Article Text

Original research
Evolution of serious and life-threatening COVID-19 pneumonia as the SARS-CoV-2 pandemic progressed: an observational study of mortality to 60 days after admission to a 15-hospital US health system
  1. Sudish C Murthy1,
  2. Steven M Gordon2,
  3. Ashley M Lowry3,
  4. Eugene H Blackstone1
  1. 1 Thoracic and Cardiovascular Surgery, Cleveland Clinic, Cleveland, Ohio, USA
  2. 2 Infectious Disease, Cleveland Clinic, Cleveland, Ohio, USA
  3. 3 Quantitative Health Sciences, Cleveland Clinic, Cleveland, Ohio, USA
  1. Correspondence to Dr Sudish C Murthy; murthys1{at}ccf.org

Abstract

Objective In order to predict at hospital admission the prognosis of patients with serious and life-threatening COVID-19 pneumonia, we sought to understand the clinical characteristics of hospitalised patients at admission as the SARS-CoV-2 pandemic progressed, document their changing response to the virus and its variants over time, and identify factors most importantly associated with mortality after hospital admission.

Design Observational study using a prospective hospital systemwide COVID-19 database.

Setting 15-hospital US health system.

Participants 26 872 patients admitted with COVID-19 to our Northeast Ohio and Florida hospitals from 1 March 2020 to 1 June 2022.

Main outcome measures 60-day mortality (highest risk period) after hospital admission analysed by random survival forests machine learning using demographics, medical history, and COVID-19 vaccination status, and viral variant, symptoms, and routine laboratory test results obtained at hospital admission.

Results Hospital mortality fell from 11% in March 2020 to 3.7% in March 2022, a 66% decrease (p<0.0001); 60-day mortality fell from 17% in May 2020 to 4.7% in May 2022, a 72% decrease (p<0.0001). Advanced age was the strongest predictor of 60-day mortality, followed by admission laboratory test results. Risk-adjusted 60-day mortality had all patients been admitted in March 2020 was 15% (CI 3.0% to 28%), and had they all been admitted in May 2022, 12% (CI 2.2% to 23%), a 20% decrease (p<0.0001). Dissociation between observed and predicted decrease in mortality was related to temporal change in admission patient profile, particularly in laboratory test results, but not vaccination status or viral variant.

Conclusions Hospital mortality from COVID-19 decreased substantially as the pandemic evolved but persisted after hospital discharge, eclipsing hospital mortality by 50% or more. However, after accounting for the many, even subtle, changes across the pandemic in patients’ demographics, medical history and particularly admission laboratory results, a patient admitted early in the pandemic and predicted to be at high risk would remain at high risk of mortality if admitted tomorrow.

  • COVID-19
  • Public health
  • Adult intensive & critical care

Data availability statement

Data are available on reasonable request. Data used for this study include human research participant data that are sensitive and cannot be publicly shared due to legal and ethical restrictions by the Cleveland Clinic regulatory bodies, including the institutional review board and legal counsel. In particular, variables such as date of testing or dates of hospitalisation are HIPAA protected health information and legally cannot be publicly shared. We will make our datasets available on request, under appropriate data use agreements with the specific parties interested in academic collaboration. Requests for data access can be made to Dr Misra-Hebert.

http://creativecommons.org/licenses/by-nc/4.0/

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

STRENGTHS AND LIMITATIONS OF THIS STUDY

  • Identifying risk factors for mortality using random survival forests machine learning.

  • Estimating time-varying instantaneous risk of mortality (hazard function) using multiphase hazards modelling.

  • Assessing change in risk by virtual-twins methodology.

  • We do not have a complete community denominator of SARS-CoV-2 positive cases (multiple screening sites, home testing), nor did drive-through screening sites draw blood from patients for routine laboratory tests. This prevented us from understanding epidemiologically why the characteristics of patients hospitalised with COVID-19—particularly their laboratory results at hospital admission—importantly fluctuated across the pandemic.

  • Criteria for admitting patients and caring for them after admission were not standardised because the study was conducted in real time during a pandemic, amid changing SARS-CoV-2 variants, development and dissemination of COVID-19 vaccines, changing methods for respiratory support and experimentation with therapeutics.

