Risk Model Enhances Prediction of Subsequent Cardiac Events in Patients With Cancer

An AI-based risk prediction model, ONCO-ACS, showed possible favorable clinical utility as a practical tool for predicting cardiovascular death, myocardial infarction, and ischemic stroke events in patients with cancer and acute coronary syndrome. A study reporting the model’s development and validation was published in The Lancet.

Using these data, treatments for these patients could be individualized to better account for their risks of subsequent cardiac events.

“To provide targeted treatment for these patients, clinicians need more accurate tools to assess individual risk profiles,” stated first study author Florian A. Wenzl, MD, from the Center for Molecular Cardiology at the University of Zurich and the National Health Service England.

Key Findings

Researchers sought to create a risk score for events of mortality, bleeding, and ischemic events in patients with cancer and acute coronary syndrome.

Patients with cancer and acute coronary syndrome had a cumulative incidence of mortality of 27.8% (95% confidence interval [CI] = 27.3%–28.3%), 7.3% for major bleeding events (95% CI = 7.0%–7.5%), and 16.1% for ischemic events (95% CI = 15.7%–16.4%). These patients also had a distinct risk profile compared with patients with cancer without acute coronary syndrome.

The factors that influenced the ONCO-ACS risk score included tumor type, time since cancer diagnosis, metastatic disease or not, age, hemoglobin levels, heart rate, estimated glomerular filtration rate, body mass index, Killip class, cardiac arrest, and major bleeding within 6 months.

The time-dependent area under the receiver operating characteristic curve for the risk prediction model was 0.84 (95% CI = 0.83–0.85) for all-cause mortality, 0.70 (95% CI = 0.68–0.73) for major bleeding events, and 0.79 (95% CI = 0.78–0.81) for ischemic events at 6 months. When externally validated, the risk prediction tool achieved a time-dependent area under the receiver operating characteristic curve of 0.80 to 0.84 for all-cause mortality at 6 months, depending on the location of the patient, 0.67 to 0.74 for major bleeding events, and 0.70 to 0.76 for ischemic events.

When the model was applied to current guidelines, the risk prediction score suggested that most of the patients with cancer and acute coronary syndrome would qualify for invasive management and long-term dual antiplatelet therapy. “Depending on the tumor characteristics, patients [with cancer] can be at elevated risk of bleeding, of arterial blood clotting, or both—each requiring different anti-platelet medication for secondary prevention after the acute event,” noted Dr. Wenzl.

“By accounting for both cancer and heart disease, ONCO-ACS marks a step towards truly personalized medicine. It can help doctors decide who benefits from invasive procedures and intensive drug therapy, and who may be at greater risk of harm,” added senior author Thomas F. Lüscher, MD, from the National Heart and Lung Institute, Imperial College London, and the Royal Brompton and Harefield Hospitals.

Model Methods

The researchers tested models with machine learning techniques to predict all-cause mortality, major bleeding events, and ischemic events for patients with cancer and acute coronary syndrome by identifying potential relationships between patients’ baseline characteristics and their outcomes. Between the development and validation sets, the risk prediction models were tested on data from 1,017,759 patients with cancer who had a heart attack in England, Sweden, and Switzerland.

Factors that could impact cardiac events were first pulled from clinical, epidemiological, and prediction model literature, and then the prediction models were trained to assess these variables in patients both with and without cancer. Then, the factors were weighted using Shapley additive explanations and eXtreme Gradient Boosting Gain metrics. The models narrowed factors down to a set of 11 variables for predicting cardiac events.

The final model was the ONCO-ACS score, which was also externally validated on a distinct dataset, looking at different time points to test the model’s forecasting ability. The ONCO-ACS scores were also compared against the Global Registry of Acute Coronary Events (GRACE) score for 6-month mortality, Predicting Bleeding Complications In Patients Undergoing Stent Implantation and Subsequent dual anti-platelet therapy (PRECISE-DAPT) score, and the Patterns of Non-Adherence to Anti-Platelet Regimen In Stented Patients (PARIS) score for atherothrombotic events. Researchers calculated the improvement in risk discrimination and reclassification for each comparison based on the difference in the time-dependent area under the receiver operating characteristic curve at 6 months, the integrated discrimination improvement index, and the continuous net reclassification improvement.

DISCLOSURE: For full disclosures of the study authors, visit thelancet.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.