key: cord-0681937-rf4esnhf
authors: Ururahy, Raul dos Reis; Gallo, César Albuquerque; Besen, Bruno Adler Maccagnan Pinheiro; de Carvalho, Marcelo Ticianelli; Ribeiro, José Mauro; Zigaib, Rogério; Mendes, Pedro Vitale; Park, Marcelo
title: Bedside clinical data subphenotypes of critically ill COVID-19 patients: a cohort study
date: 2021
journal: Rev Bras Ter Intensiva
DOI: 10.5935/0103-507x.20210027
sha: bb43c2144b0b21fbbdbf4ff17e17fdc6c13f0a00
doc_id: 681937
cord_uid: rf4esnhf

OBJECTIVE: To identify more severe COVID-19 presentations. METHODS: Consecutive intensive care unit-admitted patients were subjected to a stepwise clustering method. RESULTS: Data from 147 patients who were on average 56 ± 16 years old with a Simplified Acute Physiological Score 3 of 72 ± 18, of which 103 (70%) needed mechanical ventilation and 46 (31%) died in the intensive care unit, were analyzed. From the clustering algorithm, two well-defined groups were found based on maximal heart rate [Cluster A: 104 (95%CI 99 - 109) beats per minute versus Cluster B: 159 (95%CI 155 - 163) beats per minute], maximal respiratory rate [Cluster A: 33 (95%CI 31 - 35) breaths per minute versus Cluster B: 50 (95%CI 47 - 53) breaths per minute], and maximal body temperature [Cluster A: 37.4 (95%CI 37.1 - 37.7)°C versus Cluster B: 39.3 (95%CI 39.1 - 39.5)°C] during the intensive care unit stay, as well as the oxygen partial pressure in the blood over the oxygen inspiratory fraction at intensive care unit admission [Cluster A: 116 (95%CI 99 - 133) mmHg versus Cluster B: 78 (95%CI 63 - 93) mmHg]. Subphenotypes were distinct in inflammation profiles, organ dysfunction, organ support, intensive care unit length of stay, and intensive care unit mortality (with a ratio of 4.2 between the groups). CONCLUSION: Our findings, based on common clinical data, revealed two distinct subphenotypes with different disease courses. These results could help health professionals allocate resources and select patients for testing novel therapies.

The severe clinical presentation of 2019 coronavirus disease (COVID- 19) requiring admission to the intensive care unit (ICU) is associated with high mortality. (1) Early clinical deterioration is mainly associated with nonpulmonary organ dysfunctions and carries the highest mortality. (2) Moreover, the precocious recognition of more severe forms of the disease is essential.

In acute respiratory distress syndrome (ARDS) patients, clinical, laboratory, and inflammatory data are capable of identifying subphenotypes of more severe presentations (3) (4) (5) and, perhaps, guiding respiratory support. (6) COVID-19 patients share some characteristics, predominantly laboratory, which are capable of disclosing the more severe ones. (7, 8) Despite the large amount of recent literature published on COVID-19, it is still a new disease, and there is a lack of clinical information about its evolution. Moreover, at bedside, promptness in the ascertainment of information is crucial for making critical decisions.

Therefore, the aim of this study was to identify if there are clinical characteristics, at ICU admission and stay, able to identify the more severe clinical presentations of COVID-19 patients.

This is a retrospective cohort study of critical COVID-19 patients. Data were retrieved from a prospectively collected database from March 19, 2020 to August 3, 2020, which was derived from a single 12bed ICU at an academic tertiary care center in São Paulo, Brazil. The Research Ethics Committee of Hospital das Clínicas of the Universidade de São Paulo approved the study protocol (number 107.443), and Informed Consent was waived because of the observational nature of the study.

All patients admitted to the ICU with suspected or confirmed critical COVID-19 were included in this analysis. Patients in whom COVID-19 suspicion was low and reverse transcription-polymerase chain reaction (RT-PCR) for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), serology for SARS-CoV-2 and/or chest tomography were not suggestive of the disease were excluded from the analysis.

