Latent Archetypes of the Spatial Patterns of Cancer

Area of Science:

• Epidemiology
• Biostatistics
• Geographic Information Systems (GIS)

Background:

• Cancer atlases reveal geographic variations in cancer risk.
• Analyzing spatial patterns requires correlating them with risk factors, a complex and time-consuming task for epidemiologists.
• Current methods for studying multiple cancers simultaneously are limited to a few known related cancers.

Purpose of the Study:

• To develop an exploratory method for identifying latent spatial risk factors across a large number of diverse cancers.
• To reduce the complexity of analyzing geographic cancer risk data.
• To enable the simultaneous investigation of shared risk factors for multiple, seemingly unrelated cancers.

Main Methods:

• The study proposes a novel method utilizing singular value decomposition (SVD) and nonnegative matrix factorization (NMF).
• This approach is computationally efficient and scalable for large datasets involving numerous regions and cancer types.
• The method was validated through a simulation study and applied to cancer atlas data from multiple countries.

Main Results:

• The proposed method effectively identifies latent spatial risk factors, reducing the number of cancer maps by up to 90% while preserving most spatial variability.
• It allows for the simultaneous analysis of spatial patterns across a wide range of cancers.
• The technique demonstrates computational efficiency and scalability.

Conclusions:

• The new method significantly streamlines epidemiological analysis by reducing the data dimensionality.
• It facilitates the discovery of high-level explanations for cancer risk affecting multiple cancers concurrently.
• This approach offers a powerful tool for understanding complex geographic patterns in cancer incidence.