Title: AI Preparedness for Cancer Registries

Speaker: Dimitris Katsimpokis, PhD, Netherlands Comprehensive Cancer Organisation (Comprehensive Cancer Organisation of the Netherlands - IKNL)

Abstract: Given the forecasted rise of cancer incidence within the next decades worldwide, AI can be a vital tool to accelerate the work of cancer registries and therefore reduce the impact of cancer. The talk will focus on several on-going projects within IKNL related to AI preparedness and cancer registries: federation, synthetic data, Machine Learning (ML) prediction modelling and Large Language Models (LLMs). Specifically, the talk will outline IKNL’s open-source tool for federated analysis (vantage6) delving into its architecture and analyses. Subsequently, we will focus on aspects of synthetic data generation (e.g., predictive power) used for research purposes by looking at survival classification of lung cancer synthetic data. Then, we will turn to two projects on prediction of time-to-event (survival) modelling, first on ovarian cancer progression-free survival prediction (comparing ML with traditional approaches), and second on training LLMs to predict overall survival in breast cancer synthetic data, the latter being a joint project with the Joint Research Center (JRC) of the European Commission. Lastly, the talk will present recent efforts to create and validate a knowledge-based chatbot helper for AI-assisted cancer registration. In the end, we will discuss the challenges of using AI in a cancer registry setting (e.g., issues related to privacy, infrastructure).

Short Biography: Dimitris Katsimpokis, PhD, is a Senior Clinical Data Scientist at the R&D Department of the Comprehensive Cancer Organisation of the Netherlands (IKNL). To support IKNL’s overall goal to reduce the impact of cancer, his research focuses on accelerating the work of the National Cancer Registry (NCR) through the use of AI in various domains: ML-based prediction modelling of cancer survival, synthetic data generation and quality control, as well as cancer knowledge representation with the help of Retrieval Augmented Generation (RAG) and Large Language Models (LLMs).

Apart from AI, his work also includes: causal inference on the role of socio-economic status in accessing expensive lung cancer medication in the Netherlands, forecasting cancer survival in the Netherlands until 2045, developing a Python package for missing value imputation implementing a generalization of the popular MissForest algorithm, as well as developing hierarchical Bayesian models in STAN. His work has been published in Journal of Clinical Epidemiology, Quality of Life Research and Psychonomic Bulletin and Review.