In the ICU, patients received organ support according to the current best evidence, without the use of antibiotics (unless coinfection or superinfection was strongly suspected or confirmed) (9) or antiviral drugs (unless in a research protocol). (10) However, prior to ICU transfer, in the emergency setting, most patients did receive at least one dose of antimicrobials, mostly ceftriaxone, azithromycin and/or oseltamivir. Thromboembolism prophylaxis was performed with 40mg of enoxaparin or 15,000 IU of unfractionated heparin. (11, 12) Corticosteroids were used as methylprednisolone 1 -2mg/kg/day for 14 days and tapered up to 28 days. (13) (14) (15) (16) Both lung protective mechanical ventilation and prone positioning were used as classically described. (17, 18) Driving pressure was used only occasionally to titrate positive end-expiratory pressure (PEEP) in some patients but not as a bedside target variable. (19) Patients were intubated only secondary to severe hypoxemia or severe respiratory distress; thus, no patient was intubated early to avoid self-inflicted lung injury. (20) Neuromuscular blockade was used only in the presence of severe asynchrony or air hunger. (21) The cumulative fluid balance was targeted to zero as soon as possible. (22) Corticosteroids were used in almost all patients. (16, 23, 24) Because of the extensive human and economic resource burdens, extracorporeal membrane oxygenation (ECMO) was used only in severely hypoxemic patients (when oxygen partial pressure in the blood over the oxygen inspiratory -PaO 2 /FiO 2 -ratio was persistently lower than 50mmHg despite rescue maneuvers), in patients ventilated up to 7 days, younger than 60 years old, and without severe comorbidities. Extracorporeal membrane oxygenation was not used to treat refractory hypercapnia; instead, high-frequency positive pressure ventilation (HFPPV) was frequently used. These ECMO criteria were not in line with the current literature (25, 26) but were adapted to be suitable for more severe disease presentations during the pandemic outbreak.

Clustering analysis was used to characterize and aggregate patients. Furthermore, the variable selection to be clustered was based on clinical simplicity, availability, and low cost. Therefore, we chose to include vital signs, namely, heart rate (HR), respiratory rate (RR), and temperature, all collected every 2 hours during the whole ICU stay. Furthermore, PaO 2 /FiO 2 at the time of ICU admission was also used for clustering. After clustering, organ dysfunction, organ support, and clinical outcome data were compared between the clusters. The creatinine level was evaluated through the worst value documented variation from baselineto partially adjust the current creatinine value to prior chronic renal impairment.

The quantitative data are presented as the mean ± standard deviation, with the exception of ICU length of stay and days on mechanical ventilation, which are presented as the median [25 th percentile and 75 th percentile]. The comparisons between survivors and nonsurvivors were performed using a t-test assuming equal variances, the Mann-Whitney test, a chisquared test or Fisher's exact test, as appropriate. The aforementioned tetrad of cardinal indicators was the substrate for clustering. These indicators were tested and selected in individual combinations until a visual (graphical) clear separation in different groups of k-means. Standardization using Z scores was adopted to mitigate the scale's differences bias. The expectation maximization method was applied through the Microsoft clustering algorithm, carried out by Power BI software, in a multistep approach, and the number of clusters (k) was defined by the means of two different systems, automatically by the program's algorithm, and by the elbow method prediction model. A combinatorial analysis of the four measuring scales was then performed. Given the same dataset, different initial conditions may generate considerably dissimilar clusters, (25, 26) which underpins this multifaceted processing. Moreover, a trinomial subanalysis allowed the elaboration of dispersion diagrams, favoring visualization, an intuitive way to perceive and to validate clusters. (25, 26) Subsequently, the method's findings were scrutinized and then merged, with avoidance of superimposed data being assured. The resulting dataset was further refined by preserving only the data points constant in all models to potentiate cluster solidity. On the other hand, the price paid was the shrinking of the sample size. Finally, the cluster's internal quality was ascertained through reclustering, now taking into account supplementary nonbinary variables, a recounted system for validating results and evaluating group stability. (25, 26) The arising groups were compared with the parent clusters, and the matching rate was measured. Confidence intervals (95%) were calculated as usual. R version 4.0.2 free-source software was used for the nonclustering analyses.

Data from 147 consecutive patients were gathered, of which the data from three patients were excluded after confirmation of alternate diagnoses. Table 1 shows the general characteristics of patients stratified according to survival, where survivors showed substantially lower Simplified Acute Physiological Score 3 (SAPS 3) values. Despite the clinically high suspicion of COVID-19, RT-PCR was positive in only 101 patients (69%). In table 2, organ failure and ICU support are shown; in the survivor's group, lower maximal SOFA with the exception of the hematological domain, less invasive mechanical ventilation, less neuromuscular blockade, less prone position, less vasopressors, less continuous renal replacement therapy and less antibiotics were needed. The ICU outcomes are shown in table 3; there were 46 nonsurvivors (31%).

