Breaking barriers in survival analysis: Unlocking the potential of additive hazards models through maximum likelihood The dominant model for time-to-event data is Cox’s proportional hazards model. It is immensely popular because it yields a single effect-measure for the effect of a prognostic factor on the time-to-event through a hazard ratio. When the proportional hazards assumption is invalid, estimates based on the model are severely biased. The additive hazards model is accommodates time-varying effects in a natural way. In this project we develop methodology for additive hazards models based on maximum likelihood, including cross-validation and regularized estimates based on penalized likelihood.

Decision support algorithms can be very useful, especially in important areas like healthcare. But they also pose the risk of propagating historic biases. This project aims to make such algorithms more transparent and reliable. We will develop a new statistic that explains to a particular person seeking advice from the algorithm, how many similar persons contributed data to the algorithm. This allows citizens to judge for themselves whether an algorithm sufficiently applies to their individual situation. Knowing this, citizens can choose if they want to use the algorithm in their decision making or not.

Flexible and user-adaptive statistical inference In this proposal we want to develop mathematical theory for testing many hypotheses simultaneously. Our new theory brings a threefold flexibility into the research: in terms of sample size, analysis and model choice. The researcher may look at the data and the intermediate results, and decide on that basis to add more data, to add more hypotheses – or focus on the most promising ones – and she does not have to restrict herself to a specific pre-specified model. In the mean time our methods retain strong statistical error guarantees.

Donor organs are scarce, and it is not possible to transplant all patients. Currently, the sickest patients are prioritized to receive organs over others who are less sick. However, less sick patients will live longer after transplantation. The aim of this study is to investigate how to distribute organs (livers) so that everyone benefits the most.

In our NWO-Complexity project "Ecology meets human health" we investigated how an ecological view of human microbiota aids our understanding of these miniature ecosystems in relation to health and disease. We show how microbial interactions can be detected by patterns of interrelationship between bacteria. In applications to intestinal diseases, we show how new models and analytical techniques help us characterize disease and describe and predict individual response to treatment. In the future, these insights may lead to the development of practical microbiome-based interventions.