Clinical Characteristics and Predictors of Outcomes of Hospitalized Patients With Coronavirus Disease 2019 in a Multiethnic London National Health Service Trust: A Retrospective Cohort Study

Abstract Background Emerging evidence suggests ethnic minorities are disproportionately affected by coronavirus disease 2019 (COVID-19). Detailed clinical analyses of multicultural hospitalized patient cohorts remain largely undescribed. Methods We performed regression, survival, and cumulative competing risk analyses to evaluate factors associated with mortality in patients admitted for COVID-19 in 3 large London hospitals between 25 February and 5 April, censored as of 1 May 2020. Results Of 614 patients (median age, 69 [interquartile range, 25] years) and 62% male), 381 (62%) were discharged alive, 178 (29%) died, and 55 (9%) remained hospitalized at censoring. Severe hypoxemia (adjusted odds ratio [aOR], 4.25 [95% confidence interval {CI}, 2.36–7.64]), leukocytosis (aOR, 2.35 [95% CI, 1.35–4.11]), thrombocytopenia (aOR [1.01, 95% CI, 1.00–1.01], increase per 109 decrease), severe renal impairment (aOR, 5.14 [95% CI, 2.65–9.97]), and low albumin (aOR, 1.06 [95% CI, 1.02–1.09], increase per gram decrease) were associated with death. Forty percent (n = 244) were from black, Asian, and other minority ethnic (BAME) groups, 38% (n = 235) were white, and ethnicity was unknown for 22% (n = 135). BAME patients were younger and had fewer comorbidities. Although the unadjusted odds of death did not differ by ethnicity, when adjusting for age, sex, and comorbidities, black patients were at higher odds of death compared to whites (aOR, 1.69 [95% CI, 1.00–2.86]). This association was stronger when further adjusting for admission severity (aOR, 1.85 [95% CI, 1.06–3.24]). Conclusions BAME patients were overrepresented in our cohort; when accounting for demographic and clinical profile of admission, black patients were at increased odds of death. Further research is needed into biologic drivers of differences in COVID-19 outcomes by ethnicity.

The United Kingdom (UK) has been the third worst affected country in the world by coronavirus disease 2019 (COVID- 19), with a reported death toll of 46 413 as of 6 August 2020 [1]. Most deaths have occurred in London [2], a densely populated, multicultural capital city where >40% of residents identify themselves as belonging to a black, Asian, or other ethnic minority (BAME) [3].
Previous studies have found that age, sex, and comorbidities including cardiovascular disease are associated with poorer COVID-19 hospitalization outcomes [4][5][6][7][8]. However, robust analyses adjusting for confounding factors and evaluating the association of the physiological and laboratory characteristics on admission with clinical outcomes are lacking. Such studies are urgently needed to allow frontline clinicians to tailor their management decisions and hospital policymakers to efficiently plan their response to  Data from UK admission to intensive care units (ICUs) and from the Office for National Statistics suggest that BAME groups are more frequently affected with COVID-19 than white patients [9,10]. This is largely thought to be due to economic deprivation driving an increased risk of COVID-19 infection and death [11][12][13]. These reports, however, have largely relied on analysis of administrative-level data from linked electronic records, and lack clinical granularity to explore patients' in-hospital features and their outcomes. Using comprehensive clinical patient-level data from consecutive patient admissions to the Imperial College Healthcare National Health Service Trust (ICHNT), one of the largest hospital Trusts in England serving a multiethnic population of >600 000 in North West London [14], we aimed to (1) describe baseline characteristics and outcomes for patients hospitalized with laboratory-confirmed severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection early in the pandemic; (2) evaluate the proportion of patients hospitalized at ICHNT for COVID-19 who are from BAME groups and whether ethnicity is associated with mortality; and (3) evaluate clinical and laboratory observations on admission associated with mortality. With a wealth of clinical trials underway, baseline analyses such as this will serve as a useful comparison for the results of interventional studies.

Study Setting
We performed a retrospective cohort study at ICHNT, including all admissions with reverse-transcription polymerase chain reaction (RT-PCR)-positive SARS-CoV-2 infection between 25 February and 5 April 2020. The cohort opened at the time of admission and censoring was done as of 1 May 2020. We excluded those who did not initially present with symptoms suggestive of COVID-19, those with clinical suspicion of COVID-19 but negative RT-PCR, those who did not require hospitalization, and transfers from other hospitals. Where a patient had 2 or more RT-PCR positive results, we included the first episode that required hospitalization.

Ethical Approvals
The study was approved by the ICHNT clinical governance team. As we report on routinely collected nonidentifiable clinical audit data, no ethical approval was required under the UK policy framework for health and social care.

