¹OFFIS - Institut für Informatik, Oldenburg, Germany
²Carl von Ossietzky Universität Oldenburg, Fakultät VI – Medizin und Gesundheitswissenschaften, Department für Versorgungsforschung, Abteilung für Assistenzsysteme und Medizintechnik, Oldenburg, Germany
³Epidemiologisches Krebsregister Niedersachsen (EKN), Registerstelle, Oldenburg, Germany
⁴Klinische Landesauswertungsstelle Niedersachsen (KLast), Oldenburg, Germany

Text

Introduction: Lower Saxony comprises nine metropolitan areas and numerous sparsely populated regions. Approximately 75% of the population live in small town (<50,000 inhabitants), suburban, and rural areas [1]. Cancer burden can vary over small geographical areas, given different distribution of risk factors [2]. By displaying cancer incidence information spatially, spatial patterns can be better identified [3]. However, when comparing small areas, local dependencies need to be considered. For this reason, small area disease mapping with Bayesian spatial smoothing was introduced as a methodological solution [2]. Using a Bayesian Conditional Autoregressive (CAR) modeling approach, a new cancer atlas has been developed at the level of regional units (RUs) with a population ranging from 5,000 to over 530,000 inhabitants.

State of the art: In recent years, Bayesian spatial models have been used to model cancer incidence at the small area level. Two examples of cancer atlases have been developed in Germany. Both, the “GEKID Kreisatlas” and the Cancer atlas Schleswig-Holstein deployed the most prominent approach of Besag, York, and Mollié (BYM) [4]. More recently, the Australian Cancer Atlas and the Dutch cancer atlas by IKNL used CAR models with the Leroux prior [2]. In each implementation, neighboring areas were defined by Queens adjacency matrices.

Concept: Incidence data for 2013–2022 were retrieved from the Epidemiological Cancer Registry Lower Saxony (EKN), covering 366 regional units. Analyses included sex- and age-specific cancer data for among others overall cancer (ICD-10 C00–C97, excluding C44) and specific entities such as lung (C33+C34), breast (C50), prostate (C61), and mesothelioma (C45). Standardized incidence ratios (SIRs) were calculated by comparing observed versus expected cancer cases, with expected cases constructed through indirect standardization to minimize sampling variations [5].

Following the more recent approaches, we used Bayesian spatial models with the Leroux prior to estimate the incidence of cancer [2]. Markov Chain Monte Carlo methods produced posterior, smoothed SIRs for each RU. The median smoothed SIR and corresponding 95% credible intervals were reported to account for uncertainty. All analyses were performed in R version 4.4.2 using packages CARBayes (version: 6.1.1), sf (version: 1.0-19) and shiny (version: 1.10.0). RUs were created through aggregation of municipal VG250 data from the Federal Agency for Cartography and Geodesy (BKG).

Implementation: The development process and adjustment of statistical analyses were supported by a visualization in R Shiny. However, the interactive web version targeting the public is currently under development using several web techniques such as React and Leaflet. Future versions of the cancer atlas Lower-Saxony plan to include survival measures. The atlas will be updated annually, to improve research on regional cancer burden.

Lessons learned: Bayesian spatial models with the Leroux prior demonstrated robustness in mapping cancer incidence and visualized relevant spatial clusters. Using an R Shiny for preliminary result visualization improved the development process.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.

References

[1] Carlow VM, Mumm O, Neumann D, Schmidt N, Siefer T. TOPOI Mobility: Accessibility and settlement types in the urban rural gradient of Lower Saxony–opportunities for sustainable mobility. Urban Plan Transp Res. 2021;9(1):207–32.
[2] Duncan EW, Cramb SM, Aitken JF, Mengersen KL, Baade PD. Development of the Australian Cancer Atlas: spatial modelling, visualisation, and reporting of estimates. Int J Health Geogr. 2019 Dec 1;18(1):21.
[3] Goodwin S, Saunders T, Aitken J, Baade P, Chandrasiri U, Cook D, et al. Designing the Australian Cancer Atlas: visualizing geostatistical model uncertainty for multiple audiences. Journal of the American Medical Informatics Association. 2024 Nov 1;31(11):2447–54.
[4] Besag J, York J, Mollié A. Bayesian image restoration, with two applications in spatial statistics. Ann Inst Stat Math. 1991 Mar;43(1):1–20.
[5] Pickle LW. A history and critique of U.S. mortality atlases. Spat Spatiotemporal Epidemiol. 2009 Oct;1(1):3–17.