key: cord-0867416-2nykwpgv
authors: Manocha, Kevin K.; Kirzner, Jared; Ying, Xiaohan; Yeo, Ilhwan; Peltzer, Bradley; Ang, Bryan; Li, Han A.; Lerman, Bruce B.; Safford, Monika M.; Goyal, Parag; Cheung, Jim W.
title: Troponin and Other Biomarker Levels and Outcomes Among Patients Hospitalized With COVID‐19: Derivation and Validation of the HA(2)T(2) COVID‐19 Mortality Risk Score
date: 2021-02-26
journal: J Am Heart Assoc
DOI: 10.1161/jaha.120.018477
sha: d30869026ebb952b6aac75eb6d055f9a3979eee4
doc_id: 867416
cord_uid: 2nykwpgv

BACKGROUND: The independent prognostic value of troponin and other biomarker elevation among patients with coronavirus disease 2019 (COVID‐19) are unclear. We sought to characterize biomarker levels in patients hospitalized with COVID‐19 and develop and validate a mortality risk score. METHODS AND RESULTS: An observational cohort study of 1053 patients with COVID‐19 was conducted. Patients with all of the following biomarkers measured—troponin‐I, B‐type natriuretic peptide, C‐reactive protein, ferritin, and d‐dimer (n=446) —were identified. Maximum levels for each biomarker were recorded. The primary end point was 30‐day in‐hospital mortality. Multivariable logistic regression was used to construct a mortality risk score. Validation of the risk score was performed using an independent patient cohort (n=440). Mean age of patients was 65.0±15.2 years and 65.3% were men. Overall, 444 (99.6%) had elevation of any biomarker. Among tested biomarkers, troponin‐I ≥0.34 ng/mL was the only independent predictor of 30‐day mortality (adjusted odds ratio, 4.38; P<0.001). Patients with a mortality score using hypoxia on presentation, age, and troponin‐I elevation, age (HA(2)T(2)) ≥3 had a 30‐day mortality of 43.7% while those with a score <3 had mortality of 5.9%. Area under the receiver operating characteristic curve of the HA(2)T(2) score was 0.834 for the derivation cohort and 0.784 for the validation cohort. CONCLUSIONS: Elevated troponin and other biomarker levels are commonly seen in patients hospitalized with COVID‐19. High troponin levels are a potent predictor of 30‐day in‐hospital mortality. A simple risk score can stratify patients at risk for COVID‐19–associated mortality.

S ince its description in late 2019, coronavirus disease 2019 (COVID-19) caused by severe acute respiratory syndrome-novel coronavirus 2 infection has evolved into a pandemic infecting over 29 million people and resulting in >900 000 deaths as of September 15, 2020. 1 Severe COVID-19 is associated with acute respiratory distress syndrome as well as systemic manifestations, which include myocardial injury, cytokine storm, and coagulopathy. [2] [3] [4] Myocardial injury as defined by troponin elevation has been described in 7.2% to 36% of patients with COVID-19 and has been associated with increased mortality. [5] [6] [7] [8] Furthermore, elevation of inflammatory markers have been associated with severe COVID-19. 9 However, the specificity of biomarker elevation for predicting outcomes such as arrhythmias and mortality is not well characterized. Using a multicenter registry, we sought to characterize the distribution of biomarker levels among patients hospitalized with COVID-19, determine the independent association between biomarker elevation and outcomes of arrhythmias and mortality, and develop a novel score that incorporates biomarker levels for estimating the risk of 30-day in-hospital mortality.

This observational cohort study consisted of all consecutive patients with confirmed COVID-19 who were admitted to New York-Presbyterian (NYP)/ Weill Cornell Medicine, a quaternary referral center and 862-bed teaching hospital; and NYP/Lower Manhattan Hospital, a 180-bed community hospital between March 3 and April 6, 2020. The study was approved by the Weill Cornell Medicine Institutional Review Board, which waived informed consent. All cases of COVID-19 were confirmed through realtime reverse-transcriptase polymerase chain reaction assays (GeneXpert, Cepheid, Sunnyvale, CA), on nasopharyngeal swabs. Using REDCap, 10 patient data were manually abstracted from NYP electronic health records to develop a COVID-19 registry as previously described. 11 The database, statistical programs, and study materials that support the findings of this study are available from the corresponding author upon reasonable request.

