This study was conducted as a prospective, multicenter cohort study. Five tertiary hospitals located in China collaborated in the research and gathered pertinent cases. The project received approval from the Institutional Review Board at Tongji Hospital during its meeting on February 2, 2016, and was assigned the unique identifier TJ-C20160202 (the research in this paper meets the ethical standards of the project). All participating centers secured local Institutional Review Board approval for the enrollment of patients with FM and NFM. The study strictly adhered to the ethical guidelines in the Declaration of Helsinki and the International Conference on Harmonization Guidelines for Good Clinical Practice.

Given the absence of a universally accepted definition of FM, previous studies have used heterogeneous criteria. Some investigations have defined FM primarily based on clinical presentation requiring inotropes or MCS^[14,15], which may introduce subjectivity due to treatment-related decision-making and institutional practice variation. To improve objectivity and reproducibility, we adopted criteria grounded in the 2013 European Society of Cardiology (ESC) Working Group Position Statement and further incorporated quantitative indicators of disease severity, including: (1) hemodynamic compromise requiring MCS or vasoactive drugs; and (2) at least one of the following objective markers: diffuse myocardial edema on cardiac magnetic resonance (≥ 50% ventricular segments with T2 ratio ≥ 2.0) or early multiorgan failure [sequential organ failure assessment (SOFA) score ≥ 2 in ≥ 2 systems within 72 h]. This composite definition was designed to reflect both the acute hemodynamic instability and the systemic inflammatory severity characteristic of FM. NFM required: Acute myocarditis cases (symptom duration < 4 weeks) not meeting FM criteria. Among the patients who consented to undergo endomyocardial biopsy (EMB), we proceeded with the examination to obtain a more definitive diagnosis and provide pathological information.

Patients with FM and NFM were consistently enrolled in the study from April 1, 2016, to December 31, 2024. The inclusion criteria for this study were as follows: (1) Histological confirmation of acute myocarditis in accordance with the Dallas criteria^[16] or alternatively, myocarditis confirmed by cardiac magnetic resonance (CMR) imaging based on the Lake Louise Criteria (LLC 2009)^[17]; (2) Acute presentation of the disease, defined as the onset of cardiac symptoms within 30 days prior to hospital admission; (3) The treatment provided during hospitalization must adhere to the identical standards established by Tongji Hospital. The exclusion criteria of this study were as follows: (1) Patients who died during hospitalization; (2) Patients with any other disease that could potentially cause changes in the structure and function of the heart.

The primary composite endpoint included long-term (greater than 1 month, not paying attention to the situation during the hospitalization period) cardiovascular death, heart transplantation, chronic heart failure {defined as left ventricular ejection fraction < 50% with New York Heart Association [NYHA] class ≥ II}^[18], and structural cardiac abnormalities (left ventricular end-diastolic diameter > 5.5 cm or left atrial diameter > 4 cm for males; > 5.0 cm or > 3.5 cm for females, respectively). All cardiac structural measurements were obtained through standardized transthoracic echocardiography performed by experienced sonographers according to current guidelines^[19]. The secondary endpoint focused on persistent troponin elevation (> 34.2 pg/mL with two consecutive measurements) at one month after symptom onset in patients with myocarditis.

All patients were scheduled for outpatient follow-up visits at intervals: 1 month, 3 months, 6 months, and 12 months post-discharge, with annual appointments thereafter. These visits were designed to monitor cardiac function and size of the heart chambers, identify potential complications, update treatment plans, and confirm the occurrence of study endpoint. To ensure the highest quality of care, the follow-up processes were overseen by a dedicated follow-up working group [TY (investigator) and ZPL (author)]. The follow-up working group adhered to strict protocols for patient information recording, as outlined in Tongji Hospital’s guidelines. In cases where follow-up was delayed, the group proactively reached out to patients via multiple communication channels to ascertain their well-being and facilitate the timely completion of the follow-up process. Personnel responsible for follow-up at each center underwent rigorous training to ensure consistency and accuracy in data collection. They conducted independent reviews of all patient information and entered it into Tongji Hospital’s secure electronic database. Additional layers of data quality control were implemented, with regular reviews overseen by Jiangang Jiang and Daowen Wang (both are authors) to ensure integrity and accuracy. Local researchers were contacted as needed to verify details and resolve any queries.

