Development of an Electronic Medical Record–Based Score for Heart Failure Prediction in Cancer Survivors

Figures

• 

  Figure 1. ROC Curves for HF Prediction in the Testing Data Set This figure compares the ROC curves of the CHERISH and ARIC-HF models in predicting 3-, 5-, and 10-year HF incidence among cancer survivors in the UK Biobank validation cohort. CHERISH consistently demonstrated superior discriminative performance across all time points, with a 10-year AUC of 0.83 (95% CI: 0.81-0.85) vs 0.70 (95% CI: 0.64-0.76; P < 0.001) for ARIC-HF. These findings support CHERISH as a robust tool for long-term incident HF risk stratification in survivorship care. ARIC = Atherosclerotic Risk In Community studies; AUC = area under the curve; CHERISH = cancer-specific HF prediction from electronic medical records in survivor health care; HF = heart failure; ROC = receiver operating characteristic.

• 

  Figure 2. Comparison of Net Benefit Between CHERISH and Default Strategies CHERISH risk score was above both “treat all” and “treat none” across a meaningful range of threshold probabilities (roughly between 0.02 and 0.2), indicating improvement in clinical net benefit. Abbreviations as in Figure 1.

• 

  Figure 3. Observed and Predicted HF Incidence by CHERISH in Deciles This figure illustrates the observed vs predicted 10-year incident HF risk across deciles of the CHERISH risk score among cancer survivors in the Ontario administrative database. Calibration was assessed by comparing model-predicted cumulative incidence against observed HF events in each decile. The CHERISH model showed good agreement across the risk spectrum, with better calibration in the mid-to-high deciles. Despite a modest reduction in AUC (0.707), the score maintained robust performance in a real-world, high-risk population, highlighting its potential for scalable implementation in large health systems. Abbreviations as in Figure 1.

• 

  Central Illustration. A New Risk Score to Predict Heart Failure Among Survivors This illustration summarizes the CHERISH risk score, a novel tool designed to predict heart failure in cancer survivors. The score incorporates 11 readily available clinical variables, which enables EMR-based incident HF risk assessment at the point of primary care without requiring additional testing or treatment history. The illustration highlights the pathway from cancer survivorship to targeted HF screening, supporting earlier identification of high-risk individuals and proactive cardiovascular care in routine clinical practice. CHERISH = cancer-specific HF prediction from electronic medical records in survivor health care; EMR = electronic medical record.