Improving the predictions of survival indicators in breast cancer patients through artificial intelligence and probabilistic modeling tools is the goal of ModGraProDep, an innovative system presented in a study published in the journal Artificial Intelligence in Medicine.
Led by lecturer Ramon Clèries, from theUniversity of Barcelona(UB) Department of Clinical Sciences of the Faculty of Medicine and Health Sciences and member of the Oncology Master Plan (ICO-IDIBELL), contributing to this study were researcher José Miguel Martínez Martínez, from theUniversity of Alicante(UA)Public Health Research Group, as well as a large team of experts in epidemiology, oncology and data mining from the Oncology Master Plan-IDIBELL, the UB, the Polytechnic University of Catalonia, the Catalan Institute of Oncology (ICO), the Girona Institute of Biomedical Research (IDIBGI), the University of Girona, the CIBER of Epidemiology and Public Health (CIBERESP, Carlos III Health Institute), the University Hospital Sant Joan in Reus, the ICO Medical Oncology Service in Girona, the Cancer Registries of Girona and Tarragona, and MC Mutual.
The UA researcher, together with a multidisciplinary team of experts in epidemiology, oncology, and statistics, has discussed, compared, and validated different methodological proposals to improve the predictions of survival indicators in breast cancer patients.
Mathematical modeling: ew frontiers in the fight against cancer
One of the applications of numerical modeling for clinical indicators in oncology is the development of predictive models that help oncologists and clinicians to classify and assess future evolution scenarios for cancer patients. In this context, the prediction of patient survival —with specific variables and ages— is decisive for evaluating treatments and identifying subgroups among patients. However, this information must often be estimated by numerical modelling means, given that there is not a sufficient sample size of population to calculate these indicators accurately.
The application of the new ModGraProDep (Modeling Graphical Probabilistic Dependencies) methodology has promoted two research projects coordinated by Professor Mireia Vilardell, from the UB Faculty of Biology Department of Genetics, Microbiology and Statistics (Statistics Section) and researcher Maria Buxó from IDIBGI.
In the first case, ModGraProDep allows us to identify the structure of a database and generate a simulated population of patients with demographic characteristics of the original cohort. With this approach, possible new patient patterns can be identified and indicators calculated (e.g., patients' survival as a function of the values of their variables).
In the second study, ModGraProDep is revealed as a technology capable of assigning values in a probabilistic way in variables from which data collection had not been possible.
The research team has also designed a web application of great clinical interest to obtain a prediction of survival indicators and mortality risk from cancer —and by other causes— of each patient up to a maximum term of twenty years.
Vilardell, M., et al. (2020) Missing data imputation and synthetic data simulation through modeling graphical probabilistic dependencies between variables (ModGraProDep): An application to breast cancer survival. Artificial Intelligence in Medicine. doi.org/10.1016/j.artmed.2020.101875.