Introduction

Few anticipated that SARS-CoV-21 and its associated illness COVID-192 would still be an important public health problem several years after its discovery and clinical detection in late 2019.3 Identifying its genetic sequence soon after its clinical emergence,4 coupled with past experience with members of this viral family,5 led many to believe the disease would soon pass. Yet, despite rapid dissemination of information regarding its unique clinical characteristics and modes of transmission, early adoption of public health measures,6 and fairly rapid understanding of the pathogenesis of severe disease,7 a pandemic emerged, only partially abated by rapid manufacture of vaccines.8

Nearly 110 million persons have contracted SARS-CoV-2 in the USA, with more than a million related deaths.9 Even now in the endemic phase of the disease, with ever-emerging variants,10 widespread but incomplete population vaccination9 and availability of antiviral therapy,11 it is important to understand changes in the clinical profile of patients admitted with serious and life-threatening COVID-19. Specifically, because of the observed substantial reduction in hospital mortality from COVID-19, from about 20% early in the pandemic to about 5% 3 years later in the USA,12 we wondered whether this was the effect of SARS-CoV-2 variant evolution with lower virulence, dissemination of vaccines, herd immunity or development of effective therapeutics. Finally, as our health system followed patients systematically after hospital discharge, we were surprised by the number who died soon afterwards, leading us to wonder if the true magnitude of death from the illness was appreciated.

Therefore, our aim was to understand risk of mortality of patients admitted to our health system hospitals with serious and life-threatening COVID-19 pneumonia as the pandemic progressed by answering the following questions: (1) What was the risk of mortality of patients in-hospital and how long did high COVID-19 risk of mortality persist after hospital discharge? (2) At hospital admission, what were the predictors of mortality in the high-risk phase of mortality we identified? (3) Could the observed decrease in mortality within our hospital system as the pandemic progressed be explained by a change in the characteristics of patients at hospital admission and their physiological response to the virus (phenotype and endotype) or by something else, such as lower virulence of emerging virus variants, protective effect of vaccination or effectiveness of advanced therapies?

We limited our investigation to what was known about these patients with serious and life-threatening COVID-19 pneumonia at their initial encounter with our healthcare system—hospital admission—and their subsequent vital status, ignoring for this study their clinical progression after admission, adverse events while they were hospitalised, and therapies used during their hospital course, to help clinicians marshal appropriate resources at this first contact with these ill patients.

Patients and methods

Patients

From 1 March 2020 to 1 June 2022, 185 636 people tested positive for SARS-CoV-2 at community testing locations within the Cleveland Clinic Health System. The focus of this study is on 26 872 (15%) of these individuals who were admitted with serious and life-threatening COVID-19 pneumonia to 1 of 14 of our 15 Northeast Ohio and Florida hospitals (table 1, figure 1A).

Table 1

Characteristics and physiological response to SARS-CoV-2 of persons admitted to Cleveland Clinic Health System hospitals with serious and life-threatening COVID-19 pneumonia

Figure 1

Trends across the pandemic in number and mortality of patients hospitalised for severe and life-threatening COVID-19 pneumonia. Vertical lines demark epochs of dominant SARS-CoV-2 variants: ·Pre-1 April 2021: Alpha B.1.1.7 11 483 (43%) ·1 April 2021 to 1 July 2021: Alpha Q 1560 (5.8%) ·1 July 2021 to 1 December 2021: Delta B.1.617.2 and AY 5280 (20%) ·1 December 2021 to 1 February 2022: Omicron BA.1 6601 (25%) ·1 February 2022 to 21 March 2022: Omicron BA.1.1 785 (2.9%) ·21 March 2022 to 21 May 2022: Omicron BA.2 878 (3.3%) ·21 May 2022 to 1 June 2022: Omicron BA2.12.1 285 (1.1%). (A) Number of cases per week with each symbol a 1-week average. (B) Hospital mortality (per cent) by week with each symbol a 1-week average. Curved line is loess (locally weighted scatterplot) smoother. (C) Actuarial and simulated 60-day mortality. Black symbols represent observed 60-day unadjusted Kaplan-Meier actuarial estimated mortality for patients admitted each month, and black line is a loess smoother. The blue curve represents simulated 60-day mortality if all hospitalised patients across the pandemic were admitted on the same date from 3 January 2020 to 6 January 2022 based on the mortality model.

