key: cord-334835-j6u8t8j2
authors: Berenguer, Juan; Ryan, Pablo; Rodríguez-Baño, Jesús; Jarrín, Inmaculada; Carratalà, Jordi; Pachón, Jerónimo; Yllescas, María; Arribas, José Ramón
title: Characteristics and predictors of death among 4,035 consecutively hospitalized patients with COVID-19 in Spain
date: 2020-08-04
journal: Clin Microbiol Infect
DOI: 10.1016/j.cmi.2020.07.024
sha: 
doc_id: 334835
cord_uid: j6u8t8j2

OBJECTIVES: We aimed to analyse the characteristics and predictors of death in hospitalized patients with COVID-19 in Spain. METHODS: Retrospective observational study of the first consecutive patients hospitalized with COVID-19 confirmed by real-time polymerase chain reaction (RT-PCR) assay in 127 Spanish centres until March 17, 2020. The follow-up censoring date was April 17, 2020. We collected demographic, clinical, laboratory, treatment, and complications data. The primary endpoint was all-cause mortality. Univariable and multivariable Cox regression analyses were performed to identify factors associated with death. RESULTS: Of the 4,035 patients, males accounted for 2,433/3,987 (61.0%), the median age was 70 years, and 2,539/3,439 (73.8%) had >1 comorbidity. The most common symptoms were a history of fever, cough, malaise, and dyspnoea. During hospitalization 1,255/3,979 (31.5%) patients developed acute respiratory distress syndrome, 736/3,988 (18.5%) were admitted to intensive care units, and 619/3,992 (15.5%) underwent mechanical ventilation. Viral or host-targeted medications included lopinavir/ritonavir 2,820/4,005 (70.4%), hydroxychloroquine 2,618/3,995 (65.5%), interferon-beta 1,153/3,950 (29.2%), corticosteroids 1,109/3,965 (28.0%), and tocilizumab 373/3,951 (9.4%). Overall 1,131/4,035 (28%) patients died. Mortality increased with age (85.6% occurring in older than 65 years). Seventeen factors were independently associated with an increased hazard of death, the strongest among them included advanced age, liver cirrhosis, low age-adjusted oxygen saturation, higher concentrations of C-reactive protein, and lower estimated glomerular filtration rate. CONCLUSIONS: Our findings provide comprehensive information about characteristics and complications of severe COVID-19 and may help to identify patients at a higher risk of death.

8 carried out a block-wise forward procedure allocating the predictor variables into five clusters: 133 sociodemographic characteristics, comorbidities, admission signs and symptoms, vital signs, and 134 laboratory parameters. A multivariable regression analysis was fitted within each block using two 135 criteria to achieve the best set of predictors: relevance to the clinical situation and statistical 136 significance (P<0.10). We used variance inflation factors to detect collinearity among predictors 137 included in the multivariable models. We carried out a sensitivity analysis in which the order of 138 entry of the blocks was inverted. We checked the proportional hazards assumption. Variables with 139 more than 25% of missing values have not been considered, and missing values were treated as a 140 separate category for analysis. Heterogeneity introduced by different hospitals was accounted for 141 by using robust methods to estimate standard errors and, thus, to calculate 95% confidence 142 intervals (CI) and P-values. Statistical analyses were done using Stata software (version 15.0; Stata 143

Corporation, College Station, Texas, USA. This study is registered with ClinicalTrials.gov, 144

The final cohort included 4,035 hospitalized patients (see web-only Supplementary Figure S1 ) 147 in which SARS-CoV-2 was detected by , pharyngeal swabs 148 Patients' characteristics, categorized by survival, are shown in Table 1 . In brief, males accounted 155 for 61.0%, the median age was 70 years, and 25.1% were > 80 years old. Most patients were 156

Spanish born whites. The age distribution of patients stratified by sex is shown in Figure 1A . 157

At least one comorbidity was present in 73.8% and 26.7% had at least three comorbid conditions. 158

The most common comorbidities were arterial hypertension (51.2%), chronic heart disease 159 (23.3%), diabetes mellitus (21.8%), chronic pulmonary disease (not asthma) (17.9%), and obesity 160 (13.8%). Only 0.7% of patients had HIV. Before admission, 19.4% patients were on angiotensin-161 converting enzyme (ACE) inhibitors, and 17.3% were receiving angiotensin II receptor blockers 162 (ARBs) ( Table 1) . 163

The median duration of symptoms before hospitalization was 4 (IQR 2 -7) days, and the most 164 commonly reported were history of fever ( (15.4%), presumed bacterial pneumonia (10.6%), heart failure (5.8%), and blood-stream infection 199 (4.9%). During the study period, 28.0% of patients died, 64.1% were discharged, and 7.8% 200 remained hospitalized. The median (IQR) time to death since the beginning of symptoms and since 201 hospital admission was 13 (9-19) days and 10 (6-16) days, respectively. Death was particularly 202 high among patients ≥ 80 years (54.9%) ( Figure 1B) and those with ≥3 comorbid conditions 203 (47.7%). Death was also very high among those with ARDS (59.3%), those who were admitted to 204 ICU (42.4%), and those who underwent mechanical ventilation (45.7%). The median (IQR) length 205 of stay was 4 (1-9) days for patients who were discharged; and 35 (32-38) days for those who 206 remained hospitalized at the censoring date. 207

