This study was a retrospective study conducted in the Guanggu Campus of Tongji hospital (Wuhan), Tongji Medical College, Huazhong University of Science and Technology, which was a designated hospital for patients with COVID-19 during the outbreak of novel coronavirus disease in Wuhan, China. The diagnosis of COVID-19 was confirmed by RNA testing of the SARS-CoV-2 in the on-site clinical laboratory according to the WHO interim guidance. 1,284 patients with laboratory-confirmed COVID-19 between February 9, 2020 and March 30, 2020 were initially enrolled. Patients aged <18 years or lacking cardiac biomarker results (e.g., MYO, CK-MB, and hs-TnI) were excluded. Ultimately, 1,229 patients were enrolled in this study. The study was approved by the ethics committee of Tongji hospital, Huazhong University of Science and Technology (TJ-IRB20210317). Written informed consent was waived due to the use of de-identified retrospective data.

All clinical, laboratory and outcome data were available and were collected according to the electronic medical records using a standardized data collection form. Data collection for early laboratory results was defined using the first examination at admission (within 24 h of admission). Data collection for the late-stage laboratory results was defined using the last assessment during hospitalization. All data were verified by two physicians blinded to the patient's identity.

Demographic characteristics, medical history, and physical examination findings at admission, and early and late laboratory results during hospitalization were collected. Demographic characteristics obtained for the study included age and gender. Medical history included hypertension (HP), diabetes (DM), chronic liver disease (CLD), coronary heart disease (CHD), cerebrovascular disease, chronic kidney disease (CKD), chronic obstructive pulmonary disease (COPD), and cancer. Physical examination findings included first body temperature, respiratory rate, pulse rate, systolic blood pressure (SBP), diastolic blood pressure (DBP), and saturation of pulse oxygen (SpO₂). Laboratory findings included hs-TnI, CK-MB, MYO, neutrophil (NEU), lymphocyte (LYM), high sensitivity C-reactive protein (hs-CRP), interleukin 6 (IL-6), D-dimer, fibrinogen (FIB), alanine aminotransferase (ALT), albumin (ALB), creatinine (Cr), estimated glomerular filtration rate (EGFR), and glucose (GLU) levels. All clinical samples were tested in the same laboratory using the same standard.

All patients were followed up from admission to discharge to observe the risks of in-hospital mortality. Survival time was defined as the time from hospital admission to discharge or death. The follow-up data were obtained from reviewing medical records by trained researchers using a double-blind method.

In this study, we focused on three commonly used biomarkers of myocardial injury (hs-TnI, CK-MB, and MYO). At our institution, these three biomarkers have been integrated as a test package into the laboratory test panel for COVID-19 patients. Serum levels of hs-TnI, CK-MB, and MYO were measured by chemiluminescence microparticle immunoassays on the Architect i2000SR platform (Abbott Laboratories, Chicago, IL) according to manufactures' instructions. The values were considered elevated if they were above the upper limit of normal (ULN) levels, which were defined as the 99th upper percentile of the biomarker distribution in the normal population. According to our laboratory normal ranges, the ULN levels of hs-TnI, CK-MB, and MYO were 34.2 pg/ml, 5.2 ng/ml, and 154.9 ng/ml, respectively, in male patients, and were 15.6 pg/ml, 3.1 ng/ml, and 106 ng/ml in female patients.

According to the Guidance for Corona Virus Disease 2019: Prevention, Control, Diagnosis, and Management edited by the National Health Commission of the People's Republic of China (16), all cases were identified into four categories of mild cases, ordinary cases, severe cases, and critical cases based on the clinical conditions at the time of admission. (1) Mild cases: the clinical symptoms are mild, and no pneumonia manifestation can be found in imaging. (2) Ordinary cases: patients have symptoms like fever and respiratory tract symptoms, and pneumonia manifestation can be seen in imaging. (3) Severe cases: meeting any of the following: respiratory distress, RR ≥ 30 breaths/min; the oxygen saturation is <93% at a rest state; arterial partial pressure of oxygen (PaO₂)/oxygen concentration (FiO₂) ≤ 300 mmHg (1 mmHg = 0.133 kPa). Patients with >50% lesions progression within 24 to 48 h in pulmonary imaging were treated as severe cases. (4) Critical cases: meeting any of the following: respiratory failure occurs, mechanical ventilation is needed; shock occurs; or complicated with other organ failures that require monitoring and treatment in the intensive care unit.