Data

Early in the pandemic, a COVID-19 Registry was established across our hospital system, capturing data for a common set of variables from electronic health records via automated feeds using standardised templates13 and manually by a study team.

Endpoint

The data-driven endpoint for this study was all-cause mortality from date of hospital admission to 60 days after discharge (hereafter, simply called ‘mortality’). Of patients discharged alive, 81% were followed at least 60 days.

Data analysis

SAS V.9.4 (SAS Institute) and R V.4.0.0 (R Foundation for Statistical Computing, Vienna, Austria) were used for data analysis. Profile of patients at hospital admission is depicted by temporal trends across the pandemic. To estimate the duration of the high-risk phase of mortality from the disease, we modelled time-varying instantaneous risk of mortality (hazard function) after hospital admission parametrically.14

R program randomForest-SRC15 was used for the analysis of time-related mortality. It incorporated demographics, medical history, home medications and vaccination status before SARS-CoV-2 infection (34 variables; online supplemental appendix E1), representing the phenotype of these patients before they contracted COVID-19. To these, we added on the patient level the date of hospital admission, symptoms and endotype as reflected in the results of routine laboratory tests at hospital admission, and on the population level the SARS-CoV-2 variant dominant in the community on the date of hospital admission (28 additional variables; online supplemental appendix E1). One thousand trees were grown with bootstrap samples of patients. Missing data were imputed on the fly during tree-building.16 Branch splits, using a log-rank rule, were based on a random subset of 8 variables with a terminal node size of 15.

Supplemental material

The importance of the association of variables with mortality was assessed by variable importance, hierarchically ordering variables by the magnitude of prediction error reduction.17 Partial dependency plots were generated to visualise the relation of variables to mortality.18

To determine whether the mortality of a given patient admitted anytime during the pandemic differed from that had the same patient been admitted earlier or later in the pandemic (virtual-twin analysis19), 60-day mortality was predicted for that patient as if the patient had been admitted on 1 March 2020 and then every day through 1 June 2022 by substituting for the actual date of admission all these counterfactual dates. These predictions were performed for every patient on every day, and the predicted mortality estimates from the entire cohort were averaged day-by-day across the pandemic. Risk-adjusted predicted mortality on 1 March 2020 and 1 June 2022 is presented with 15th and 85th percentiles, equivalent to ±1 SD of the individual patient predictions, with paired differences presented with 2.5th and 97.5th percentiles.

To investigate the difference between observed and predicted mortality across the pandemic, we performed a secondary random survival forest analysis using all variables in online supplemental appendix E1 except date of admission. From this analysis, the probability of 60-day survival was predicted for each patient. To document changes in patient risk profile across the pandemic, these survival probabilities were aggregated into seven categories: (1) ≥95% survival, (2) 90% to <95%, (3) 85% to <90%, (4) 80% to <85%, (5) 70% to <80%, (6) 60% to <70% and (7) <60%.

Patient and public involvement

There was no patient or public involvement in this study.

Results

Patient characteristics

Patients presented to our Health System with serious and life-threatening COVID-19 pneumonia in waves (figure 1A). The average age was 62±19 years (table 1 and online supplemental figure E1A). Early in the pandemic, more males than females were admitted, but this quickly approached a 50:50 ratio (online supplemental figure E1B). Trend in patient race exhibited a complex temporal pattern (online supplemental figure E1C). Many patients were on chronic medications, the most common being steroids (25%, table 1). Until mid-February 2021, no admitted patient had been vaccinated against COVID-19; by the study end, 5350 had been, 67% (524/787) of those admitted during May 2022.

Symptoms at hospital admission fluctuated across the pandemic, with fever declining from nearly 100% to a plateau of about 30%; cough progressively declined from about 75% to about 50% (online supplemental figure E2, A and B). Dyspnoea and gastrointestinal symptoms fluctuated about 40% (online supplemental figure E2, C and D).

Results of routine laboratory tests at hospital admission fluctuated across the pandemic (online supplemental figure E3). In particular, C-reactive protein declined early and again late in the experience (online supplemental figure E3-D), as did hepatic function enzymes (online supplemental figure E3-E).