Demographics (age, sex, and race) and pre-existing comorbid conditions (coronary artery disease, heart failure, hypertension, diabetes mellitus, prior history of atrial fibrillation, pulmonary disease, renal disease [defined as creatinine ≥2.0 mg/dL or need for hemodialysis], and active cancer) were abstracted from the electronic health record. Hypoxia on presentation was defined as use of supplemental oxygen in the emergency department within 3 hours of presentation as abstracted from respiratory flowsheets and nursing documentation. Chest radiographic findings were abstracted from the initial and any follow-up radiology reports and categorized based on the most abnormal findings. Transthoracic echocardiography findings on left ventricular ejection fraction and right ventricular function were abstracted. Right ventricular dysfunction was defined as tricuspid annular plane systolic excursion <17 mm or right ventricular longitudinal myocardial velocity <9. 5 

The primary outcome of the study was 30-day all-cause in-hospital mortality. The secondary outcomes were atrial and ventricular arrhythmias. Atrial arrhythmias were defined as atrial fibrillation, flutter, or tachycardia. Ventricular arrhythmias were defined as frequent premature ventricular contractions, nonsustained ventricular tachycardia, sustained ventricular tachycardia, and ventricular fibrillation. Arrhythmias were identified by review of all ECGs and telemetry strips during hospitalization. ECG, telemetry findings, and complication events were reviewed and adjudicated by study investigators (KM, BP, XY, JK, and JWC). Disagreements on adjudication were resolved by consensus. .04 ng/mL (limit of detection 0.03 ng/mL; maximum 500 ng/mL), BNP >100 pg/ mL (limit of detection 2 pg/mL; maximum 5000 pg/ mL), CRP >0.9 mg/dL (limit of detection 0.4 mg/dL; maximum 91.2 mg/dL), ferritin >322 ng/mL (limit of detection 0.5 ng/mL; maximum 16 500 ng/mL), and d-dimer >230 ng/mL (limit of detection 150 ng/mL; maximum 55 000 ng/mL). Myocardial injury was defined as TnI level ≥0.50 ng/mL. Within this cohort, the quartiles for each biomarker level were determined. Patients were categorized as having severe biomarker level elevation on the basis of a recorded peak level ≥75th percentile for the study population. Bivariate associations between each candidate predictor and the primary outcome (30-day in-hospital mortality) were determined with logistic regression. Baseline demographic characteristics (age, sex, and race), comorbid conditions (including coronary artery disease, heart failure, prior history of atrial fibrillation/atrial flutter, hypertension, diabetes mellitus, pulmonary disease, renal disease, immunosuppression, smoking status, and cancer), marker of disease severity at presentation (hypoxia in the first 3 hours of admission), and severe biomarker (TnI, BNP, CRP, ferritin, and d-dimer) elevation with bivariate significance (P<0.10) for the primary outcome were selected. Multivariable logistic regression models were then used to identify independent predictors of 30-day mortality. Collinearity of variables was assessed with the variance inflation factor. A risk score for 30-day in-hospital mortality was derived from the multivariable model by assigning weighted points based on the beta coefficients for each significant independent predictor (P<0.05). To make the score clinically interpretable, age was categorized as <65, 65 to 74, and ≥75 years (for the study population, 65 years of age was the 50th percentile cut-off and 75 years of age was the 75th percentile cut-off). Given that the derivation cohort consisted of patients who had all 5 biomarkers measured, a validation cohort of patients who had predictive biomarkers measured but were missing the nonpredictive biomarkers could be formed. This validation cohort was used for testing of the mortality risk score ( Figure S1 ). Validation of the models was conducted by bootstrapping 1000 resamples of the validation cohort. The discriminative capacity and agreement between the observed and predicted probability of mortality for the risk score model was assessed with a receiver operating characteristic curve analysis with C-statistic and calibration plot, respectively. Survival curves were then generated using the Kaplan-Meier method and compared by using the log-rank statistic. All analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC) and SPSS version 24 (IBM, Armonk, NY). All tests were 2-sided with P<0.05 indicating statistical significance.

