Publikation
Harmonizing regional cancer registry data to facilitate Germany-wide epidemiological analyses
Sebastian Germer; Markus Sauerberg; Ole Johanns; Christiane Rudolph; Christopher Gundler; Stefan Meisegeier; Khalid Abnaof; Soo-Zin Kim-Wanner; Anna Krauß; Manuela Langholz; Sabine Luttmann; Natalie Rath; Katharina Rausch; Alexander Katalinic; Heinz Handels; Alice Nennecke; Henrik Kusche
In: International Journal of Medical Informatics (IJMEDI), Vol. 220, Page 106586, Elsevier, 11/2026.
Zusammenfassung
Background
Data heterogeneity remains a major limitation in epidemiological and clinical analyses of cancer. In Germany, cancer registration is organized at the federal state level. While standardized oncological datasets exist for reporting and data transfer, the use of free-text fields and open regular expressions introduces heterogeneity within and across registry datasets. This variability complicates multi-registry analyses and necessitates extensive data cleaning, thereby limiting the reproducibility and comparability of population-based cancer studies.
Methods
We implemented a pipeline that converts XML-based cancer registry records into a standardized tabular format and harmonizes variable expressions using both publicly available and internally curated and published reference data. The pipeline was designed for independent implementation by cancer registries and supports reproducible, large-scale data integration. To evaluate the pipeline, we performed analyses regarding the number of unique values as well as conformant values before and after pipeline application. Also, we designed a technical feasibility study focusing on early-stage non-small cell lung cancer.
Results
The pipeline was evaluated using data from 14 German cancer registries covering four cancer types: lung, thyroid, breast, and non-Hodgkin lymphoma. The approach improved overall data adherence to reference lists, reduced inconsistencies, and increased statistical power by combining data from over one million cancer patients. The harmonized dataset also enables meaningful analyses of rare cancer entities or subgroups that would otherwise be underrepresented in single registries, as can be seen in the conducted feasibility study.
Conclusion
By addressing data heterogeneity as a key barrier in cancer epidemiology, the proposed pipeline enhances both data conformity and analytical usability. It facilitates comprehensive, cross-registry analyses with greater statistical power and supports more reliable, generalizable insights into cancer trends across Germany. It can be extended to additional cancer types in the future.