Crude mortality

Crude hospital mortality fell from 11% among 229 patients admitted in March 2020 to 3.7% among 254 patients admitted in March 2022, a 66% reduction across the pandemic (p for trend <0.0001) (figure 1B). However, mortality continued after hospital discharge: 9.94% (68% CI 9.76% to 10.1%) at 30 days and 12.3% (CI 12.1% to 12.5%) at 60 days (figure 2A). Instantaneous unadjusted risk of death peaked at 0.446%/day (CI 0.436% to 0.457%) 9 days after hospital admission (figure 2B), falling thereafter and plateauing at 0.0366%/day (CI 0.0350% to 0.0383%) at 60 days, 4.0 times greater than an age-sex-race-matched US population reference of 0.0093%/day. Across the pandemic, 60-day mortality fell from 17% in May 2020 to 4.7% in May 2022, a 72% decrease (p for trend <0.0001) (figure 1C).

Figure 2

Time-related mortality after hospital admission for severe and life-threatening COVID-19 pneumonia: overall and according to SARS-CoV-2 variant and vaccination status. (A) Unadjusted mortality to 360 days after admission is depicted by Kaplan-Meier estimates (circles with vertical bars for 68% CIs equivalent to ±1 SE) superimposed on solid parametric estimates enclosed within a 68% confidence band. The number of patients remaining at risk at various intervals is shown beneath horizontal axis. Dash-dot-dash line represents mortality of an age-sex-race-matched US population. (B) Corresponding instantaneous risk of mortality, which peaks and falls to a constant level by about 60 days after hospital admission. (C) Unadjusted Kaplan-Meier estimates for each SARS-CoV-2 variant (see Key), with 68% confidence bars. (D) Unadjusted Kaplan-Meier estimates according to COVID-19 vaccination status with 68% confidence bars. (E) Partial dependency plot of risk-adjusted mortality according to SARS-CoV-2 variant from at-admission model. (F) Partial dependency plot of mortality according to COVID-19 vaccination status from at-admission model. Key: ·Black: Pre-1 April 2021: Alpha B.1.1.7 11 483 (43%) ·Red: 1 April 2021 to 1 July 2021: Alpha Q 1560 (5.8%) ·Orange: 1 July 2021 to 1 December 2021: Delta B.1.617.2 and AY 5,280 (20%) ·Blue: 1 December 2021 to 1 February 2022: Omicron BA.1 6601 (25%) ·Green: 1 February 2022 to 21 March 2022: Omicron BA.1.1 785 (2.9%) ·Purple: 21 March 2022 to 21 May 2022: Omicron BA.2 878 (3.3%) ·Brown: 21 May 2022 to 1 June 2022: Omicron BA2.12.1 285 (1.1%, too few to graph).

Actuarial mortality was highest for alpha variant B.1.1.7, lower for delta B.1 and omicron BA.1, and lowest for alpha-Q, omicron BA.1.1 and omicron BA.2 (figure 2C). Mortality was higher for COVID-19-vaccinated than unvaccinated patients (figure 2D).

Predictors of time-related mortality

The random survival forest model (prediction error 18%) identified older age as the most important and strongest predictor of mortality (figure 3), with an upward inflection beginning at about age 50 years, followed by a rapid increase with each year of age. After age, laboratory values at hospital admission emerged as highly important risk factors, including biomarkers of renal and liver dysfunction, a systemic inflammatory response, and haematological derangement (figure 3). After accounting for all these variables, neither SARS-CoV-2 variant (figure 2E) nor introduction of COVID-19 vaccination (figure 2F) remained importantly associated with mortality.

Figure 3

Illustration of strength and shape of a sample of risk-adjusted (partial dependency plots) associations with 60-day mortality of older age, higher blood urea nitrogen, lower albumin, higher aspartate aminotransferase, higher CRP and lower lymphocyte count. These are representative of demographics, renal and liver function, inflammatory response and haematological derangements found at hospital admission of patients with COVID-19. Accompanying these partial dependency plots at the foot of the graph is variable importance (VIMP) of each of these and others of the 13 most important variables in the at-admission model of mortality (error=18%). AST, aspartate aminotransferase; BUN, blood urea nitrogen; CRP, C-reactive protein; RBCs, red blood cells; RDW, red cell distribution width.