Data Collection
Individual patient electronic medical records were used to extract demographic, clinical, laboratory, radiological, comorbidity, and outcome data including need for ICU admission and invasive ventilation, using a standardized data template. Demographic characteristics included patients' sex, age, ethnicity, and whether they were a health worker. Pre-existing comorbidities were recorded individually and used to calculate the Elixhauser comorbidity score, a points-based system validated in the UK to evaluate risk of hospital mortality [15,16]. Ethnicity is normally registered on initial presentation to the emergency department for all patients at ICHNT. Given challenges with capacity and infection control at ICHNT during the present epidemic, this variable was often recorded as "other ethnicity. " For completeness, these entries were checked against London Ambulance Service records, previous clinic letters and GP referral letters, and reclassified as appropriate. Where ethnicity was still unable to be confirmed, we report this as unknown.
We recorded clinical observations at the time of initial presentation to hospital, including New Early Warning Score (NEWS-2), respiratory rate, mean arterial blood pressure, temperature, oxygen saturation (SaO 2 ), and oxygen requirements. SARS-CoV-2 testing was performed by RT-PCR of specimens collected by nasopharyngeal swabs. Laboratory results of serum creatinine, urea, glucose, and lactate were only included if performed within 4 hours of admission. All other blood tests and chest radiography were included if performed within 24 hours of admission. Chest radiographic findings were reported as per recommendations from the British Society of Thoracic Imaging guidelines (Supplementary Data) [17].

Outcomes of Interest and Statistical Analysis
We assessed the primary outcome of vital status as discharged alive or died and an intermediate outcome of clinical deterioration, defined as either requiring a ≥60% fraction of inspired oxygen (FiO 2 ) to maintain SaO 2 >94%, invasive ventilation, or admission to ICU. We used standard χ 2 , Student t, or Wilcoxon rank-sum tests, as appropriate, to describe the prevalence of covariates at admission among patients with a completed outcome.
We further performed adjusted and unadjusted logistic regression to assess (1) predictors of COVID-19 hospital mortality and (2) differences in mortality by ethnicity. Numeric variables (eg, age, hemoglobin, creatinine) were treated as continuous if their density distribution plot showed a normal distribution in either the natural or logarithmic scale, else they were coded into categorical. Variables with a P value below .157 in unadjusted regression were selected for adjusted regression, as per the Akaike information criterion [18]. In the adjusted regression model, variables were considered statistically significant with a P value <.05 and thus kept as final candidates in the prediction model. Variables with >20% missing values were excluded from regression analysis.
We evaluated time from hospital admission to clinical deterioration and final outcome accounting for competing risks and right-censored data (ie, patients still in hospital at the time of censoring) using the Nelson-Aalen and Kaplan-Meier estimators, respectively (Supplementary Data). The effect of selected covariates on the hazard of clinical deterioration and death was assessed using Cox proportional hazard models. The proportional hazards assumptions were evaluated using Schoenfeld residuals [19].
All analyses were performed using R Studio (version 1.2.5033). The funders of this study had no role in the study design, data collection, analysis, interpretation, or reporting.

Description of Cohort
Seven hundred fifty-six patients had a positive SARS-CoV-2 nasopharyngeal swab and were hospitalized at ICHNT between 25 February and 5 April 2020, of which 142 were excluded as their clinical presentation was for other non-COVID-19 symptoms (but subsequently tested positive for SARS-CoV-2 during hospitalization). Of 614 patients included in analyses, 559 had completed outcomes and 55 (9%) were still hospitalized at the date of censoring ( Figure 1). Their median age was 69 years (interquartile range [IQR], 25), 382 (62%) were male, 235 (38%) were white, 244 (40%) were from a BAME group, and for 135 (22%) the ethnic background was unknown.
The mean time from symptom onset to hospital admission was 7.2 days (standard deviation, 5.6 days), with cough, fever, and shortness of breath being highly prevalent at presentation; 192 (31%) had gastrointestinal symptoms. On initial clinical assessment, 196 patients (32%) had NEWS-2 of ≥7, with 420 (69%) having an SaO 2 <95% on room air or requiring supplemental oxygen to maintain normal SaO 2 . Among patients with full blood count available (n = 605), 262 (43%) had anemia <130 g/L, 385 (64%) had lymphopenia <1.1 × 10 9 /L, and 97 (16%) had thrombocytopenia <130 × 10 9 /L. Biochemical and inflammatory marker abnormalities were also found in a high proportion of patients, most notably a reduced glomerular filtration rate (eGFR) and a raised C-reactive protein (CRP) ( Table 1). Abnormalities in their inflammatory and cardiac injury markers (eg, D-dimer, troponin) were also highly prevalent, albeit these investigations were significantly underrecorded and thus not included in regression analyses (Table 1).
Nearly half of the cohort (n = 276) received at least 60% oxygen during their admission and 80 (13%) received invasive ventilation. Of patients with completed outcomes, 381 (68%) were discharged alive and 178 (32%) patients died in hospital. The median length of hospital stay, accounting for patients with pending outcomes, was 7 days (IQR, 6-8 days) and the median time to clinical deterioration was 15 days (IQR, 8-32 days) (Supplementary Figure 2).