This study aimed to compare the incidence of primary endpoint events between patients with FM and those with NFM. Based on historical clinical data from Tongji Hospital, the incidence of primary outcome events was 45% among FM and 30% among NFM. The study was designed with a significance level of 0.05, a power of 80%, and a 1:1 ratio for the two groups. For the two-sided test, the sample size was calculated using^[20,21]:

(1) $$ n=\frac{\left[Z_{\alpha} \sqrt{2 p q}+Z_{\beta} \sqrt{p_{1}\left(1-p_{1}\right)+p_{2}\left(1-p_{2}\right)}\right]^{2}}{\left(p_{1}-p_{2}\right)^{2}} $$

Where Z_α represents the Z-score corresponding to the significance level (α = 0.05), Z_β represents the Z-score corresponding to the power of 80% (β = 0.20), p₁ and p₂ are the proportions of the primary outcome event in the FM and NFM groups, respectively, and q = 1 − p, where p refers to an average proportion of the primary outcome event. This formula was used to determine the required sample size for the study. The result indicating that the required sample size for each group is n = 163, leading to a total sample size of 326 for the study, was determined through statistical calculations.

Considering a 5% patient dropout rate, we should enroll 344 patients. Additionally, since this study primarily calculates hazards risk based on regression analysis, it is necessary to ensure at least 10 events per variable (EPV) for modeling. Based on the directed acyclic graph (DAG), we plan to adjust for a maximum of 3 variables. Given that NFM is expected to have 52 events, no further sample size increase is required.

After discovering that the missing data amounted to less than 5% and was determined to be Missing at Random, we opted for multiple imputation utilizing the predictive mean matching method^[22,23].

Continuous variables were presented as median with interquartile range (IQR) for non-normally distributed data. To assess statistical differences between groups, we employed the Mann-Whitney U test for non-normally distributed variables. Categorical variables were summarized using frequencies (percentages) and compared between groups using the Chi-square (χ²) test or Fisher’s exact test, depending on their applicability. A two-sided P-value of less than 0.05 was deemed statistically significant in all cases.

Main analysis: Recognizing the limitations of purely data-driven variable selection strategies (e.g., selecting covariates based on statistical significance in univariate analyses), we adopted a causal inference framework using DAGs based on prior clinical knowledge and published literature. DAGs allow explicit specification of assumed causal relationships between exposure, confounders, mediators, and outcomes^[24,25]. This approach helps avoid inappropriate adjustment for mediators or collider variables, which may otherwise introduce bias. Therefore, age and sex were identified as minimal sufficient adjustment variables, while variables such as treatment strategies, complications, and laboratory markers were considered potential mediators and were not included in the primary adjusted model. Three models were constructed: Model 1 (Crude): Isolated association between myocarditis type (FM vs. NFM) and outcomes. Model 2 (Minimally Adjusted): Incorporated age and sex as the confounders (primary analysis). Model 3 (Comprehensive adjusted): Evaluated time from onset to hospital admission (TFOHA) given its dual potential as either a mediator or context-dependent confounder. Several clinically relevant variables, including length of stay (LOS), complications, laboratory parameters, and treatment strategies, were intentionally excluded from adjustment as they were identified through DAG as likely mediators (consequences rather than causes of myocarditis severity). Treatment strategies were particularly avoided as covariates since they represent confounding by indication, being intrinsically linked to the underlying disease severity or type of myocarditis.

Supplementary analyses: Survival analyses were performed using Kaplan-Meier curves with log-rank testing for unadjusted comparisons between groups. To account for potential confounding, we adopted a two-stage modeling approach: First, covariates were screened through univariate Cox regression (significance threshold P < 0.05). Significant predictors were then incorporated into multivariate Cox proportional hazards models to estimate adjusted HR with 95% confidence intervals (CIs). For secondary outcomes, we similarly employed univariate followed by multivariate logistic regression to derive adjusted odds ratios (ORs) with 95%CI. For the primary outcome only, subgroup analyses were performed, including variables such as age, sex, TFOHA, LOS, smoking status, drinking status, and cardiac troponin I at admission.

Explore analyses: Given the potential association between persistent troponin elevation and primary outcomes, we conducted comprehensive exploratory analyses. These included multi-model adjusted Cox regression and Kaplan-Meier survival analyses treating persistent troponin elevation (secondary outcome) as an independent variable and the composite primary endpoint as the dependent variable. We further quantified this relationship using restricted cubic splines (RCS) to allow for non-linear associations.