Continuous variables were presented as mean ± standard deviation or median (inter-quartile range, IQR), as appropriate. Categorical variables were presented as n (%). Event frequencies were compared by chi-square test. Differences in cardiac biomarker levels between the early- and late-stage groups were compared using the Wilcoxon signed-ranks test (two-tailed). Other comparisons between the two groups were made using the independent samples t-test (normally distributed continuous variables) or the Mann-Whitney U-test (non-normally distributed continuous variables).

To better understand the overall performance of myocardial biomarkers, we performed both standard receiver-operating characteristic (ROC) analysis and time-dependent ROC curve analysis (17, 18). The area under the ROC curve (AUC) was calculated to evaluate the performance of each biomarker. The optimal cut-off point was assessed by standard ROC analysis and determined using the maximization of the Youden's index. Time-dependent AUC curves with 95% confidence intervals (CIs) were performed to compare the prognostic accuracy of cardiac biomarkers.

Cumulative survival curves for in-hospital death were assessed using Kaplan-Meier product-limit estimation with log-rank tests. Univariate Cox proportional hazards regression and mixed-effects Cox models were used to calculate hazard ratios (HRs) for risk factors.

Multivariate Cox proportional hazards model analysis was performed to identify independent prognostic factors associated with in-hospital mortality in patients with COVID-19. Considering the possible collinearity issues, the least absolute shrinkage and selection operator (LASSO) regression was previously performed to screen potential prognostic factors (19). Variables with non-zero coefficients in the LASSO regression were selected for further proportional hazard assumption (Schoenfeld test) to assess the applicability of the variables to the multivariate Cox model (20). If the proportional hazard assumption was not violated (p > 0.05), the multivariate Cox proportional hazard model was then performed. Only variables with a p-value < 0.05 in the multivariate analysis were considered as independent prognostic factors. The predictive performance of the multivariate Cox model was evaluated by Harrell's concordance index (C-index).

The cases with missing biomarker data were excluded listwise with statistics software. Statistical analyses were performed using SPSS 22.0 (SPSS, Chicago, IL, USA) and R software version 3.6.3 (http://www.r-project.org). p-value < 0.05 was considered statistically significant.

Of the 1,229 eligible patients, 66 (5.4%) died during follow-up. Table 1 presents the basic clinical characteristics and physical examination findings for the overall study population and for patients stratified by mortality. Patients who died were older, more likely to be male, and more likely to have a history of stroke and chronic kidney disease than those who were alive. Besides, they had lower SpO₂ and DBP levels and higher body temperature and respire rate levels at admission (Table 1).

Table 2 shows the levels of early and late laboratory parameters in the overall study population, survivors, and non-survivors. Non-survivors had lower LYM, ALB, and EGFR and higher levels of hs-TnI, CK-MB, MYO, NEU, hs-CRP, IL-6, D-dimer, ALT, Cr, and GLU in the early stages of hospitalization compared to survivors. These trends could also be found in the later levels of laboratory results (Table 2).

In terms of dynamic characteristics, the levels of hs-TnI, CK-MB, MYO, NEU, hs-CRP, IL-6, D-dimer, FIB, ALT, and GLU in survivors decreased significantly in the late stage of the disease. In contrast, in non-survivors, their levels increased significantly in the late stage (see Supplementary Table 1).

Figures 1A-C demonstrate the levels of myocardial injury biomarkers. As shown, myocardial biomarker levels were significantly higher in non-survivors compared to survivors in both early and late stages. In terms of dynamic changes, myocardial biomarker levels were significantly decreased in survivors while elevated in non-survivors.

There were 18 (1.5%), 912 (74.2%), 231 (18.8%), and 68 (5.5%) patients with mild, ordinary, severe, and critical conditions, respectively. Myocardial biomarker levels in patients stratified according to clinical classification are shown in Figures 1D-F. Patients in the severe group had higher levels of hs-TnI and MYO in the early stages and higher levels of hs-TnI in the late stages compared to the mild/ordinary group. Patients in the critical group had significantly higher myocardial biomarker levels in both early and late stages than in the severe group.

The percentages of patients with myocardial marker levels below the cut-off (for definition, see later section), between the cut-off and ULN, and above the ULN are illustrated in Supplementary Figures 1A,B. The percentages of deaths in patients with myocardial marker levels below the cut-off, between the cut-off and ULN, and above the ULN were shown in Supplementary Figures 1C,D.

Supplementary Figure 2 shows the correlation matrix heat map for 16 clinical parameters in the early and late stages of the disease. MYO levels were significantly correlated with creatinine, EGFR, hs-CRP, IL-6, hs-TnI, CK-MB, and age. Hs-TnI levels were significantly correlated with ALB, LYM, EGFR, hs-CRP, IL-6, and D-dimer.

To evaluate the associations between in-hospital mortality of COVID-19 and covariates including myocardial injury markers, we performed Cox regression analysis where the Wald χ² and p-values were calculated. After adjusting for age, gender, and comorbidities (HP, DM, CLD, CHD, CKD, COPD, stroke, and cancer), hs-TnI, CK-MB, and MYO were significantly associated with in-hospital mortality of COVID-19. Notably, the Wald χ² values for MYO were relatively higher than those for hs-TnI and CK-MB (Supplementary Table 2).

To assess the association of each myocardial marker elevation above ULN with in-hospital mortality, Cox regression-derived hazard ratios were calculated. After adjusting for age, gender, and comorbidities, the HRs for the risks of in-hospital mortality for elevated early-stage levels of Hs-cTnI, CK-MB, and MYO were 7.47 (95% CI, 4.39-12.72, p < 0.001), 10.37 (95% CI, 5.87-18.30, p < 0.001), and 7.96 (95% CI, 4.75-13.35, p < 0.001). Adjusted HRs for elevated late-stage levels of Hs-cTnI, CK-MB, and MYO were 36.35 (95% CI, 20.13-65.66, p < 0.001), 27.37 (95% CI, 16.19-46.28, p < 0.001), and 77.54 (95% CI, 39.09-153.82, p < 0.001) (see Supplementary Table 3).

Kaplan-Meier curves between groups categorized by ULN are shown in Supplementary Figure 3. As illustrated, patients with hs-TnI, CK-MB, and MYO levels above the ULN had significantly decreased survival rates.

The standard ROC curves were conducted to compare the relative accuracy, sensitivity, specificity, and positive and negative predictive value of each biomarker. The best cut-off values were calculated based on the ROC analysis. Our results show that using the cut-off values determined by the ROC curves improved prognostic accuracy over the use of ULN values.

Table 3 summarizes the overall performance of the myocardial biomarkers. When incorporating the early levels of myocardial biomarkers, the AUC was 0.89 (95% CI 0.84-0.94) for MYO, 0.84 (95% CI 0.78-0.89) for hs-TnI, and 0.78 (95% CI 0.72-0.85) for CK-MB. The optimal cut-offs were 80.8 ng/ml for MYO with a sensitivity of 84.8% and a specificity of 86.2%, 7.9 pg/ml for hs-TnI with a sensitivity of 77.3% and a specificity of 77.4%, and 1.2 ng/ml for CK-MB with a sensitivity of 71.2% and a specificity of 76.2% (Table 3).

When incorporating the late levels of myocardial biomarkers for ROC analysis, the AUCs were 0.96 (95% CI 0.92-1.00) for MYO, 0.94 (95% CI 0.90-0.98) for hs-TnI, and 0.92 (95% CI 0.87-0.96) for CK-MB, respectively. The optimal cut-off point was 98.0 ng/ml for MYO with a sensitivity of 90.9% and a specificity of 96.0%, 15.7 pg/ml for hs-TnI with a sensitivity of 89.4% and a specificity of 94.4%, and 1.5 ng/ml for CK-MB with a sensitivity of 84.8% and a specificity of 91.1% (Table 3).

