Patients with prostate cancer can face elevated cardiovascular risk, shaped by traditional risk factors, cancer-related features, and treatment exposures.. However, commonly used cardiovascular risk calculators were developed in non-cancer populations and may require laboratory values that are not routinely available during oncology visits.
A study published in JCO Oncology Practice developed and externally validated prostate cancer-specific cardiovascular risk classification scores using routinely available clinical variables.
The original article, titled “Development and External Validation of a Machine Learning–Based Risk Classification Score for Prostate Cancer–Specific Cardiovascular Risk,” was published on May 29, 2026, in JCO Oncology Practice.
Authors: Nickolas Stabellini, Tarek Nahle, Viraj Shah, Sai S. Kollapaneni, Omar M. Makram, Jennifer Cullen, Neal L. Weintraub, Harikrishnan H. Kunhiraman, Umang Swami, Morteza Mahmoudi, Nathanael Fillmore, Isla P. Garraway, Darryl Leong, Jehonathan Pinthus, John Shanahan, Alberto J. Montero, Neeraj Agarwal, and Avirup Guha.
Why This Study Matters
Cardiovascular disease is a major source of morbidity and mortality among patients with cancer and long-term cancer survivors. In prostate cancer, cardiovascular risk may reflect conventional cardiovascular risk factors, cancer biology, and treatment exposures, including androgen deprivation therapy.
General cardiovascular risk tools, including the pooled cohort equations, SCORE2, and PREVENT, were not developed specifically for oncology populations. Many of these tools also require inputs such as lipid panels, which may not be available during routine prostate cancer visits. This study aimed to develop practical, prostate cancer-specific cardiovascular risk scores using variables commonly available in oncology workflows.
Read more about Prostate Cancer Cure Rate on OncoDaily.
Study Design
The investigators developed and externally validated machine learning-based risk classification scores for three outcomes: composite cardiovascular disease, atherosclerotic cardiovascular disease, and heart failure. Composite cardiovascular disease was defined as new-onset heart failure, ischemic stroke, or myocardial infarction. Atherosclerotic cardiovascular disease was defined as ischemic stroke or myocardial infarction.
The development cohort included 1,815 men with prostate cancer diagnosed between January 2010 and August 2014 at University Hospitals Seidman Cancer Center. Two external validation cohorts were used. Validation cohort 1 included 4,022 men from RADICAL PC 1. Validation cohort 2 included 1,729 men with prostate cancer from Wellstar Medical College of Georgia.
Patients with prior cardiovascular disease and those with missing diagnosis or follow-up dates were excluded according to cohort-specific criteria. Feature selection was performed using XGBoost and LASSO regression. Final predictors were then modeled and translated into additive point-based risk scores. Patients were categorized as low, intermediate, or high risk.
Study Population
In the development cohort, median follow-up was 10.3 years. Median age at diagnosis was 67 years. In this cohort, 76% of patients were White and 21% were Black. Current or previous smoking was reported in 15% of patients. Advanced-stage disease was present in 8.9%, high-risk Gleason score in 22%, and androgen deprivation therapy use in 9.4%.
The validation cohorts differed from the development cohort. Validation cohort 1 included higher proportions of patients with smoking history, advanced-stage disease, hypertension, diabetes, and androgen deprivation therapy use. Validation cohort 2 had a higher proportion of Black patients and higher rates of hypertension and diabetes.
At 10 years in the development cohort, cumulative incidence was 15.0% for composite cardiovascular disease, 9.4% for atherosclerotic cardiovascular disease, and 10.1% for heart failure.
What Variables Were Included in the Scores
For atherosclerotic cardiovascular disease, the final score retained six predictors: age at diagnosis greater than 65 years, Black race, current or previous smoking, advanced stage, hemoglobin greater than 14.9 g/dL, and platelets greater than 239,000/µL. For heart failure, the final score retained three predictors: age at diagnosis greater than 71 years, advanced stage, and diastolic blood pressure greater than 98 mmHg.
For composite cardiovascular disease, the final score retained six predictors: age at diagnosis greater than 70 years, Black race, current or previous smoking, household members of two or more, high-risk Gleason score, and never-smoking status. Androgen deprivation therapy was included in feature selection for some outcomes but was not retained in the final point scores.
How the Scores Performed
For composite cardiovascular disease, the prostate cancer-specific score showed a 10-year time-dependent AUC of 0.71 in the development cohort and 0.59 in validation cohort 2. For atherosclerotic cardiovascular disease, the score showed a 10-year time-dependent AUC of 0.66 in the development cohort and 0.58 in validation cohort 2.
For heart failure, the score showed a 10-year time-dependent AUC of 0.70 in the development cohort and 0.54 in validation cohort 2. Shorter-term performance was also assessed. For 2-year composite cardiovascular disease, time-dependent AUC was 0.79 in the development cohort, 0.66 in validation cohort 1, and 0.64 in validation cohort 2.
For 2-year atherosclerotic cardiovascular disease, time-dependent AUC was 0.70 in the development cohort, 0.61 in validation cohort 1, and 0.65 in validation cohort 2. For 2-year heart failure, time-dependent AUC was 0.82 in the development cohort, 0.60 in validation cohort 1, and 0.62 in validation cohort 2.
Comparison With Conventional Risk Scores
The prostate cancer-specific score showed better discrimination than pooled cohort equations and SCORE2 across several comparisons, although performance varied by outcome, cohort, and time horizon. Performance compared with PREVENT was heterogeneous. PREVENT showed strong discrimination in some settings, while the prostate cancer-specific score was designed for oncology workflows where laboratory-based inputs may not always be available.
Potential Role in Prostate Cancer Care
The study suggests that prostate cancer-specific cardiovascular risk scores may provide a practical framework for risk stratification during routine oncology care.
Because the scores use routinely available variables, they may support point-of-care cardiovascular risk assessment and help identify patients who may benefit from preventive strategies, closer monitoring, or referral to cardio-oncology services.
This approach may be particularly useful in settings where conventional calculators are difficult to apply because key laboratory inputs are unavailable.
Limitations
The study had several limitations. The development cohort came from a single health system, which may limit generalizability. Although two external validation cohorts were used, differences in coding, demographics, treatment patterns, and care delivery may have affected model performance.
Outcomes were identified using electronic health record and ICD codes, which may introduce misclassification. Some clinically relevant factors, including radiotherapy details, family history, and lifestyle variables, were not included because they were not consistently available across datasets.
The strong influence of age in the models highlights the need for additional covariates in future work. The validation cohorts also differed from the development cohort, and validation cohort 1 had shorter follow-up, meaning 10-year performance could not be assessed in that cohort.
Future work should include larger, diverse, prospective, multicenter validation and should assess implementation in real-world oncology workflows.
Key Takeaway
This study developed and externally validated prostate cancer-specific cardiovascular risk classification scores using variables commonly available in oncology practice.
The scores showed moderate discrimination in the development cohort and more modest performance in external validation. While further validation is needed before broad implementation, these scores provide a practical foundation for scalable cardiovascular risk stratification in patients with prostate cancer.
The full article is available in JCO Oncology Practice.