Using decision trees to identify intersectional subgroups at risk for cancer screening non-attendance: three European case studies

As in many relevant public health areas, attendance in cancer screening programs is stratified by social dimensions, yet current additive approaches fail to capture the complexity of discrimination leading to health inequalities. In fact, social dimensions interact, shaping experiences of discrimination in accessing cancer screening. This dissertation advances the study of complex social inequalities by developing different analytical strategies that explore the use of decision trees under the framework of intersectionality to identify subgroups at risk of non-attendance. Three European case studies were analysed: breast cancer screening (BCS) in Germany, BCS in Spain, and colorectal cancer (CRC) screening in Sweden.
Three analytical strategies were explored: (i) comparing decision tree-based and evidence-informed approaches (BCS Germany), (ii) using decision trees to reduce intersectional complexity (CRC Sweden), and (iii) employing decision trees as predictive tools (BCS Spain). Findings reveal key Individual-regional interactions. In Spain, regions significantly influenced BCS attendance, reflecting economic disparities and screening program timelines. In Germany, partnership cohabitation was a protective factor, while certain regions had higher non-attendance risks. In Sweden, organized screening programs mitigated inequalities, while opportunistic screening revealed disparities based on gender, migration background, and income.
This dissertation contributes to identifying intersectional subgroups at risk of non-attendance by inductively selecting social dimensions and modelling non-linear and nuanced interactions between categories across subgroups. It also advances the methodological field of quantitative intersectionality by proposing decision trees to simplify intersectional complexity, improving results' interpretability. Overall, it enhances understanding of cancer screening inequalities and proposes methodological tools for public health research.