Key result
Only ~33% of CV risk-prediction models achieve acceptable discrimination in cancer survivors.
Why the study?
Do cardiovascular risk-prediction models accurately predict incident cardiovascular events in adult cancer survivors?
Meta-Analysis (n=983,438)
Do cardiovascular risk-prediction models accurately predict incident cardiovascular events in adult cancer survivors?
General and cancer-specific cardiovascular risk models show variable and often limited predictive performance in cancer survivors, highlighting the need for externally validated models tailored to specific cancer populations.
Answers are grounded in this paper and cite their sources.
Variable model performance warrants caution for risk stratification in cancer survivors; extends need for validated, cancer-specific tools.
This systematic review and meta-analysis included 34 studies encompassing 983 438 adult cancer survivors and evaluated the performance of both general-population and cancer-specific cardiovascular risk-prediction models across vascular events, cancer therapy-related cardiac dysfunction or heart failure, arrhythmias, and composite cardiovascular event outcomes. In total, 27 unique risk scores were assessed. Approximately one-third of model-outcome evaluations demonstrated acceptable-to-good discrimination, defined as an area under the curve of ≥0.70. Among general-population models, four risk scores demonstrated acceptable-to-good discrimination for vascular events, three models were acceptable for CTRCD/heart failure, and two models for arrhythmias. The New Zealand CVD score was the only composite outcome model with an AUC ≥0.70. Cancer-specific scores, such as HFA-ICOS and the Ezaz score, showed high specificity for selected outcomes and may support rule-in decisions, although their lower sensitivity limits use for ruling out risk of CTRCD/heart failure. In our subgroup analysis, model performance varied by cancer type, with more consistent discrimination in breast cancer cohorts and substantially poorer performance in survivors of haematologic malignancies, indicating limited transportability across cancer populations. In summary, these findings provide a comprehensive overview of available cardiovascular risk-prediction tools and may assist clinicians in selecting appropriate models for primary cardiovascular prevention in cancer survivors while highlighting the need for exposure-aware, externally validated models tailored to specific cancer populations.
No takes yet. Share an insight, caveat, or question.
Kazemian et al. (2026) conducted a meta-analysis in Cancer survivors (n=983,438). Cardiovascular risk-prediction models was evaluated on Model discrimination (AUC/C-index) for cardiovascular events. Approximately one-third of cardiovascular risk-prediction models demonstrated acceptable-to-good discrimination in cancer survivors, with general-population models performing best for vascular events.
Synapse has enriched 10 related papers. Consider them for comparative context: