%) patients in time period 3 (appendix pp 8–9). By comparison, the proportion of patients admitted to wards on non-invasive ventilation remained low, decreasing from 3390 (8·5%) of 39 721 patients in time period 1 to 115 (4·7%) of 2442 patients in time period 3 (appendix pp 10–11). By the end of the first wave, 42·7% of patients (1184 of 2775) admitted to hospital received no respiratory support (1184 [48·5%] of 2442 patients admitted to wards received no respiratory support; table; appendix pp 10–11). More information on the characteristics of patients admitted to critical care units and wards, and the proportions of patients receiving respiratory support are included in the appendix (pp 8–23).The proportion of patients who received steroids increased from 7354 (15·5%) of 47 453 patients (2069 [26·8%] of 7732 patients in critical care units) in time period 1 to 910 (32·8%) of 2775 patients (223 [67·0%] of 333 patients in critical care units) in time period 3, mainly in patients receiving respiratory support (table; appendix pp 8–23, 29).Unadjusted weekly in-hospital mortality at 28 days declined from 32·3% (95% CI 31·8–32·7) in the period from March 9 to April 26, 2020, to 16·4% (15·0–17·8) in the period from June 15 to August 2, 2020 (figure 1B; appendix p 24). This reduction in weekly in-hospital mortality at 28 days did not differ substantially in the sensitivity analyses, in which patients without an outcome were reclassified as survivors (appendix p 31), nor when patients who died in hospital more than 28 days after admission were included (a further 5% of patients died after 28 days; see appendix p 32 for subgroup analysis).In-hospital mortality was higher with increasing age, increasing number of comorbidities, and male sex (figure 2). Over the course of the first wave, in-hospital mortality declined for all demographic categories, most notably in older patients (7867 [48·5%] of 16 233 patients aged ≥80 years died in time period 1 vs 239 [24·8%] of 965 patients aged ≥80 years in time period 3) and comorbid populations (figure 2; appendix p 7). Markers of increased severity of illness at presentation to hospital were associated with increased in-hospital mortality. In-hospital mortality declined for all markers of severity of illness over time (figure 3) and for patients treated on wards and in critical care units (appendix pp 24, 28, 29).There was a 35% reduction in the odds of in-hospital mortality per 7-week time period (odds ratio [OR] 0·65 [95% CI 0·63–0·67], pfigure 4). After adjustment for age, sex, deprivation, and hospital, the odds of in-hospital mortality per 7-week time period was 0·58 (0·56–0·60; pfigure 5). There was a significant interaction between respiratory support and week of admission (p

Figure 4OR for in-hospital mortality for week of admission per 7-week period

Show full caption

In-hospital mortality unadjusted for week of admission (A); adjusted for age, sex, deprivation, and hospital (B); adjusted for age, sex, deprivation quintile, severity of illness (respiratory rate, oxygen saturations, Glasgow coma score, serum urea concentration, and C-reactive protein), and number of comorbidities (C); and adjusted for age, sex, deprivation quintile, severity of illness, number of comorbidities, and potential mediators (maximal level of care, respiratory support, and treatment with steroids). Error bars represent 95% CIs. OR=odds ratio.

Figure 5Causal graph with natural effects models mediation analysis

Show full caption

The OR for the total natural indirect effect was 0·95 (95% CI 0·94–0·95, p<0·0001; percentage of effect through indirect path 22·2%; joint mediators) and for the pure natural direct effect was 0·84 (0·82–0·87, p<0·0001; percentage of effect from direct path 77·8%). IMD=index of multiple deprivation. YXM and YM=unmeasured confounders. OR=odds ratio. *Arrows from steroid treatment mediator confounders are not shown for clarity.

There were substantial reductions in unadjusted in-hospital mortality between time period 1 and time period 3 in patients receiving no respiratory support (1339 [14·0%] of 9564 patients in time period 1 vs 71 [6·0%] of 1183 patients in time period 3), oxygen only (9496 [35·1%] of 27 054 patients admitted to wards in time period 1 vs 243 [21·3%] of 1141 patients admitted to wards in time period 3; and 228 [22·7%] of 1004 patients admitted to critical care units in time period 1 vs ten [12·8%] of 78 patients admitted to critical care units in time period 3), and non-invasive ventilation in critical care units (587 [33·2%] of 1768 patients in time period 1 vs 39 [24·8%] of 157 patients in time period 3) across all age groups (appendix pp 8–23). However, in-hospital mortality remained persistently high for patients receiving invasive mechanical ventilation (2034 [41·0%] of 4961 patients in time period 1 vs 41 [41·8%] of 98 patients in time period 3) and non-invasive ventilation on the ward (1626 [48·0%] of 3387 patients vs 51 [44·3%] of 115 patients). These differential changes in in-hospital mortality persisted after adjustment for demographic and severity of illness variables (appendix p 30).Discussion

