Model Aims to Predict Immunotherapy-Related Myocarditis Fatalities

Immune checkpoint inhibitor (ICI)–related myocarditis presenting within 30 days of initiating treatment was found to be associated with an increased risk for myocarditis fatality. Researchers developed a machine learning model to predict the likelihood of immune-related cardiac fatality.

“Our analysis indicates that the first month of a patient receiving ICI therapy is the crucial period for determining patients’ risk of myocarditis fatality. If a patent on ICIs develops myocarditis in those first 30 days, that’s a flashing warning light,” said study author Hassan Mohammed Abushukair, MD, Postdoctoral Researcher at the Oklahoma University Stephenson Cancer Center, who presented the findings at the American Association for Cancer Research (AACR) Annual Meeting 2026 (Abstract 5212). “This gives clinicians an actionable timeframe for determining whom ICI therapy may be dangerous for.”

The study also characterized triple-M overlap syndrome (TMOS)—a combination of three toxicities: myocarditis, myositis, and myasthenia gravis—to determine risk factors, incidence, and mortality rates.

Study and Model Methods

Researchers gathered cases of ICI-related myocarditis, myositis, and myasthenia gravis (n = 4,635) in patients with cancer from the World Health Organization (WHO) Vigibase pharmacovigilance database.

The cases were separated into seven groups: myocarditis alone, myositis alone, myasthenia gravis alone, myocarditis plus myositis, myocarditis plus myasthenia gravis, myositis plus myasthenia gravis, and TMOS.

“In this study, we report the largest clinical series of this rare fatal syndrome, aimed at delineating clinical features, fatality predictors, and temporal trends,” the study authors wrote in their abstract.

The researchers also used a subset of patients with ICI-related myocarditis (n = 822) to train and develop a machine learning model for prediction of myocarditis fatality. The cohort was divided into an 80:20 data split for training and testing the model. They also used an external real-world data set of patients with ICI-related myocarditis (n = 37) for independent validation of the model.

The model was developed using Random Forest and XGBoost machine learning techniques, they then added additional baseline troponin T/I protein information to an augmented model. The augmented model was externally validated on a cohort of 68 patients.

“TMOS and the conditions that compose it can easily cause fatalities for the subset of ICI-treated patients who develop these side effects. But clinicians need to know who may be at the greatest risk for fatal outcomes, and we do not yet have that level of understanding,” said Dr. Abushukair. “Our analysis aimed to identify how we can more systematically approach risk stratification for patients who may develop fatal cardiac and autoimmune side effects from ICI treatment.”

Immune-Related Toxicity Findings

Of the overall cases gathered from the WHO Vigibase, there were 2,641 cases of ICI-related myocarditis, including 1,911 (72.4%) that represented myocarditis alone and 730 cases (27.6%) that included myocarditis with myositis and/or myasthenia gravis; the cohort included 207 patients (7.8%) with TMOS.

TMOS was more likely to occur in patients with melanoma (35.7%), older patients (75 vs 68 years; P < .0001), men (66.2% vs 56.7%; P = .0121), and those treated with ICI doublets (25.1% vs 21.5%), as compared with myocarditis alone. Hepatitis occurred most often with TMOS vs myocarditis alone (16.4% vs 3.9%; P < .0001).

Myocarditis fatalities were more common in patients who developed TMOS (37%) vs myocarditis alone (20%), myocarditis and myositis (26%), or myocarditis and myasthenia gravis (29%).

Myocarditis alone started later after beginning ICI (median, 60.8 days) than myocarditis plus myositis (27 days), myocarditis and myasthenia gravis (27 days) and TMOS (26 days; P < .05).

When myocarditis occurred within the first month of ICI initiation, it was independently associated with a greater risk for myocarditis fatality, with adjustments made for age, treatment regimen, cancer type, and co-reactions (odds ratio [OR] for ≤ 1 vs 1–3 months = 0.41; 95% confidence interval [CI] = 0.22–0.73; P = .0036; OR for ≤ 1 vs 3–12 months = 0.44; 95% CI = 0.21–0.86; P = .0212).

Model Performance

The fatality predictor model generated a probability percentage for how likely a patient was to die from their immune checkpoint inhibitor–related myocarditis based on a number of factors including their age, the timing of their myocarditis, and co-occurring immune-related adverse events.

The machine learning model demonstrated an area under the curve of 0.755 on the training data set, 0.727 on the internal testing set. The top features for predicting myocarditis fatalities were early-onset myocarditis and cardiorespiratory co-reactions. On external validation, the augmented model with added baseline troponin information achieved an area under the curve of 0.785.

“I believe the model we’re developing is a great illustration of how even simple clinical data analysis can be used to address fatality in cancer treatment. Ultimately, we envision a helpful bedside tool to rule out high-risk fatality from TMOS and its constituent conditions,” said Dr. Abushukair. “With a greater understanding of the risks that these ICI side effects pose, clinicians and patients alike can be more attuned to which symptoms to be on the lookout for. Our hope is that this will create a safer paradigm for ICI treatment.”

DISCLOSURES: Dr. Abushukair reported no potential conflicts of interest. For full disclosures of the other study authors, visit abstractsonline.com.

ASCO AI in Oncology is published by Conexiant under a license arrangement with the American Society of Clinical Oncology, Inc. (ASCO^®). The ideas and opinions expressed in ASCO AI in Oncology do not necessarily reflect those of Conexiant or ASCO. For more information, see Policies.