Between March 3 and April 6, 2020, 1053 consecutive patients with COVID-19 were admitted to NYP/ Weill Cornell Medicine and NYP/Lower Manhattan Hospital. Of these, 446 (42.4%) patients had all 5 biomarkers (TnI, BNP, CRP, ferritin, and d-dimer) checked and comprised the primary study cohort. As of May 10, 2020, 242 (54.3%) patients were discharged, 109 (24.4%) were still hospitalized, and 95 (21.3%) had died. The median length of follow-up was 15 (IQR, 6-34; range 0-62) days. The median maximum TnI was 0.05 (IQR, 0-0.34) ng/mL, BNP was 84 (IQR, 25-300) pg/mL, CRP was 27.2 (IQR, 12.7-83.6) mg/ dL, ferritin was 1219 (IQR, 598-2000) ng/mL, and ddimer was 2083 (IQR, 532-6106). The distribution of biomarkers levels are shown in Figure 1 . Based on nominal reference biomarker values, 50.7%, 45.7%, 97.8%, 90.6%, and 94.0% had abnormal levels of TnI, BNP, CRP, ferritin, and d-dimer, respectively. Overall, 99.6% had elevation of at least 1 of the 5 biomarkers. Overall, 93 (20.9%) patients had evidence of myocardial injury with peak TnI levels ≥0.50 ng/mL. Baseline characteristics of the primary study cohort stratified by presence or absence of severe biomarker elevation (ie, maximum levels above the 75 th percentile cut-off) are summarized in Table 1 . Patients with severe elevations of TnI and BNP were older and had a significantly higher proportion of patients with coronary artery disease, heart failure, and prior stroke. Patients with severe elevations of CRP, ferritin, and ddimer had higher proportion of patients with hypoxia upon presentation.

The radiographic and echocardiographic findings of patients stratified by presence of severe biomarker elevation are summarized in Table 2 . Compared with patients without severe ferritin level elevation, patients with ferritin level ≥200 ng/mL had increased prevalence of abnormal chest radiographs and bilateral pulmonary infiltrates. Among patients who received echocardiograms, patients with BNP ≥300 pg/nL had lower left ventricular ejection fractions and higher prevalence of right ventricular dysfunction compared with patients with BNP <300 pg/nL. There was no significant difference in left ventricular ejection fraction or right ventricular function among patients with or without severe elevation of the other biomarkers. A comparison of the antiviral and anti-inflammatory therapy administered to patients with or without severe biomarker elevation is summarized in Table 3 . Across all groups of patients with severe biomarker elevations, significantly higher proportion of patients received steroid therapy. Overall, cardiac catheterization was performed in 3 patients because of troponin elevation. One patient had ST-segment-elevation myocardial infarction and underwent coronary angiography, which revealed a culprit right coronary artery lesion, which was successfully stented. Another patient had unstable angina and non-ST-segment-elevation myocardial infarction and was found to have triple vessel coronary artery disease. Coronary artery bypass graft surgery was planned but the patient developed COVID-19associated respiratory failure, staphylococcal aureus pneumonia, complete heart block requiring pacemaker implantation, and died of multisystem organ failure. The third patient developed new cardiomyopathy and severe eosinophilia and underwent right heart catheterization and endomyocardial biopsy, which was negative for myocarditis.