Overall in-hospital mortality within 28 days of admission substantially decreased during the first wave of the pandemic in the UK. At the peak of admissions in late March and early April, 2020, patients were substantially more unwell at presentation to hospital and presented later from their onset of symptoms than those presenting to hospital at later months. There was a reduction in the level of respiratory support received; use of invasive ventilation reduced over time, and the proportion of patients receiving non-invasive ventilation increased. By late June and early August, almost half of patients admitted required no supplementary oxygen. The reduction in in-hospital mortality during the first wave was observed across all demographic groups, and was not fully accounted for by changes in the case-mix or a reduction in illness severity at hospital admission. One-fifth of the reduction in in-hospital mortality could be accounted for by changes in treatment, including respiratory care and steroid treatment.

ISARIC4C has recruited patients from hospitals across the UK, accounting for approximately two-thirds of patients admitted to hospital in the UK with COVID-19 in the first wave. Data were collected from arrival in the Emergency Department to discharge for patients managed both on wards and in critical care units, enabling us to review admission and mortality rates in whole hospitals rather than just in critical care units.

We showed a reduction in in-hospital mortality during the first wave that cannot be fully explained by baseline patient demographics or measured presenting severity of illness markers. These trends are consistent with those observed in New York hospitals,20Horwitz LI Jones SA Cerfolio RJ et al.Trends in COVID-19 risk-adjusted mortality rates. where mortality also significantly and progressively declined between March and August, 2020. Mortality rates in critical care units in the UK have also reduced over this time.21Doidge JC Mouncey PR Thomas K et al.Trends in intensive care for patients with COVID-19 in England, Wales and Northern Ireland., 22Dennis JM McGovern AP Vollmer SJ Mateen BA Improving survival of critical care patients with coronavirus disease 2019 in England. Most patients admitted to hospital during the first wave were older (ie, median age 70 years), had two or more comorbidities, and were of White ethnicity, and in-hospital mortality was highest in these groups. The case-mix has changed over the course of the pandemic, with an increase in the proportion of patients younger than 50 years and female patients who, both in our study and in other studies, have lower mortality rates than older groups and male patients.23WHO
Coronavirus disease (COVID-19)., 24Bonanad C Garcia-Blas S Tarazona-Santabalbina F et al.The effect of age on mortality in patients with COVID-19: a meta-analysis with 611,583 subjects. However the declines in mortality were observed in all age groups, ethnic groups, and in both sexes. Shielding of susceptible groups was formally introduced in the UK on March 23, 2020,25Department of Health and Social Care
Shielding update. and patients presenting before shielding was introduced might have been more susceptible but not identifiable in our dataset. Severity of illness at presentation to hospital decreased, and the proportion of patients requiring no respiratory support increased at the end compared with at the start of the first wave. SARS-CoV-2 is transmitted predominantly by respiratory droplets; therefore, physical distancing, lockdown (between March 23, 2020, and July 4, 2020),26UK Government
Prime Minister's statement on coronavirus (COVID-19): 23 March 2020. and widespread adoption of face masks could have reduced viral load (ie, the infectious dose) at the point of transmission,27Goyal A Reeves DB Cardozo-Ojeda EF Mayer BT Schliffer JT Slight reduction in SARS-CoV-2 exposure viral load due to masking results in a significant reduction in transmission with widespread implementation. thus reducing severity of illness in infected patients.28Pujadas E Chaudhry F McBride R et al.SARS-CoV-2 viral load predicts COVID-19 mortality. Compared with March and April, 2020, patients presented earlier in the disease course in June and July, 2020, and length of stay of non-survivors in critical care units increased, consistent with patients presenting earlier in their illness and fewer in extremis. Our data do not include community factors; however, it is possible that changes in health-seeking behaviour might have led to patients attending hospitals more easily and earlier.Hospital admissions in the UK peaked at approximately 3000 patients admitted per day in early April, 2020, with the ICU caseload peaking shortly after this point1Public Health England
Coronavirus (COVID-19) in the UK., 2COVID-19: admissions to Hospital – Update. before gradually declining to a plateau of approximately 100 patients admitted per day by July, 2020.1Public Health England
Coronavirus (COVID-19) in the UK. The UK has a small health-care workforce, with few hospital and critical care beds per capita relative to other high income countries.29Resilient health and care. Learning the lessons of COVID-19 in the English NHS., 30Organisation for Economic Co-operation and Development
