Comparing CISNET Breast Cancer Models Using the Maximum Clinical Incidence Reduction Methodology

Area of Science:

• Oncology
• Biostatistics
• Health Services Research

Background:

• Collaborative modeling is used to assess cancer screening strategies.
• Model structure and assumptions significantly influence predictions of cancer incidence and mortality.
• This study examines factors affecting breast cancer screening model predictions within the Cancer Intervention and Surveillance Modeling Network (CISNET).

Purpose of the Study:

• To investigate how pre-clinical duration, screening sensitivity, and treatment improvements affect breast cancer incidence and mortality predictions.
• To understand the influence of model structure and assumptions on collaborative cancer screening model results.

Main Methods:

• Utilized the Maximum Clinical Incidence Reduction (MCLIR) method to compare 4 scenarios: no screening, one-time perfect screening with perfect treatment, one-time mammogram with perfect treatment, and one-time mammogram with guideline-concordant treatment.
• Compared changes in diagnosed breast cancers and cancer mortality across simplified screening scenarios.

Main Results:

• Models predicted wide ranges in incidence (19%-71%) and mortality reduction (33%-67%) under perfect screening conditions.
• Models assuming earlier tumor inception showed substantially higher incidence and mortality reductions.
• Under realistic conditions (mammogram at age 62 with current treatment), the range of predicted outcomes was considerably smaller.

Conclusions:

• The timing of tumor inception and pre-clinical phase length substantially impact model predictions for clinical incidence and mortality reduction.
• This finding will enhance transparency in future CISNET breast cancer analyses.
• The MCLIR approach can improve the interpretation of model variations and be applied to other disease screening settings.