Bernasconi A, Zanga A, Trama A, Guzzinati S, Andreotti A, Ada WG. and ROSANNA validation WG.
49th Annual Meeting of the Group for Cancer Epidemiology and Registration in Latin Language Countries (GRELL)
Porto, 27-30 maggio 2025
Abstract
Objectives: Thanks to their ability to link to cancer cases several administrative data, population-based cancer registries (CRs) have the unique opportunity to study long-term outcomes in cancer survivors. The objective of this work is to discuss the use of a causal model to study 5-years risk of cardiovascular diseases (CVDs) in young women surviving breast cancer; with a focus on the issues of external validity and generalizability of the results when transporting its results to patients living in another area.
Methods: To train a causal bayesian network model, we fused a clinical and a population-based cohort of 1375 young women (18-39 years old) who survived at least 1 year of breast cancer, living in the Lombardy region in Italy. The model included information on cancer prognostic factors, treatments, and some major cardiovascular risk factors (i.e. hypertension, dyslipidemia and type 2 diabetes). To externally validate the model, data were collected from 6 additional CRs, covering 5 different regions nationwide.
Results: Comparing the different cohorts, we observed similar distributions of age, histology and major cardiovascular risk factors, while minor differences in the treatment choices (especially in pre-surgical treatments). As expected, CVDs are very rare, ranging from 1% to 4%. Geographic variability in the model predictive ability will be presented using standard classification metrics and interpreted also into the light of health migration for the main breast cancer surgery, that varies widely in the 6 considered areas (from 5% to 30%).
Conclusions: With this use-case, we proposed a generalizable methodological framework to be of interest for researchers working in CR that aim at making epidemiological use of their data. In the next 3 years, this model will be further externally validated and extended in the context of a pilot study nested into the Joint Action PreventNCD in four different European national CR (Denmark, Norway, Belgium and Estonia).