Intensive care beds capacity. At the start of the pandemic, the UK had high occupancy of hospital beds and little spare capacity.29Resilient health and care. Learning the lessons of COVID-19 in the English NHS. However, during the rapid response to COVID-19, an additional 2711 beds compatible with mechanical ventilation across England were made available, reflecting a 53% increase in capacity.31Mateen BA Wilde H Dennis JM et al.A geotemporal survey of hospital bed saturation across England during the first wave of the COVID-19 pandemic. National reported occupancy in critical care units never exceeded 62%, although local peaks in critical care unit occupancy were much higher than this value, with 52 hospitals across England reaching 100% occupancy at one or more points during the first wave of the pandemic.30Organisation for Economic Co-operation and Development
Intensive care beds capacity., 31Mateen BA Wilde H Dennis JM et al.A geotemporal survey of hospital bed saturation across England during the first wave of the COVID-19 pandemic. The effect of increased patient numbers could have been different depending on the type and size of the hospital, with smaller hospitals reaching full capacity sooner than larger hospitals. Even before the pandemic, critical care capacity strain was associated with increased mortality.32Harris S Singer M Sanderson C Grieve R Harrison D Rowan K Impact on mortality of prompt admission to critical care for deteriorating ward patients: an instrumental variable analysis using critical care bed strain. The rapid increase in critical care beds required redeployment of non-critical care staff, and in some UK regions, increased nursing staff-to-patient ratios, which could have influenced early patient outcomes.33Sprivulis PC Da Silva JA Jacobs IG Jelinek GA Frazer ARL The association between hospital overcrowding and mortality among patients admitted via Western Australian emergency departments.In the first time period late March, early April, we found that a higher proportion of patients were admitted to critical care units than later. Patients in critical care units at this time were considerably younger than those on wards; even so, in-hospital mortality was much higher for COVID-19 than for other severe acute respiratory infections, such as viral pneumonia.7Intensive Care National Audit and Research Centre
Reports. ICNARC report on COVID-19 in critical care: England, Wales and Northern Ireland. During the peak admission period, the proportion of patients aged older than 80 years and the proportion of patients with two or more comorbidities admitted to critical care units was lower than after this time point; however, this pattern reflected the demographic of patients admitted to wards during the peak admission period.A fifth of the reduction in in-hospital mortality in our study can be explained by changes in respiratory support and steroid treatment, together with associated changes in clinical decision making associated with respiratory support. At the beginning of the pandemic, much attention was given to early intubation and to different ventilator management practices based on different presumed phenotypes. These practices changed as clinicians started to manage acute respiratory distress syndrome (ARDS) from COVID-19 as routine ARDS. The proportion of patients receiving invasive ventilation reduced over time; however, in-mortality remained persistently high after adjusting for patient demographics and illness severity, as shown in our study. It would be unwise to interpret this association as causal, as use of invasive mechanical ventilation is reserved for patients with the most severe illness, and there was an overall reduction in mortality in patients with COVID-19 admitted to critical care units in our study. There are several potential explanations for this finding. First, the change in case-mix might not have been adequately captured by multimorbidity and age; a higher proportion of older patients (ie, those older than 70 years) and patients with comorbidities were ventilated later in the first wave, potentially at a time when there was more critical care capacity than at the start of the first wave. Second, critical care practices changed, with increasing use of non-invasive ventilation over time, and only patients presenting in extremis or those who did not respond to a trial of non-invasive ventilation received invasive mechanical ventilation. This change in practice could be partly due to the changing case-mix, but also due to increasing clinician familiarity with the use of non-invasive ventilation, and an improving ability to identify which patients might benefit from this intervention.34Montgomery CM Humphreys S McCulloch C et al.Critical care work during COVID-19: a sociological analysis of staff experiences. It is possible that patients admitted early in the first wave who received invasive mechanical ventilation would have received non-invasive ventilation if they had been admitted later in the first wave, and that they would have survived regardless of the mode of ventilation. Therefore, compared with patients who received invasive mechanical ventilation early in the first wave, those admitted later who received invasive mechanical ventilation were a more severely ill population who had failed to respond to treatments and would die if not offered invasive mechanical ventilation. This notion is supported by the significant increase in the effect of respiratory care over time; allocation to respiratory support was linked to better prediction of outcome by clinicians over time in our study. Ongoing trials comparing the use of non-invasive ventilation and invasive mechanical ventilation in critically ill patients with COVID-19 will help to overcome this selection bias and confounding by indication to ascertain wheth