Risk Prediction in Cardio-Oncology: Conceptual and Methodological Considerations

Figure 1. Cardiovascular Risk Assessment Across Cancer Trajectories This figure illustrates the need for cardiovascular risk assessment on the basis of cancer prognosis. It contrasts a lower risk patient (case 1) entering long-term survivorship with a higher risk patient (case 2) facing recurrence and ongoing therapy. The figure shows that risk prediction time points (eg, baseline, pretherapy, surveillance) should be tailored to the clinical course, whether remission or chronic disease management. EGFR = epidermal growth factor receptor; NSCLC = non–small cell lung cancer.

Figure 2. Purpose of the Model This figure demonstrates how the clinical purpose drives the design of a prediction model. It shows that the model’s purpose informs definitions of the target population, outcomes of interest, index date, and prediction time horizons. The figure emphasizes that methodological choices should align conceptually with the clinical question to ensure the model is relevant and effective. CV = cardiovascular; LVEF = left ventricular ejection fraction.

Figure 3. Evolving Index Dates and Vantage Points for Risk Assessment in Cardio-Oncology Using case 1, this figure maps the longitudinal evolution of cardiovascular risk assessment. It highlights 3 distinct assessment points: pretreatment baseline, midtreatment reevaluation, and long-term survivorship. The figure illustrates that index dates are not fixed; as patients progress through therapy, prediction objectives shift from preventing acute toxicity to managing long-term survivorship risks, requiring models that can incorporate emerging data and cumulative risks. Abbreviations as in Figure 2.

Figure 4. Model Evaluation: Discrimination, Calibration, and Clinical Utility This figure presents 3 essential pillars for comprehensive evaluation of risk prediction models: discrimination, calibration, and clinical utility. It links statistical performance (discrimination and calibration) with clinical utility (relevance for decision-making). The figure emphasizes that models should not be judged solely by statistical performance; they must also demonstrate practical value in clinical workflows to support patient care.

Central Illustration. The Continuous Evolution of Risk Prediction Models in Cardio-Oncology The illustration depicts the 4-stage cycle required for personalized care: 1) define; 2) develop and validate; 3) evaluate; and 4) implement and monitor. This circular framework emphasizes that risk prediction is a continuous process requiring ongoing monitoring and recalibration rather than one-time, linear activity. The illustration highlights that models should evolve alongside electronic health record data and changing clinical guidelines to maintain accuracy and utility.