Changing patient profile and associated changes in mortality

Based on the at-admission model and virtual-twin simulation, 60-day mortality from date of admission had all 26 872 patients been admitted in March 2020 was predicted to be 15% (individual-patient 15th–85th percentiles 3.0%–28%) and fell to 12% (2.2%–23%) in May 2022 (paired difference 2.8% (2.5–97.5 percentiles 0.13%–8.1%), p (paired t-test) <0.0001). This represents a 20% decrease across the pandemic compared with the 72% decrease in observed 60-day crude mortality (figure 1C, blue line). Divergence of observed and risk-adjusted predicted mortality—a Simpson’s Paradox20—was in large part attributable to change across the pandemic in profile of demographics and pre-COVID-19 conditions (patient phenotype, online supplemental figure E1), but more particularly to change in profile of patients’ physiological response to the disease (patient endotype, online supplemental figure E3) as assessed by at-admission laboratory values. When coalesced into seven predicted risk groups, progressively fewer highest-risk patients were admitted over time, contributing to the decline in observed crude mortality and increase in survival (figure 4A, table 2 and online supplemental table E1). Simultaneously, there was a mild reduction of lowest-risk patients, but expansion of patients with intermediate-risk profiles.

Figure 4

Distribution of seven risk profiles of patients with serious and life-threatening COVID-19 pneumonia across date of hospital admission and observed survival for each profile. (A) Predicted 60-day survival using the at-admission model for mortality for survival profiles 1–7 are as follows: (1) ≥95%, (2) 90% to <95%, (3) 85% to <90%, (4) 80% to <85%, (5) 70% to<80%, (6) 60% to <70%, (7) <60%. Note particularly the gradual reduction in percentage of hospital admissions of the highest risk patients (pink and purple at bottom of graph) and increase in number of patients in lower risk categories at the top of the graph. (B) Survival stratified by predicted risk profile. Each symbol represents an event estimated by the Kaplan-Meier method, and vertical bars are 68% CIs equivalent to ±1 SE. Individual graphs are identified according to predicted survival profile.

Table 2

Profile of patients hospitalised with serious and life-threatening COVID-19 pneumonia according to 7 predicted 60-day survival categories by at-admission model: (1) ≥95%, (2) 90% to <95%, (3) 85% to <90%, (4) 80% to <85%, (5) 70% to <80%, (6) 60% to <70% and (7) <60% (see figure 4A and online supplemental table E1)*

Unadjusted actuarial survival of patients in these seven risk groups demonstrated large differences, particularly among the three highest risk groups (figure 4B), consisting of elderly patients, more often white, with more comorbidities, higher admission C-reactive protein and neutrophil counts, and lower lymphocyte counts (table 2). In contrast, the lowest risk group (34% of hospitalised patients) had a median age of 48 years and had fewer comorbidities and derangements in their admission laboratory profile.

Discussion

What this study confirms

  • Non-risk-adjusted crude hospital mortality among patients hospitalised with severe and life-threatening COVID-19 pneumonia trended markedly downward across the pandemic, confirming what others have reported.12

  • Non-risk-adjusted crude hospital mortality varied with SARS-CoV-2 variant, as shown by the US Centers for Disease Control and Prevention.21

  • Age was the strongest risk factor for mortality after hospital admission, followed by laboratory values at hospital admission.22

  • We identified several mortality risk groups with widely disparate phenotypic and endotypic characteristics, as did Cheyne et al, which they termed ‘phenoclusters.’23

  • Mortality by vaccination status among patients hospitalised with serious life-threatening COVID-19 pneumonia was related to demographics and disease severity.24

What this study finds new

  1. Patients hospitalised with serious and life-threatening COVID-19 pneumonia remain at high risk of mortality for at least 60 days after admission and thereafter are at higher risk of death than the age-sex-race-matched US population. This indicates that hospital mortality as a metric underestimates the lethality of COVID-19. Although time-related survival and mortality curves extending beyond hospital admission and discharge appear in a number of publications, to our knowledge, duration of the early risk of mortality from COVID-19 extending beyond hospitalisation has not been quantified and investigated. This study is unique in reporting time-related mortality to 60 days after hospital admission and its modulators (risk factors).

