Markov for Screening [Marie Skłodowska-Curie Postdoctoral Fellowship]

Cancer screening programs increasingly rely on large-scale longitudinal health data, including biomarkers, imaging data, and patient follow-up information. Current statistical models can describe disease progression but often fail to fully capture the two-way interaction between disease stages and biomarker evolution. This limitation may reduce the accuracy of risk prediction and screening recommendations. The project builds on the Finnish ProScreen prostate cancer screening trial, one of the largest screening studies worldwide integrating MRI and biomarker data.

The MarkovForScreening project aims to develop a new generation of interpretable statistical and machine learning methods based on Bidirectional Multistate Markov Models (Bi-MMMs). These models will jointly analyze disease progression and biomarker evolution to improve personalized risk prediction, screening strategies, and decision-making. The project combines biostatistics, machine learning, Bayesian inference, and geometric data analysis to address modern challenges in health data science.

The project will provide healthcare professionals, policymakers, and citizens with more reliable evidence on the benefits and risks of cancer screening. By improving prediction accuracy and reducing unnecessary interventions, it seeks to support more personalized and cost-effective healthcare. The resulting open-source methods will be applicable beyond prostate cancer, including other diseases such as breast cancer and diabetes, contributing to advances in public health and personalized medicine across Europe.

Samuel Gruffaz (PI)

samuel.gruffaz [at] tuni.fi