Identifying intersectional groups at risk for missing breast cancer screening: Comparing regression- and decision...

Area of Science:

• Public Health
• Epidemiology
• Health Services Research

Background:

• Malignant neoplasm of the breast is a leading cause of death for women in Germany.
• Nationwide breast cancer screening (BCS) programs aim for early detection but face challenges with participation rates and socio-demographic inequalities.
• Identifying high-risk groups for non-participation is complex due to intersecting disadvantages.

Purpose of the Study:

• To identify intersectional groups of women at higher risk of not attending BCS in Germany.
• To compare the effectiveness of an evidence-informed regression strategy versus a decision tree-based regression strategy for identifying these groups.

Main Methods:

• Analysis of data from the German 2019 European Health Interview Survey (N=23,001).
• Two logistic regression models were developed: one evidence-informed and one decision tree-based (Classification and Regression Tree).
• Both models used cross-classified intersectional groups based on PROGRESS-Plus characteristics, adjusted for age.

Main Results:

• The evidence-informed approach identified low-income women born outside Germany, living rurally, and not cohabiting as high-risk (OR=9.48).
• The decision tree approach identified widowed women living alone or with dependents, in specific federal states, as high-risk (OR=3.43).
• The decision tree model demonstrated higher discriminatory accuracy (AUC=0.6726 vs 0.6618) and identified nuanced at-risk groups.

Conclusions:

• Decision tree-based regression offers superior accuracy in identifying intersectional groups at risk for non-participation in breast cancer screening.
• This approach can reveal under-studied populations and inform targeted interventions to improve BCS uptake and reduce health inequalities.
• Findings highlight the need to consider complex social dimensions beyond single inequality factors for effective public health strategies.