Sensitivity analyses: To assess the robustness of our findings, we performed sensitivity analyses in two distinct subpopulations: (1) The age-restricted cohort (≥ 15 years) to minimize developmental confounding; (2) A propensity score-matched (PSM) cohort [1:1 matching on age and sex with a caliper width of 0.1 standard deviation (SD)] to address potential demographic imbalances. The analyses were conducted using R (version 4.3.1) with the following primary packages: survival and survminer for survival analysis, MatchIt for propensity score matching, mice for multiple imputation, rms for RCS analysis, and ggplot2 for visualization. Additional packages including dplyr and tableone were used for data management and baseline table generation.

Figure 1 outlines the process of patient inclusion, screening, and follow-up. Among all the patients initially included, 13 individuals (all with FM) died due to cardiovascular causes. The final analysis enrolled 355 patients, comprising 181 with FM and 174 with NFM. Table 1 summarizes the key characteristics of the study population. Notably, the FM versus NFM group exhibited a higher proportion of female patients (51.4% vs. 24.1%, P < 0.001), an older median age [34.0 (21.0 to 50.0) vs. 23.5 (17.0 to 37.8) years, P < 0.001], a longer median TFOHA [3.0 (2.0 to 6.0) vs. 3.0 (1.3 to 4.0) days, P = 0.043], and a lower median left ventricular ejection fraction (LVEF) [30.0 (24.0 to 45.0) % vs. 57.0 (51.0 to 61.0)%, P < 0.001]. Furthermore, the FM group demonstrated a significantly higher incidence of both primary outcome (49.2% vs. 36.2%, P = 0.001) and secondary outcome (39.2% vs. 14.9%, P = 0.001) compared to the NFM group. A significant difference was observed in the median follow-up duration for primary outcomes [18.00 (6.00 to 48.00) months in FM vs. 36.00 (6.00 to 54.00) months in NFM, P = 0.007]. We provide descriptive analyses for each individual component of the primary composite endpoint, though these were not prespecified for formal statistical testing. The observed event rates were as follows: cardiovascular death or heart transplantation occurred in 7(3.9%) FM patients versus 1 (0.6%) NFM patient; chronic heart failure developed in 48 (26.5%) FM patients compared to 13 (7.4%) NFM patients; and cardiac structural abnormalities were identified in 34 (18.7%) FM patients versus 49 (28.1%) NFM patients.

Supplementary Table 1 provides detailed clinical profiles of patients who reached the composite endpoint of cardiovascular death or heart transplantation.

Table 2 illustrates the treatment strategies employed for patients during their hospital stay. In comparison to the NFM group, the FM group exhibited a significantly higher utilization of intra-aortic balloon pump (IABP), with 169 (93.4%) versus 0 patients, P < 0.001. Similarly, a higher proportion of patients in the FM group used extracorporeal membrane oxygenation (ECMO), specifically 55 (30.4%) compared with 0 patients in the NFM group, P < 0.001. Moreover, a higher proportion of patients received immunoglobulin treatment in the FM group (136, 96.5%) compared to the NFM group (106, 73.1%), P < 0.001. Similarly, a greater number of patients in the FM group received glucocorticoids with 140 patients (99.3%) compared to 130 patients (89.7%), P = 0.001.

To better differentiate confounders from mediators, we developed a DAG [Supplementary Figure 1] and employed a multi-model adjustment approach as the main analysis. In the unadjusted model (Model 1), patients with FM demonstrated a significantly higher risk of the primary outcome compared to NFM patients (HR 1.52, 95%CI 1.10-2.11). After adjustment for sex and age in Model 2, the association attenuated to a marginal risk increase (adjusted HR 1.40, 95%CI 1.00-1.97), which was no longer statistically significant. Further adjustment for TFOHA in Model 3 yielded an HR of 1.57 (95%CI 0.97 to 2.52), maintaining the non-significant trend [Figure 2A]. For the secondary outcome, all three models demonstrated consistent statistically significant trends. After adjustment for sex, age, and TFOHA, the OR was 3.44 (95%CI 2.01-6.03). Detailed results are presented in Figure 2B.

