Biostatistics research @ Radboudumc

Advanced modelling for longitudinal and survival data, including federated inference

Survival analysis is a branch of statistics in which one is interested in the time to a specific event. Examples are the time to death after the diagnosis of cancer, or the time to recovery after a specific infection. One of the major challenges in survival analysis is that for some patients this time at which the event occurs is not observed. This can happen for many reasons.

One of our projects is focused on Bayesian Federated Inference (BFI) for survival models. BFI can be used if data sets in separate centers can not be merged, for instance due to privacy legislation. BFI constructs from local inferences in separate data centers what would have been inferred had the data sets been merged.

Main projects

Collaboration RU science faculty and Radboudumc

Over a period of 8 years, two consecutive PhDs students will support collaborative research across the two RU faculties to act as seed for larger collaborations. In the first PhD project the focus is on the development and application of new mathematical, statistical and computational techniques for inference in case the sample size is small compared to the number of covariates in a statistical model. The second project will focus more on implementation of novel methodology in clinical research practice.

Hanarth Fund: Improved Transparent AI methods for personalized prediction

(2022-2025)

In this project we aim to develop and apply novel methodology for analysing small data sets: correction methods for model overfitting and methodology for doing federated analyses. The latter aims to construct from local inferences in separate centres what would have been inferred had the datasets been merged. Especially if datasets in medical centres are small and cannot be combined due to e.g., privacy legislation, applying this approach improves the accuracy of parameters estimates and prediction. The methodology will be applied to data on salivary gland cancer and other clinical data to obtain new insights in possible predictive factors of these diseases.


Main publications

Jonker MA, Pazira H, Coolen ACC, Bayesian Federated Inference for estimating Statistical Models based on non-shared multicenter data. arXiv preprint, 2024a, accepted by statistics in medicine. (https://doi.org/10.48550/arXiv.2302.07677)

Ramjith J, Bender A, Roes KCB, Jonker MA. Recurrent events analysis with piece-wise exponential additive mixed models. 2022. Statistical Modelling.


GitHub

Pazira H, Massa E, Jonker MA, The R package BFI and a manual: hassanpazira.github.io/BFI and github.com/hassanpazira/BFI, 2024.


People

Marianne Jonker, Jordache Ramjith, Hassan Pazira, Emanuele Massa