On Cluster Structures of Finnish Cancer Incidence Data

Abstract

Introduction: The global burden of cancer is increasing. Part of this development is attributable to the estimated growth and aging of the population. In particular, aging is 1 of the main risk factors for cancer. However, there are many other risk factors beyond aging, including certain lifestyle and environmental factors. In addition, changes in diagnostic thresholds, increasing coverage of screening, and other similar factors affect cancer incidence rates. Therefore, even after excluding the effect of aging of the population, cancer incidence rates have not remained constant over time. To study these changes, the focus of this study is to identify and analyze cluster structures of the Finnish cancer incidence data from 1963 to 2023. Methods: To uncover the cluster structures, a proximity measure that is based on the shape of the curves is used. For unstandardized data, the proximity measure is shown to be invariant under simple location shift, and for standardized data, also under simple scaling, making the proximity measure suitable for assessing the similarities or dissimilarities of trends in time. As the group-building algorithm, agglomerative hierarchical clustering, combined with the average linkage method, is used. Results: The cluster structures were identified for 12 different subgroups, determined by age and sex. In many cases, cancers for which there exists a national screening program, including breast and cervical cancer, or an individualized testing tool, including prostate cancer, formed clusters of their own. Melanoma of the skin and lung & tracheal cancer are other 2 cancer types that often separated as their own clusters, possibly due to certain lifestyle factors. Conclusion: The study demonstrates the potential of the proposed proximity in the given context. In addition, the analysis of the cluster structures provides some insight into the Finnish cancer epidemiology.