Time-dependent ROC curve analysis was performed based on early levels of myocardial biomarkers to better understand the performance of each myocardial biomarker (see Figure 2). The AUC values for predicting 14-day survival for MYO, hs-TnI, and CK-MB were 0.88, 0.82, and 0.84, respectively (Figure 2D). For predicting 28-day survival, the AUC values for MYO, hs-TnI, and CK-MB were 0.84, 0.77, and 0.77, respectively (Figure 2E). The time-dependent AUC curves for myocardial markers are shown in Figure 2F. As shown, the AUC values of MYO were superior to those of hs-TnI and CK-MB throughout the follow-up period (Figure 2F).

Patients with biomarker levels above the newly established cut-off value had a significantly higher risk of in-hospital mortality than patients with biomarker levels below the cut-off value. After adjusting for confounders including age, sex, comorbidities (HP, DM, CLD, CHD, CKD, COPD, stroke, and cancer), and disease severity indicators (hs-CRP, D-dimer, NEU) in a multivariate Cox model, the HRs for risk of mortality with increased early levels of hs-TnI, CK-MB, and MYO were 2.31 (95% CI, 1.20-4.48, p = 0.013), 1.82 (95% CI, 1.01-3.31, p = 0.048), and 8.31 (95% CI, 3.94-17.52, p < 0.001), respectively. Adjusted HRs for late levels of hs-TnI, CK-MB, and MYO were 18.19 (95% CI, 7.87-42.05, p < 0.001), 8.22 (95% CI, 3.81-17.71, p < 0.001), and 34.41 (95% CI, 14.15-83.68, p < 0.001), respectively (Table 4).

Kaplan-Meier curves showed an early separation of mortality curves for patients with biomarker levels above the ULN and patients with biomarker levels between the cut-off and the ULN, indicating a significantly increased risk of death in these patients (Figure 3).

Overall, these results suggest that all three commonly used cardiac biomarkers are significantly associated with in-hospital mortality. MYO provided better prognostic performance than hs-TnI and CK-MB in predicting mortality of COVID-19.

To identify independent prognostic factors associated with COVID-19 mortality, we further developed multivariate Cox models. Potential confounders including age, sex, comorbidities (HP, DM, CLD, CHD, CKD, COPD, stroke, and cancer), physical examination on admission (temperature, respiratory rate, pulse, SBP, DBP, and SpO₂), and laboratory parameters (hs-TnI, CK-MB, MYO, NEU, LYM, hs-CRP, IL-6, D-dimer, FIB, ALT, ALB, Cr, EGFR, and GLU) were controlled for. LASSO regression was previously performed to address possible collinearity issues.

When incorporating variables including age, sex, comorbidities, physical examinations, and early-stage laboratory results in the LASSO regression, five variables (MYO, NEU, hs-CRP, IL-6, and D-dimer) with non-zero coefficients were identified (see Figure 4A). They all had p-values < 0.05 in the multivariate model, suggesting an independent prognostic effect of early levels of MYO, NEU, hs-CRP, IL-6, and D-dimer on COVID-19 mortality (see Figure 4B).

Three variables (MYO, hs-CRP, and D-dimer) were identified when incorporating variables including age, sex, comorbidities, physical examinations, and late-stage laboratory results in the LASSO regression (Figure 4C). They were all found to have p-values < 0.05, indicating that late levels of MYO, hs-CRP, and D-dimer had an independent prognostic effect on mortality in COVID-19 (see Figure 4D).

Given that COVID-19 patients with underlying cardiovascular disease are more likely to exhibit elevated myocardial marker levels than patients without cardiovascular disease (6), we performed a subgroup analysis in patients without a history of cardiovascular disease (see Supplementary Figure 4). The results confirmed that MYO provided better prognostic performance than hs-TnI and CK-MB and that MYO was an independent factor associated with in-hospital mortality in COVID-19. We further performed a subgroup analysis of patients with severe or critical conditions. Similarly, the results showed that MYO provided a better prognostic performance with an independent prognostic effect on in-hospital mortality (Figure 5).