The clustering process led to two well-defined assemblies ( Figure 1 and 5 )°C] were higher during the ICU stay. All the respiratory, cardiovascular and renal support metrics differed between the groups, both in frequency and duration, with an increased intervention need in cluster B. The white cell counts in Cluster B were appreciably increased when set against the findings of Cluster A, as were the CRP levels. Thrombotic events occurred more often in Cluster B, and the maximal plasma D-dimer levels was also higher in this cluster. Finally, the SAPS 3 and maximal Sequential Organ Failure Assessment -SOFA (in all six domains) score differences surfaced in the cluster comparison, which reinforced a highly relevant mortality rate variance that was 4•2 times higher in Cluster B that that in Cluster A.

The daily mean variation amplitude was wide for the three physiological parameters. In Cluster A, the HR average oscillation was 72 -99 beats per minute, the RR was 16 -28 breaths per minute, and the temperature was 35.5 -37.0°C. In Cluster B, the observed fluctuations were 92 -126 beats per minute, 19 -36 breaths per minute, and 35.8 -38.9°C, respectively. Assuming the upper limits of the range as the boundary of Cluster A, considering the whole group of patients, only in 8.6% of the observed time the HR was compatible with the Cluster A subphenotype. The same occurred in 25.6% and 13.6% of the observed time for RR and temperature, respectively ( Figure 2 ). The parameter interrelationships were also heterogeneous. The three variables stood together consistently compatible with Cluster B in 60.3% of the observed time, and only in 0.6% of the observed time were the three variables together compatible with Cluster A (Figure 2 ). k -minimized through the squared Euclidean distances within clusters. HR -heart rate; Temp -body temperature; PaO 2 /FiO 2 -oxygen partial pressure in the blood over the oxygen inspiratory fraction; RR -respiratory rate. The merge refinement represents the probabilistic distribution of the clusters according to the expectation maximization algorithm. 

Considering only ICU COVID-19 patients, heterogeneity remains a marked feature. In our patients, there were several clinical-laboratory differences in regard to general characteristics, organ failure, and organ support between severe COVID-19 patients who survived and those who did not survive their ICU stay. However, simple clinical variables such as HR, RR, and body temperature during the ICU stay and the PaO 2 /FiO 2 ratio at the time of ICU admission were able to separate the COVID-19 patients into two different subphenotypes.

Some patient characteristics were different between the survivors and nonsurvivors at ICU admission, such as the SAPS 3 score, age, PaO 2 /FiO 2 ratio, lactate and acid-base status, all of which are in line with the current literature. (27) (28) (29) Models for the prediction of unfavorable evolution of COVID-19 have been proposed. There are different outcome prediction models, taking into account demographic data, (2) laboratory data, (2) and the combination of clinical plus radiologic features. (2) Otherwise, no study has been dedicated to exploring only bedside clinical data. In this way, also based upon the premise of different courses of disease, the clustering of COVID-19 in subphenotypes has been reported. The approach adopted by Azoulay et al. (2) included clinical and laboratory multiparametric analyses, eliciting findings consistent with risk-prediction studies. The refinement of the clustering method resulted in a reduced sample size in both clusters; moreover, this technique reduces the sensitivity of cluster characteristics, otherwise enhancing their specificity. (30) It is interesting to note that the time spent with vital signs within the range of Cluster A was low, probably because those patients had a shorter ICU stay. Moreover, this physiological behavior brings a consistent clinical meaning of a good outcome pool of patients.

The purpose of the present study, which used predominantly clinical data, was to offer a cost-effective alternative for resource allocation guidance that eventually may aid in the selection of candidates for testing novel therapies or even for the early implementation of treatments in the future.

The limitations of our study include the sample size, the single-center source of the patients, the subjectivity that permeated the selection of variables for clustering, and the lack of validation in an external cohort. In compensation, our proposition was built in such a way that the wide heterogeneity of resource availability across centers would not become a constraint to prospective studies in different or larger populations. Furthermore, since patient stratification is a critical task in clinical decision making, bedside guiding elements could thus facilitate and hasten these judgements. An additional strength of the study is the considerable premorbid similarity between the individuals from both groups, which minimized the confounding factors. Additionally, the academic tertiary health service status, together with Brazil's (and São Paulo's) broad sociocultural diversity, may have contributed to reducing the underrepresentation of population subsets.

This study was able to identify two clinically distinct subphenotypes of COVID-19 patients in accordance with disease severity. Maximal heart rate, body temperature, respiratory rate and the intensive care unit admission oxygen partial pressure in the blood over the oxygen inspiratory ratio are bedside variables that can help identify more severe COVID-19 patients.

