From baseline data to outcomes: are AI models really competitive?
Mat/06 - Probabilità e Statistica
The author will provide some ideas on the role of statisticians in the pharmaceutical industry and present the results of the project she worked on during the recent internship at the Department of Clinical Statistics Europe of Bayer HealthCare. In particular, in clinical trials a huge amount of information is routinely collected per subject, but only a small fraction of these data is used in standard analyses. The success of Artificial Intelligence across different fields encourages adopting AI-based methods for extracting deeper insights also from clinical data. The author developed multivariate predictive models for time-to-event outcomes (overall survival) to understand if there is valuable information not accounted for by standard methodology. Traditional Cox models using typical low-dimensional, standard clinical predictor variables were compared with machine learning methods (random forests) and neural networks architectures, using the totality of collected baseline data from a randomized multicenter phase III trial.
20-03-2019, h. 16.30
Aula Lagrange, Dipartimento di Matematica "G. Peano"