An optimization framework to guide the choice of thresholds for risk-based cancer screening

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Risk stratification for cancer screening often overlooks model performance alongside intervention availability.
• Artificial intelligence (AI) models offer potential for improved cancer risk assessment but require careful integration into screening programs.

Purpose of the Study:

• To develop and apply a framework for selecting risk groups by considering both AI model performance and available interventions.
• To guide breast cancer screening intervals using an AI model to minimize advanced cancer incidence.

Main Methods:

• Developed a framework using linear programming to define risk groups that minimize expected advanced cancer incidence under resource constraints.
• Applied the framework to breast cancer screening intervals using an AI model, with performance estimated from a case-control study and other parameters from screening trials.

Main Results:

• The AI model recommended tailored screening intervals: 1 year for the highest 4% risk, 3 years for the middle 64%, and 4 years for the lowest 32%.
• This approach was expected to reduce advanced cancer diagnoses by approximately 18 per 1000, maintaining similar average screening frequency.
• Sensitivity analyses indicated robustness of threshold choices to model parameters, though advanced cancer reduction estimates require further validation.

Conclusions:

• The developed framework effectively defines risk thresholds for risk-adapted screening, aiming to reduce the population health burden of cancer.
• Further evaluation through health-economic modeling and real-world studies is recommended to validate the framework's impact on advanced cancer reduction.