Hospital outcomes stratified by presence of severe biomarker elevation are shown in Table 4 . The presence of severe elevation of all 5 biomarkers was associated with increased intensive care unit admission, hypotension requiring vasopressor support, and respiratory failure requiring mechanical support. Compared with patients without severe troponin elevation, patients with troponin level ≥0.34 ng/mL had significantly more atrial tachyarrhythmias (36.6% versus 18.3%; P<0.001), ventricular tachyarrhythmias (36.6% versus 17.7%; P<0.001), and 30-day in-hospital mortality (45.5% versus 13.2%; P<0.001). Severe elevation of BNP was also associated with higher prevalence of arrhythmias and death. Severe elevations of ferritin and d-dimer were associated with increased death. The receiver operating characteristic curves for the 5 biomarker levels and their association with 30-day in-hospital mortality are shown in Figure 2 . Of the studied biomarkers, troponin levels had the highest area under the receiver operating characteristic curve of 0.789 (95% CI, 0.739-0.838) for association with 30-day mortality. The univariate and multivariable predictors of 30-day in-hospital mortality are shown in Table 4 . Age and hypoxia on presentation were independently associated with mortality. Among the biomarkers, only severe elevation of troponin (adjusted odds ratio, 4.38: 95% CI, 2.32-8.28; P<0.001) was significantly associated with mortality after adjustment for comorbidities.

A mortality risk score was developed using a multivariable logistic regression model for 30-day mortality using the 3 independent predictors of hypoxia on presentation, age category, and troponin level ≥0.34 ng/ mL. The HA 2 T 2 mortality risk score was calculated for each patient by summing points assigned to each predictor: 2 points for severe troponin elevation, 1 point for age 65 to 74 years and 2 points for age ≥75 years, and 1 point for hypoxia upon presentation (minimum risk score 0 and maximum risk score 5). Among the derivation cohort of 446 patients, patients with a HA 2 T 2 score of 0 (n=59) had a 30-day mortality of 0.0% while patients with a HA 2 T 2 score of 5 (n=29) had a 30day mortality of 65.5% (P-for-trend <0.001) (Figure 3) . Overall, patients with a HA 2 T 2 score of <3 had a 30day in-hospital mortality of 5.9% while patients with a HA 2 T 2 score of ≥3 had a 30-day mortality of 43.7%. The survival curves of patients with HA 2 T 2 score ≥3 and <3 are compared in Figure 4 . The C-statistic for the HA 2 T 2 score and 30-day mortality was 0.834 (95% CI, 0.792-0.876). Using the independent validation cohort of 440 patients who had at least 1 troponin level checked but were missing at least 1 of the other 4 biomarkers, the C-statistic was 0.784 (95% CI, 0.729-0.838). Validation with the bootstrapping method yielded a C-statistic of 0.792 (95% CI, 0.733-0.843). Good agreement between the observed and predicted risk was seen in both the derivation and validation cohorts ( Figure 5 ).

In this study of biomarkers and mortality in a multicenter cohort of patients hospitalized with COVID-19, we identified several important findings. First, abnormal levels of troponin, BNP, CRP, ferritin, and d-dimer are common among patients hospitalized with COVID-19. Second, biomarker level elevation tracked with complications such as hypotension requiring vasopressor therapy, respiratory failure requiring mechanical ventilation, arrhythmias, and death. Third, age, hypoxia on presentation, and severe troponin elevation defined as ≥0.34 ng/mL were independently associated with 30-day in-hospital mortality. Based on these 3 risk factors, the HA 2 T 2 score was developed and validated to predict 30-day mortality with acceptable discriminative capacity. Notably, patients with HA 2 T 2 score ≥3 had >40% 30-day mortality while those with a HA 2 T 2 score <3 had <6% 30-day mortality.