  2. Over the course of the pandemic, both phenotypic and endotypic characteristics of hospitalised patients have fluctuated, some widely. We uniquely show that the substantial reduction of mortality among patients hospitalised with COVID-19 is in large part related to these temporal changes in risk profile of these patients across the pandemic. Specifically, a patient predicted to be at high risk of mortality at the beginning of the pandemic will still be at high-risk tomorrow.

  3. The differences in mortality according to SARS-CoV-19 variants and COVID-19 vaccination status were uniquely found to be related to the changing at-admission patient profile across the pandemic.

Findings in context of substantial change in crude mortality

We were surprised by our finding that trends across the pandemic in the profile of patients admitted with serious and life-threatening COVID-19 appeared to account for the largest proportion of the substantial decline in crude mortality of these patients. Why did the patient profile change? We speculate that several mechanisms may have been at play.

Decrease in vulnerable populations

Changing phenotypes: Change in patient demographics, pre-COVID-19 medical conditions and symptoms of COVID-19 constitute the phenotype of our hospitalised patients.25 26 Phenotypic changes across the pandemic could have resulted from the deaths of people in the most vulnerable groups in the general population. The most vulnerable are the elderly, particularly those above age 50 years, as identified early in the pandemic.27 They also have the most chronic medical conditions.

Within the US general population, Hispanics have higher age-standardised crude mortality from COVID-19 than whites.28 We also observed in our study a widely fluctuating hospital admission pattern according to race that seemed to coincide somewhat with dominant viral variant; however, once adjusted for characteristics at hospital admission, neither race nor ethnicity was an important risk factor, as confirmed by others who have looked beyond phenotypic factors.29 30

Changing endotypes: Patients’ comorbid conditions reflected in our high-risk groups are associated with specific on-admission laboratory test results that differ from those without these conditions. These conditions, such as cardiorenal dysfunction, combined with systemic responses to the virus, such as biomarkers of inflammation, changed across the pandemic.31–34

SARS-CoV-2 variants

After accounting for patient phenotype and endotype at hospital admission, SARS-CoV-2 variants were not found to be risk factors per se. This does not mean that the population’s response to each viral variant is the same.35–37 As noted, the demographics of admitted patients fluctuated importantly across the pandemic, as did symptomatology, seemingly corresponding with dominant viral variants.38–40 Thus, one might infer that patient response to viral variants was well reflected in patient characteristics at admission, resulting in no remaining ‘independent’ information from knowing the viral variant.

Vaccination

Crude mortality of hospitalised patients who had been vaccinated against COVID-19 was higher than that of the unvaccinated well beyond hospital dismissal; yet, like viral variants, vaccination ‘disappeared’ as a risk factor once patient characteristics were accounted for. These patients were primarily within intermediate-risk groups, neither the lowest nor highest (see online supplemental table E2).

It is clear that despite the emergence of SARS-CoV-2 variants and declining effectiveness of the initial vaccine platforms to generate vaccine-induced immunity, protection against hospital admission for serious and life-threatening COVID-19 pneumonia and COVID-related death has remained robust.41–43 Prevention of severe illness in the vaccinated cohort—which began with the vulnerable elderly—development of herd immunity and early treatment and mitigation of infection44 are reducing the high-risk populations from the cohort of those hospitalised with COVID-19, which in turn reduces hospital mortality.45 46 For vaccinated patients who are hospitalised, their phenotype and endotype tended not to put them into the highest risk groups, the end result of which accounts for removing vaccination as a risk factor.

Selection

That patients with both the least and most physiological derangements diminished as the pandemic progressed suggests that the triage of patients to be admitted to the hospital changed. The rise of widespread telehealth during this time facilitated home monitoring and treatment of patients with antivirals. This permitted identifying (1) patients who demonstrated sufficient clinical improvement for management outside the hospital, (2) patients with clinical deterioration for hospital admission and (3) those for whom therapy was deemed futile for selective hospital admission.47 48 Severity of the disease can be estimated by symptom-complex phenotype (such as respiratory rate, oxygen saturation, blood pressure, heart rate, temperature, alertness), limited laboratory analysis (endotype: blood urea nitrogen, C-reactive protein) and demographics (age, sex).49–53

It has been suggested that although viraemia peaks within the first week of infection, peak host primary immunological response may not be evident for perhaps as long as a week later.54 Thus, we cannot say that vulnerability to the virus, differences in physiological response to viral variants, and vaccination had unimportant effects on mortality of hospitalised patients. Rather, the extent to which these and other factors altered characteristics of the population selected for hospital admission across the pandemic is one possible factor responsible for reduction in crude mortality.

