The development of statistical data analysis methods helps us gain a better understanding of the causes of cancer and epidemiology, and of cancer screening in Finland.
STATISTICAL METHODS FOR CANCER RESEARCH
Cancer research is supported and facilitated by using the appropriate statistical methods. The assessment of cancer survival is a key part of methodological cancer research. It has focused on more reliable methods of the studies on long-term survival and better computational regional comparative methods. Methods based on computational statistics have also been developed for assessing population attributable fractions.
The modelling of the hidden factors of cancer was investigated in a cooperative project with Aalto University, in addition to finding new ways to evaluate the overdiagnosis associated with screening. Also, new methods of assessing the heritability and familial aggregation of cancer are studied and possibilities for utilizing self-learning computational statistics in cancer coding are currently under investigation.
STATISTICAL METHODS FOR CANCER REGISTRATION
We are also investigating the use of machine learning to support the registration of cancers. Here the main issues are the processing of linguistically challenging texts, producing structured data based on the processed texts, and combining different data sets using statistical methods. Machine learning methods free up registration workers to focus on tasks where a human is most needed when a part of the cancer registration tasks can be performed reliably by a machine.
All of the projects described above are currently ongoing.
Most important publications:
Lead researcher: Janne Pitkäniemi
Research team: Karri Seppä, Joonas Miettinen (machine learning in cancer registration), Tiina Hakanen, Matti Rantanen, Heidi Ryynänen
Funding: Cancer Society of Finland, Cancer Foundation of Finland.
Collaborating institutions: University of Helsinki / Department of Computer Science, Aalto University, University of Oulu, Department of Public Health