In supplementary analyses, survival analysis demonstrates a significantly higher risk of the primary composite endpoint in FM patients than in NFM patients (log-rank P = 0.010; Figure 3). The detailed information of univariate and multivariate regression is presented in Table 3. Compared with NFM, the initial HR and 95%CI of FM patients were 1.52 (1.10 to 2.11) for primary outcome. However, after multivariate adjustment, the HR and 95%CI changed to 0.90 (0.35 to 2.30). For the secondary outcome, FM patients initially exhibited a higher risk, with an OR and 95%CI of 3.67 (2.22 to 6.22). After multivariate adjustment, the OR and 95%CI were 1.61 (0.73 to 3.54), which was not significant. We believe that data-driven methods may lead to over-adjustment, while the results based on the DAG are more accurate.

To assess the consistency of results across clinically relevant variables, we performed prespecified analyses stratified by sex, age, TFOHA, LOS, smoking status, drinking status, and cardiac troponin I (cTnI). Continuous variables (age, TFOHA, LOS, cTnI) were dichotomized at median values. As shown in Figure 4, the direction of effect favored a consistent overall result across all subgroups, with no significant heterogeneity detected (all interaction P-values > 0.05).

Unadjusted analysis demonstrated that persistent troponin elevation (secondary outcome) was significantly associated with an increased risk of the primary composite endpoint (HR 1.97, 95%CI 1.42-2.74). This association remained significant after adjustment for age and sex (adjusted HR 1.66, 95%CI 1.18-2.33; Supplementary Figure 2).

Further analysis examined the log2-transformed average of two consecutive troponin measurements taken at 1-month post-discharge. A positive linear relationship was observed between troponin I levels and the risk of the primary outcome once troponin exceeded the upper limit of the normal reference range (34.2 pg/mL, log2-transformed value of 5). This dose-dependent association persisted in both unadjusted [Figure 5A] and adjusted models [Figure 5B].

To address potential heterogeneity between pediatric and adult populations, we conducted analyses in patients aged ≥ 15 years. Supplementary Table 2 presents the baseline clinical characteristics of these adult patients, while Supplementary Table 3 details their treatment regimens. Unadjusted survival analysis [Supplementary Figure 3] demonstrated that FM patients maintained a significantly higher risk of developing the primary composite endpoint, consistent with findings in the overall population. This non-significant association remained robust after multi-model adjustment (Supplementary Figure 4A for primary outcome and Supplementary Figure 4B for secondary outcome). Similar results were observed in the relationship between secondary and primary outcomes [Supplementary Figure 5].

PSM based on DAG effectively balanced age and sex distributions between groups (other factors served as mediating variables and did not require balancing), as evidenced by the baseline characteristics and treatment details presented in Supplementary Tables 4 and 5. Survival analysis of the matched cohort showed no significant difference between groups (log-rank P = 0.084, Supplementary Figure 6). Subgroup analyses of the balanced population [Supplementary Figure 7] yielded results concordant with those from the overall cohort.

Firstly, EMB was not conducted for all patients in this study. However, despite expert consensus and other statements recommending EMB for patients with acute myocarditis to guide diagnosis or treatment decisions^[1,36], this recommendation has not been confirmed through large-scale clinical trials. Historical research also indicates a relatively low rate of EMB completion. For instance, Ammirati et al.^[11] reported that only 26.7% of patients with acute myocarditis underwent EMB, while in a multicenter study, EMB was performed in 183 (44%) suspected FM patients^[37]. Similarly, a Japanese single-center study of 22 patients did not include EMB^[38]. Compared to these previous studies, our research has made progress in addressing this limitation.

Secondly, our study did not incorporate molecular analysis using gene test, and immunohistochemistry on myocardial biopsy tissue samples was not performed in all centers. This was primarily due to a lack of compelling evidence supporting the use of these procedures, as well as financial constraints faced by the research team.

Thirdly, while our sample size met pre-specified power calculations, it remained insufficient to definitively exclude modest but clinically important risk differences (particularly for rare outcomes such as heart transplantation). This limitation underscores the need for larger, collaborative studies to validate these findings.

Our study demonstrates that patients with FM under the current treatment regimen based on Chinese guidelines did not show a significantly increased long-term risk of the primary composite endpoints (cardiovascular death, heart transplantation, chronic heart failure, and cardiac structural abnormalities) compared with NFM patients. However, it should be noted that, given the borderline P value and the limited statistical power, this finding should be interpreted with caution. Notably, FM patients had a substantially higher incidence of persistent cTnI elevation at one-month follow-up than NFM patients. In both FM and NFM groups, persistent cTnI elevation was associated with an increased risk of developing the primary composite endpoints, with the magnitude of risk showing a positive correlation with the degree of cTnI elevation.