Predictors of Mortality
Out of those with completed outcomes (n = 559 patients), the median age, sex, and comorbidity profiles were significantly different between those who died and those who were discharged alive (  Figure 3). Other admission   In adjusted logistic regression, clinical and laboratory observations with an increased odds ratio of mortality that remained statistically significant were fever (≥38°C) on admission, having a SaO 2 <95% while on supplementary oxygen or if >60% FiO 2 was administered on admission, elevated white cell count, thrombocytopenia (continuous), hypoalbuminemia (continuous), and a reduced eGFR (  Table 5).

Outcomes by Ethnicity
Two hundred thirty-five (38%) patients were of white background, 133 (22%) were black, 94 (15%) Asian, 17 (3%) "other, " Analyzed with Kaplan-Meier estimator. Values for "All" account for censoring of those with uncompleted outcomes. Values for "Died" and "Survived" restricted to those with completed outcomes. P values as assessed with Cox proportional hazards models and Schoenfeld residuals. For "Survived, " estimation of time to clinical deterioration was not possible with these methods, as <50% in the subgroup had the event of interest. c Chest radiograph classification as per the British Society of Thoracic Imaging.  (Table 3). However, white patients were significantly older, with an average age of 70.6 years  other" ethnicity) (P < .01) and also had a higher burden of pre-existing comorbidities as defined by the Elixhauser score (P = .02) ( Table 3 and Supplementary Figure 4). Despite their younger age composition and lower Elixhauser score, the severity of disease on presentation (NEWS-2) was similar between patients of BAME and white ethnicity (P = .28) ( Table 3 and Supplementary Figure 4). Furthermore, when stratifying ethnic groups by the median age of the overall cohort (ie, 69 years), we found that the cumulative probability of death of black patients was in fact higher than that of their white counterparts (Figure 2).
We thus performed unadjusted and adjusted logistic regression analyses to assess the odds of death of BAME groups compared to white patients. The unadjusted odds of death did not differ across ethnic groups (Table 3 and Supplementary Table  4). However, when adjusting for age, sex, comorbidities, and NEWS-2 on admission, the odds of death of black compared to white patients was higher (aOR, 1.85 [95% CI, 1.06-3.24]; P = .04; AUROC = 0.78) ( Table 3). If admission severity by  Table 3). The proportion of patients with clinical and laboratory abnormalities on admission associated with increased odds of death also varied by ethnic group (Supplementary Figure 6). Notably, eGFR alterations (P < .01) and fever (P = .01) were more prevalent among BAME groups. The prevalence of the rest of the prognostic markers did not show significant variations by ethnicity.