Findings in context of other early-warning models

Our at-admission risk model incorporates standard demographic data, medical history, vaccination status, symptoms and routine admission laboratory data, and yet it seems to predict the outcome of hospitalised patients well and can be classified among the several early-warning models developed by others.52 55–60 Discrepancies among these models are in part related to what variables are considered. If only demographics and coexisting medical conditions are included, then the model will be less predictive of outcome than if derangements of routine clinical laboratory results are also included and, presumably, than if results of other more specific diagnostic tests are included.

Among models that account for routine clinical laboratory results, haematological and serum chemistry data are critical.61–63 During the ‘incubation’ period, and perhaps in the early phase of the disease, haematological constituents seem unperturbed.64 In patients whose disease progresses to hospitalisation, there are important haematological changes, a profound derangement of serum chemistries and an increase in inflammatory mediators and cytokines that give rise to a severe inflammatory response syndrome.65 66

The genesis of life-threatening illness seems to be the induction of multiple interleukins, macrophage proinflammatory proteins and tumour necrosis factor-alpha, thrombosis and endothelial dysfunction, and platelet activation, among innumerable other identified homeostatic derangements.67 At some transition point, which we suspect is just prior to hospital admission, individuals begin to declare their fate, as predicted by at-admission endotype. Lymphopenia and neutrophilia are important predictors of severe disease, with decreased lymphocyte count at admission being associated with intensive care unit (ICU) destination, and higher neutrophil count identified as a risk factor for death.68–70 Increased levels of inflammatory markers at admission, such as C-reactive protein, procalcitonin, lactate dehydrogenase and ferritin, portend an ominous prognosis.64 68 71 72 Many of these variables have been similarly shown to predict poor outcome in our study.

Post-hospitalisation mortality

Longitudinal studies of patients who have been hospitalised with COVID-19 demonstrate elevated risk that persists beyond hospital discharge.43 56 59 73 74 The shape of published Kaplan-Meier mortality curves is consistent with our findings of an early peaking high hazard phase that merges by 60 days into a lower late phase of risk. This pattern of risk is presumably related to persisting disease and debility postdischarge that may lead to mortality at home or at long-term nursing facilities.75 What has not been pointed out by others is the persistently elevated late constant hazard after the peaking phase. This elevated late hazard above that expected in the US population beyond 60 days76 suggests the chronic and insidious nature of late COVID-19.77 78

Limitations

This is a study of the COVID-19 pandemic experience in a single US multihospital system and may not be generalisable. It focuses only on patients hospitalised with COVID-19. Others from our hospital system have reported validated models for predicting hospitalisation for COVID-19.79 In addition, of the first 356 admissions for COVID-19, 49 (14%) were diagnosed with SARS-CoV-19 after hospital admission (11 after false-negative testing).80 As more rapid COVID-19 testing evolved, both delayed test results and false negatives fell to negligible levels. Information driving the decision of healthcare providers to admit a given patient is unavailable.

Decisions related to data elements to be recorded in the COVID-19 Registry were made near the beginning of the pandemic. Hence, some potentially important variables, such as laboratory tests for ferritin, were not included or even performed.

SARS-CoV-2 variant information was known only for hospitalised caregivers within our system. Therefore, as a surrogate, we bracketed eras of the prevalence of each variant in the general US population. Although we knew each patient’s COVID-19 vaccination status, vaccination as a variable is confounded by the timing of its introduction, number of initial doses and boosters received, and vaccine platform, information not captured in our dataset.