Elevation of multiple biomarkers has been shown to correlate with severe COVID-19. Troponin level elevation has been reported in 7.2% to 36% of patients hospitalized with COVID-19. [5] [6] [7] [8] In addition, increased levels of inflammatory biomarkers such as procalcitonin, ferritin, erythrocyte sedimentation rate, CRP, and IL-6 have been associated with severe COVID-19. 8 Finally, severe d-dimer elevation has also been reported, which is consistent with observations that COVID-19 is associated with increased rates of thrombotic complications. 12 These findings support the notion that severe COVID-19 is a systemic disease process that likely involves multiple mechanisms including myocardial injury, inflammation, endothelial injury, and thrombosis. [2] [3] [4] Indeed, we found significant elevations across 4 of the 5 studied biomarkers in the majority of patients studied in the derivation cohort. However, the relative prognostic significance of these biomarker level elevations in COVID-19 had not been previously elucidated. In our study, although severe elevations of BNP, CRP, ferritin, and d-dimer were associated with mortality, severe elevation of troponin was the only independent predictor of death among the biomarkers, as the presence of TnI ≥0.34 ng/mL was associated with an adjusted odds ratio >4 for 30-day in-hospital mortality.

To our knowledge, our study is the first to compare troponin with other biomarkers to assess its relative association with COVID-associated mortality. Based on our findings, myocardial injury appears to be a common pathway through which COVID-19 can lead to mortality. Whether myocardial injury is predominantly the proximate cause or simply a marker of COVID-19-associated mortality is unclear. The mechanisms of troponin elevation associated with COVID-19 include microvascular injury, stress cardiomyopathy, acute coronary syndrome, hypoxia with supply-demand mismatch, systemic inflammatory response, and direct viral injury. 13 Although there have been case reports of COVID-19-associated myocarditis diagnosed with cardiac magnetic resonance imaging and endomyocardial biopsy, 14,15 its incidence is unclear. Furthermore, while some patients with COVID-19 with troponin elevation and ST-segment elevation have obstructive coronary disease, a substantial proportion of these patients do not undergo coronary angiography. 16 In our study, only 2 patients underwent coronary angiography, with 1 patient undergoing percutaneous coronary revascularization for ST-segment-elevation myocardial infarction. This suggests that in the vast majority of patients in our cohort who had troponin elevation, coronary angiography was not pursued because it was not felt that revascularization would result in significant clinical benefit. In our study cohort, among the patients who underwent echocardiographic imaging, there were no significant differences in the frequency of left or right ventricular dysfunction among patients with or without severe troponin elevation.

Currently, several clinical variables, laboratory findings, and chest radiographic findings have been proposed for use in prediction tools to determine the risk of severe COVID-19 progression and mortality. Liang et al developed a risk tool using 10 variables including age, comorbidities, radiographic findings, lactic dehydrogenase levels, neutrophil levels, and direct bilirubin levels to estimate the risk of COVID-19-associated critical illness. 17 Similarly, Galloway et al proposed a risk tool with 12 variables that incorporated age, radiographic findings, neutrophil levels, CRP, and albumin to determine risk of intensive care unit admission or death caused by COVID-19. 18 Using only the variables of age, hypoxia upon presentation, and troponin level, the HA 2 T 2 score cut-off of 3 can differentiate patients at low versus high risk of in-hospital mortality. Although it has been argued that troponin measurement in patients with COVID-19 has limited utility in that it often cannot be used to determine the need for coronary intervention, 19 it has become clear that troponin levels can be used as an important prognostic marker. 20 It should be noted that low-level troponin elevation among patients hospitalized with COVID-19 is very common, but only a relatively high cut-off value of 0.34 ng/mL for TnI (75th percentile for our derivation cohort patients) had the greatest prognostic value.

Although elevation across a broad range of biomarkers appears to be associated with COVID-19 severity, our study shows that troponin is the most potent predictor of mortality. Based on our receiver operating characteristic analysis, the utility of tracking a wide range of biomarkers other than troponin for prognostication purposes to assess whether rapid institution of antiviral therapies such as remdesivir or anti-inflammatory agents such as dexamethasone in patients with high HA 2 T 2 scores can lower mortality. The COVID-19-associated mortality score may also be used to guide prognostication. For patients with the combination of advanced age, significant comorbidities, and a high HA 2 T 2 score, this tool can be used to aid in goals-of-care discussion.

