Risk Prediction in Cardio-Oncology: Conceptual and Methodological Considerations: JACC: CardioOncology...

Area of Science:

• Cardio-oncology
• Biostatistics
• Clinical Epidemiology

Background:

• Cardio-oncology involves managing cardiovascular disease in cancer patients.
• Risk prediction models are crucial for guiding clinical decisions in cardio-oncology.
• Methodological challenges specific to cardio-oncology necessitate careful model development.

Purpose of the Study:

• To provide a state-of-the-art review on developing risk prediction models in cardio-oncology.
• To illustrate how prediction objectives, index dates, and time horizons align with clinical decisions.
• To offer a framework for rigorous and clinically meaningful risk prediction tool development.

Main Methods:

• Review of case scenarios to align prediction objectives with clinical decisions.
• Summary of considerations for data source, population, sample size, and variable selection.
• Discussion of challenges in incorporating treatment data (e.g., immortal time bias).
• Review of various modeling approaches (regression, competing risk, dynamic, machine learning).
• Outline of best practices for model evaluation (discrimination, calibration, clinical utility).
• Principles for implementation (workflow integration, transparency, updating).

Main Results:

• Aligning prediction objectives, index dates, and time horizons is critical.
• Careful consideration of data sources, population, sample size, and variables is essential.
• Addressing challenges like immortal time bias and confounding by indication is necessary.
• Various modeling techniques and evaluation metrics are available for model development.
• Implementation requires workflow integration, transparency, and ongoing updates.

Conclusions:

• A comprehensive framework is presented to support the development of robust cardio-oncology risk prediction models.
• These models can guide prevention, monitoring, and treatment decisions for cancer patients with cardiovascular disease.
• The review emphasizes the need for rigorous methodology and clinical relevance in cardio-oncology risk prediction.