RR Ururahy: conceptualization, methodology, formal analysis, investigation, and writing (original draft); CA Gallo: methodology, investigation, and writing (review and editing); BAMP Besen: investigation and writing (review and editing); MT Carvalho: investigation and writing (review and editing); JM Ribeiro: investigation and writing (review and editing); R Zigaib: investigation and writing (review and editing); PV Mendes: investigation and writing (review and editing); M Park: conceptualization, methodology, formal analysis, investigation, writing (review and editing), and project administration.

The data that support the findings of this study are available from the dataset of Hospital das Clínicas of the Universidade de São Paulo's intensive care unit. Restrictions apply to the availability of these data, which were used under license for the current study and thus are not publicly available. Data are, however, available from the authors upon reasonable request and with the permission of the entity's Research Ethics Committee.

Objetivo: Identificar apresentações mais graves de COVID-19.

Métodos: Pacientes consecutivamente admitidos à unidade de terapia intensiva foram submetidos à análise de clusters por meio de método de explorações Conclusão: Nossos achados, baseados em dados clínicos universalmente disponíveis, revelaram dois subfenótipos distintos, com diferentes evoluções de doença. Estes resultados podem ajudar os profissionais de saúde na alocação de recursos e seleção de pacientes para teste de novas terapias.

COVID-19 in-hospital mortality and mode of death in a dynamic and nonrestricted tertiary care model in Germany

Increased mortality in patients with severe SARS-CoV-2 infection admitted within seven days of disease onset

Hostresponse subphenotypes offer prognostic enrichment in patients with or at risk for acute respiratory distress syndrome

Biomarkers for acute respiratory distress syndrome and prospects for personalised medicine

Subphenotypes in acute respiratory distress syndrome: latent class analysis of data from two randomised controlled trials

Development and validation of parsimonious algorithms to classify acute respiratory distress syndrome phenotypes: a secondary analysis of randomised controlled trials

Clinical phenotypes of critically ill COVID-19 patients

Coronavirus disease 2019 (COVID-19): cytokine storms, hyper-inflammatory phenotypes, and acute respiratory distress syndrome

Cavalcanti AB; COALITION COVID-19 Brazil II Investigators. Azithromycin in addition to standard of care versus standard of care alone in the treatment of patients admitted to the hospital with severe COVID-19 in Brazil (COALITION II): a randomised clinical trial

Hydroxychloroquine with or without Azithromycin in Mild-to-Moderate Covid-19

American Society of Hematology 2018 guidelines for management of venous thromboembolism: prophylaxis for hospitalized and nonhospitalized medical patients

A comparison of enoxaparin with placebo for the prevention of venous thromboembolism in acutely ill medical patients. Prophylaxis in Medical Patients with Enoxaparin Study Group

Effect of prolonged methylprednisolone therapy in unresolving acute respiratory distress syndrome: a randomized controlled trial

Methylprednisolone infusion in early severe ARDS: results of a randomized controlled trial

Ancukiewicz M; National Heart, Lung, and Blood Institute Acute Respiratory Distress Syndrome (ARDS) Clinical Trials Network. Efficacy and safety of corticosteroids for persistent acute respiratory distress syndrome

Association between administration of systemic corticosteroids and mortality among critically ill patients with COVID-19: a meta-analysis

Ventilation with lower tidal volumes as compared with traditional tidal volumes for acute lung injury and the acute respiratory distress syndrome

Prone positioning in severe acute respiratory distress syndrome

PEEP titration to minimize driving pressure in subjects with ARDS: a prospective physiological study

P-SILI is not justification for intubation of COVID-19 patients

Early neuromuscular blockade in the acute respiratory distress syndrome

Comparison of two fluid-management strategies in acute lung injury

Dexamethasone in hospitalized patients with Covid-19

Azevedo LC; COALITION COVID-19 Brazil III Investigators. Effect of dexamethasone on days alive and ventilator-free in patients with moderate or severe acute respiratory distress syndrome and COVID-19: The CoDEX Randomized Clinical Trial

Extracorporeal membrane oxygenation for severe acute respiratory distress syndrome

Extracorporeal membrane oxygenation for severe respiratory failure in adult patients: a systematic review and meta-analysis of current evidence

SAPS 3--From evaluation of the patient to evaluation of the intensive care unit. Part 2: Development of a prognostic model for hospital mortality at ICU admission

ERICC (Epidemiology of Respiratory Insufficiency in Critical Care) investigators. Clinical outcomes of patients requiring ventilatory support in Brazilian intensive care units: a multicenter, prospective, cohort study

Differences in acid-base behavior between intensive care unit survivors and nonsurvivors using both a physicochemical and a standard base excess approach: a prospective, observational study

Interactive visual exploration and refinement of cluster assignments