DISCUSSION
We present detailed demographic, clinical, and outcome data for 614 laboratory-confirmed SARS-CoV-2 cases at a large multiethnic London NHS Trust. Our study provides methodologically robust evidence that, beyond the widely reported factors associated with increased odds of COVID-19 mortality (age, sex, and severe hypoxemia on admission), thrombocytopenia, leukocytosis, hypoalbuminemia, and reduced eGFR are also significantly associated with increased in-hospital death. Furthermore, we present an in-depth analysis of the clinical association between ethnicity and patient outcomes during the COVID-19 epidemic in the UK, a current topic of high global health priority. In addition to an overrepresentation of BAME patients in our hospitalized COVID-19 cohort (when compared to historic admissions to ICHNT; see Supplementary Table  3), we find an association of increased odds of death among black (compared to white) patients, when adjusted for age, sex, burden of comorbidities, and severity of disease on admission.
Emerging evidence on the associations between ethnicity and COVID-19 outcomes has been conflicting. In the UK, reports from gray literature highlight a higher risk of death for BAME than for white patients, which has been partially explained by sociodemographic differences and pre-existing comorbidities [8,11]. An Office for National Statistics report, however, relied on historical population census data (self-reported health indices) to determine these associations [11], and a preprint study focused on comorbidities without admission severity or other clinical observations [8]. Our study thus adds an important clinical perspective, using comprehensive granular patient-level data, which national datasets cannot often provide.
We find that the underlying profiles of the different ethnic groups varied, with black and "other" ethnic minority groups being younger, and Asian and white groups having a higher burden of pre-existing comorbidities. A high prevalence of undiagnosed comorbidities among BAME groups has been previously identified [20], although the significance of this in the current pandemic is yet unknown. We observed that after adjusting for potential confounders, such as age, sex, comorbidities, and severity of admission, black (compared to white) patients were at increased odds of death. This clinically relevant trend was also seen for Asian patients, although it was marginally nonstatistically significant. Interestingly, another recent UK study found ethnicity to be 1 of 8 covariates predicting ICU admission but not mortality [21]. In that study, though, the majority of nonwhite patients were black, whereas our study had high representation of both black and Asian patients.  Table 3 that white patients were significantly older than black, Asian, and other ethnic minority groups. *P < .05.
Together, our findings and those from other emerging studies suggest that important biological factors may be potentially driving differences in COVID-19 hospitalization outcomes by ethnicity, which cannot solely be explained by socioeconomic differences.
The symptom profile in our cohort was largely consistent with previous studies [4-6, 12, 21]. However, both in our study and others from the UK [12], a much higher proportion of patients reported shortness of breath on admission compared with what has been reported in China (65%-67% vs 25%) [22]. Furthermore, a higher proportion of our cohort showed abnormal inflammatory markers and thrombocytopenia. These discrepancies could reflect differing thresholds and criteria for hospitalization between the UK and other countries. High mortality rates have been seen in many settings, for example, 26% among all admissions in an Italian study and 88% for mechanically ventilated patients in a study from the United States [6,23]. We found a crude mortality rate of 29% overall and of 52% among ventilated patients, albeit 55 patients in our study remained hospitalized at the time of censoring. The extent to whether intercountry variations in clinical practice and capacity restraints affects in-hospital mortality is currently unknown and warrants further investigation.
A key strength of our study is that we have systematically identified associations between demographic and clinical factors associated with outcomes in a cohort of >600 patients through multiple regression, survival and competing risks analyses. Previous studies have focused on descriptive statistics [4-6, 12, 22], with few robustly correcting for covariates in their evaluation of outcomes [21].
However, limitations to our study should be acknowledged. First, although we attempted to be as comprehensive as possible, the retrospective nature of this study limited the availability of data. Missing values, including of key variables such as BMI, ethnicity, and pre-existing comorbidities, could be possibly related to the severity of disease on admission and we may have thus missed important associations with mortality. We tried to overcome these data limitations as thoroughly as possible by cross-checking medical clerking on admission and discharge, with patients' referral letters and historic records, where available. Also, many of the nonroutine laboratory tests (including D-dimer, brain natriuretic peptide, and troponin) were only introduced systematically at ICHNT a few weeks into the pandemic. Although descriptive statistics suggest there are differences in the prevalence of these variables by patient outcomes, they were excluded from the multiple logistic regression given incomplete recording. Strengthening of data recording has been implemented in our hospitals. This, together with standardized laboratory investigation protocols, will improve future observational studies, as well as the monitoring of patient outcomes in intervention trials.
Finally, this study did not consider the characteristics and outcomes of those who were hospitalized with clinical features of COVID-19 but were RT-PCR swab negative and those who might have had hospital-acquired SARS-CoV-2 infection. The profiles of these patients and whether their outcomes are different to the patients presented here require further research. Furthermore, we could not quantify the impact of changes in clinical practice over the time of our study. However, only a few patients were enrolled in clinical trials early in the pandemic, so changing clinical practices are unlikely to impact on our study findings.

CONCLUSIONS
This is one of the first studies in the UK to analyze the associations between demographic and clinical characteristics of COVID-19 patients and their hospitalization outcomes by ethnicity. We have employed robust methodological analyses to account for the effect of patients with uncompleted outcomes. The admission characteristics with the strongest association with increased odds of death were fever, severe hypoxemia, thrombocytopenia, leukocytosis, hypoalbuminemia, and a reduced eGFR. BAME groups were overrepresented in our study population, compared with historical admissions to our hospital trust. After accounting for age, comorbidity profile, and severity of disease on admission, patients of black and Asian ethnicity appear to have the worst hospitalization outcomes. Research is urgently needed into complex multifactorial interactions underpinning differences in COVID-19 disease outcomes by ethnicity. Such analyses would have a vital role in informing changes in clinical and public health practice locally and abroad.
Disclaimer. The study sponsors had no role in study design, data collection, data analysis, data interpretation, report writing, or decision to submit for publication. The views expressed are those of the authors and not necessarily those of the funders. The corresponding author had full access to all study data and had final responsibility for the decision to submit for publication.
Financial support. We acknowledge funding from the MRC Centre for Global Infectious Disease Analysis (MR/R015600/1); this award is jointly funded by the UK MRC and the UK Department for International Development (DFID) under the MRC/DFID Concordat agreement and is also part of the EDCTP2 program supported by the European Union. K. H. acknowledges joint center funding from J-IDEA.
Potential conflicts of interest. P. A. reports a research grant from Dr Foster Ltd (a wholly owned subsidiary of Telstra Health). All other authors report no potential conflicts. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.