208 Independent predictors of death in the different clusters of variables are shown in Table 3 . In the 209 final adjusted analysis, we found 17 factors independently associated with an increased hazard of 210 death: male sex, older age, arterial hypertension, obesity, liver cirrhosis, chronic neurological 211 disorder, active cancer, dementia, dyspnoea, confusion, low age-adjusted SaO2 on room air, higher 212 white cell blood count (WBC), higher neutrophil-to-lymphocyte ratio, lower platelet count, 213 prolonged INR, lower eGFR, and higher concentrations of CRP (Figure 2) . No collinearity was 214 detected, the proportional hazards assumption was fulfilled, and the results were not changed 215 when the order of entry of the blocks was inverted. Kaplan-Meier plots for death according to age 216 and sex are shown in Figure 3 . The adjusted hazard ratio (aHR) of death for being admitted early 217 in the epidemic (before 13 March) vs later was 1.07 (95% confidence interval [CI]: 0.90; 1.28), 218 P=0.407. The variable unilateral or bilateral lung opacities had missing values in 29% individuals 219 and was not included in the final model. However, when this variable was included in the model, 220

the aHR (95%CI) of death for bilateral opacities in comparison with unilateral opacities was 1.32 221 (0.11; 1.55) P=0.002. We also carried out two post-hoc analyses (data not shown). In the first one, 222 the predictors of mortality among patients ≤ 65 years were not substantially different from those 223 found in the whole data set. In the second analysis, the mortality hazard did not change depending China (3, 13) and one from UK. The majority of patients in all four cohorts were male. However, in 232 comparison with Chinese patients, those from Spain and the UK were on average, two decades 233 older and had a prevalence three times higher of comorbid conditions. It is not surprising thus 234 that mortality was substantially higher in Spain (28%) and the UK (26%) than China (1.4 and 235 3.2%). Presenting features were similar in all cohorts. However, dyspnoea was less frequent in 236

Chinese patients suggesting a more severe course in the older Spanish and British patients. In our 237 cohort, age was the main determinant of death, as has been in other series of hospitalized patients 238

with 8, 9, 14, 19) . Independently of the higher prevalence of comorbidities, it cannot 239 be ruled out that older patients could not have been prioritized to receive ICU treatment. Death 240 was also significantly higher among men than in women, as has also been described in other 241 cohorts (3, 8, 9, 13, 14) . There are sex differences in innate and adaptive immune responses that 242 might have an impact on the inflammatory response and outcomes of COVID-19 and deserve 243 further investigation (20). Hypertension was not only the most common comorbidity in our 244 cohort, as in other studies, but also an independent predictor of mortality. The association 245 between hypertension and poor outcomes in COVID-19 does not seem to be simply a matter of 246 high prevalence; alternative explanations include pre-existing hypertensive end-organ or 247 endothelial damage and interactions between COVID-19 and antihypertensive medications (21). 248

Many patients with hypertension were receiving ACE inhibitors or ARBs, but they did not increase 249 J o u r n a l P r e -p r o o f 14 mortality. Obesity was the fifth most common comorbidity in our cohort but one with the highest 250 hazard of mortality. Obesity has been found to increase the risk of hospitalization and severe 251 outcomes during influenza seasons (22). Recent studies with COVID-19 patients indicate that 252 younger hospitalized individuals were more likely to be obese (23) and that obesity is associated 253 with severe pictures (23-25) and increased mortality (14). Other underlying conditions associated 254

with an increased hazard of death were active cancer and cirrhosis and as has been reported 255 elsewhere (26, 27); meaning that clinicians should consider patients with these underlying 256 conditions as a high-risk category for . We identified several routine laboratory 257 markers as predictors of mortality, including the neutrophil-to-lymphocyte ratio, an indicator of 258 systemic inflammation that has been found of prognostic utility in sepsis (28), and COVID-19 (29, 259

Our study is limited by the retrospective design and the high number of sites, which might have 261 jeopardized the quality of the data. We tried to solve this by selecting simple and well-defined 262 variables and by careful monitoring of the data. Admission criteria might have differed between 263 the sites; nevertheless, we controlled the site effect in the analysis. We could not include in the 264 multivariable model some potentially interesting laboratory parameters, nor changes in 265 laboratory findings over time. The study's strengths include the large sample size, which allowed 266 the identification of a high number of predictors of death at admission, the analysis of clinical and 267 laboratory variables, and the inclusion of sites from areas with different incidence rates. 268

In summary, here we report the clinical characteristics of a large cohort of patients with COVID-19 269 consecutively admitted to hospitals in Spain during the first month of the epidemic. Our findings Ministerio de Ciencia, Innovación y Universidades -co-financed by European Development 293

Regional Fund "A way to achieve Europe", Operative program Intelligent Growth 2014-2020. 

Distribution -No./with data (%)

Abbreviations: IQR, interquartile range; eGFR, estimated glomerular filtration rate Median (IQR) -cells/ x10 9

<1,000 cells/ μL -No./with data (%)

mg/dL

Median (IQR)

/with data (%)

pg/mL

blood cell count; eGFR, estimated glomerular filtration rate; CKD-EPI, chronic kidney disease epidemiology collaboration; APTT, activated partial thromboplastin time; INR, international normalized ratio