This is a retrospective study relying on manual chart abstraction for its data, which may be susceptible to error or misinterpretation. Next, because of its retrospective nature, we were unable to perform a comprehensive review of other biomarkers of interest such as IL-6 and procalcitonin. Furthermore, given that biomarkers were not checked as part of a prospective protocol, there was likely selection bias because patients with multiple biomarkers checked are more likely to have greater disease severity than patients who had fewer biomarkers measured. In our study, 446 (42.3%) of the 1053 patients admitted during the study period had all 5 biomarkers measured, but 886 (84.1%) had a least 1 troponin measured, which allowed us to validate our mortality risk score on a sizeable cohort. However, a larger validation cohort using a completely independent patient sample (eg, patients from a distinct institution) would have provided a more robust assessment of the predictive value of the HA 2 T 2 mortality risk score. We categorized patients on the basis of the maximum biomarker level measured, which can affect our results depending on when during the hospital course the level was found and how often patients had biomarkers serially measured. We did not track biomarker levels over time to analyze ranges and trends for each patient. Next, we did not assign specific causes of death in our study. Furthermore, because 24.4% of the study patients remained hospitalized at the time of closure of study data collection, we focused on a 30-day in-hospital mortality end point instead of an overall mortality end point over time. Our study did not examine out-of-hospital deaths after discharge from COVID-19 hospitalization. Finally, this study covered a time period during which New York City was severely affected by COVID. Since then, advances and changes in treatment approaches may have reduced mortality rates among patients hospitalized with COVID-19, which may affect the generalizability of our findings. For example, in our study, 85%, 35%, and 6% of patients were treated with hydroxychloroquine, steroids, and remdesivir, respectively. With the publication of several studies showing a mortality benefit with systemic steroid use among critically ill patients with COVID-19, 21 

Use of a simple risk score, which incorporates troponin levels, age, and presence of hypoxia on presentation, can help stratify patients at risk for in-hospital mortality

Affiliations From the Department of Medicine, Division of Cardiology, Weill Cornell Medicine -New York Presbyterian Hospital

Clinical features of patients infected with 2019 novel coronavirus in Wuhan

Hlh Across Speciality Collaboration UK. COVID-19: consider cytokine storm syndromes and immunosuppression

Coagulopathy and antiphospholipid antibodies in patients with Covid-19

Clinical characteristics of 138 hospitalized patients with 2019 novel coronavirus-infected pneumonia in Wuhan

Cardiovascular Implications of Fatal Outcomes of Patients With Coronavirus Disease 2019 (COVID-19)

Association of cardiac injury with mortality in hospitalized patients with COVID-19 in Wuhan, China

Prevalence and impact of myocardial injury in patients hospitalized with COVID-19 infection

Association of inflammatory markers with the severity of COVID-19: a meta-analysis

Research electronic data capture (REDCap)-a metadata-driven methodology and workflow process for providing translational research informatics support

Clinical characteristics of Covid-19 in New York City

Abnormal coagulation parameters are associated with poor prognosis in patients with novel coronavirus pneumonia

COVID-19 for the cardiologist: a current review of the virology, clinical epidemiology, cardiac and other clinical manifestations and potential therapeutic strategies

Cardiac involvement in a patient with coronavirus disease 2019 (COVID-19)

COVID-19-related myocarditis in a 21-year-old female patient

ST-segment elevation in patients with Covid-19-a case series

Development and validation of a clinical risk score to predict the occurrence of critical illness in hospitalized patients with COVID-19

A clinical risk score to identify patients with COVID-19 at high risk of critical care admission or death: an observational cohort study

High-sensitivity cardiac troponin can be an ally in the fight against COVID-19

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

We thank the following Weill Cornell Medicine medical students for their contributions to the COVID- 19 

Dr Cheung has received consulting fees from Abbott, Biosense Webster, Biotronik, and Boston Scientific and fellowship grant support from Abbott, Biosense Webster, Biotronik, Boston Scientific, and Medtronic. Dr Safford has received research grant support from Amgen. The remaining authors have no disclosures to report.