We restricted our analyses to the earliest available clinical information before and at hospital admission, in part to enable generating a useful model to predict mortality and direct allocation of resources—an ‘early-warning’ model. Thus, we did not consider additional hospital events, such as mechanical ventilation or extracorporeal membrane oxygenation, escalation of care to an ICU,81 subsequent laboratory test results during hospitalisation74 or therapeutics. Such longitudinal models are most useful when implemented to support real-time decision-making.74 82

The study was conducted in real time during the pandemic, amid changing variants,83 changing strategies for respiratory support, experimentation with therapeutics, innovation in patient isolation and telehealth. Consequently, care of patients during the pandemic was not standardised. Furthermore, we do not know why patients died. If, as is documented earlier, the virus is quickly cleared, then it appears that for many, deaths may have been due to a hyperinflammatory response with multisystem organ failure. What seemed to be heart attacks could have been acute coronary thrombosis.

Are the results we have reported unique to COVID-19, or might other serious infections follow the same pattern of mortality? We have no viral or other pulmonary infectious disease to use as a comparator for our study. Yang et al  59 found that prior emergency department risk scores for sepsis and bacterial pneumonia had low predictive value for COVID-19 pneumonia. Radiomics analysis of computed chest tomographic images appears able to differentiate COVID-19 from other viral pneumonias.84 Whether the heterogeneity found in COVID-19 that kills previously healthy young adults as well as the elderly with comorbid illness is unique remains unknown.

Conclusions and relevance

The evolution of the pandemic as addressed by us and others51 strongly suggests that COVID-19 remains lethal, although for a decreasing at-risk group of the general population. The observed mortality of patients hospitalised with COVID-19 in our health system has substantially decreased over time. That there appear to be fewer patients who manifest severe disease in response to exposure and infection suggests that something has changed to favourably affect disease outcome. Clearly, dissemination of vaccination programmes and herd immunity are at play, and there may be a change in viral virulence with the development of variants that have led to a smaller fraction of the general population presenting with advanced disease. Early use of antiviral agents may blunt the disease, and fewer people now need to seek advanced medical attention.

However, our sobering results indicate that it is unlikely that we are done with lethal COVID-19. Unfortunately, under-reporting of COVID-19-related lethality is likely when only in-hospital outcomes are examined. With continued emergence of variants having possibly different virulence, interval outcomes assessment is critical. The nimble adaptability of the SARS-CoV-2 virus and only modestly changing risk-adjusted fate of those who develop severe and life-threatening COVID-19 pneumonia mandate that continued individual and public health efforts remain in place. For those who develop severe and life-threatening COVID-19 pneumonia, a unique finding of our study is that a patient admitted early in the pandemic and predicted by our model to be at high risk of mortality would still be at high risk of mortality if admitted tomorrow.

Data availability statement

Data are available on reasonable request. Data used for this study include human research participant data that are sensitive and cannot be publicly shared due to legal and ethical restrictions by the Cleveland Clinic regulatory bodies, including the institutional review board and legal counsel. In particular, variables such as date of testing or dates of hospitalisation are HIPAA protected health information and legally cannot be publicly shared. We will make our datasets available on request, under appropriate data use agreements with the specific parties interested in academic collaboration. Requests for data access can be made to Dr Misra-Hebert.

Ethics statements

Patient consent for publication

Ethics approval

The Cleveland Clinic Institutional Review Board approved data collection, with informed consent waived (IRB Record 136 under IRB 20-283, 27 April 2020).

Acknowledgments

The authors thank Alex Milinovich and Greg Strnad for management of the COVID-19 Registry, Ben Kramer for data verification, Beth Lieber for statistical programming, Leslie Jelinski for reference management, Brian Kohlbacher for graphic design expertise and Tess Parry for editorial assistance.

References

Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

Footnotes

  • Contributors SCM: guarantor and principal investigator, conceived the study, wrote the protocol and submitted it to the HDISC COVID-19 subcommittee and subsequently to the Institutional Review Board, drafted and revised the manuscript, and performed several literature reviews. SMG contributed to and reviewed multiple iterations of the document and provided infectious disease insights, graphs and references. AML, guarantor, performed all statistical analyses, wrote extensive analysis reports, constructed figures and tables, verified the accuracy of all data and inferences related to the data and analyses, and reviewed all iterations of the manuscript. EHB, co-principal investigator, revised the original protocol, supervised data gathering and review, worked with AML in statistical analysis and selecting graphs, and worked with SCM on many drafts of the manuscript and all revisions thereof. SCM attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted; that the manuscript is an honest, accurate and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned have been explained.

  • Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

  • Competing interests None declared.

  • Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.