Author: Erica Tavazzi

On 15 January 2025, the Aula Magna of the Department of Information Engineering hosted a meeting dedicated to the research efforts of UNIPD within Spokes 2 and 3 of the DARE (Digital Life-Long Prevention) initiative.

During the event, Professor Barbara Di Camillo presented the pilot project in which our research group is actively involved, titled “Digitally-Empowered Management of Type 2 Diabetes: From the Diagnosis to the Prediction of Complications.”

It was a privilege to contribute to this gathering, which highlighted the exceptional quality and scientific relevance of the ongoing projects within the DARE initiative. The exchange of ideas and insights reaffirmed the initiative’s potential to make significant strides in digital health research.

Last week, our PhD student Elena Marinello presented her research titled “Machine Learning Models Highlight the Impact of Pollution and Weather Patterns on Relapse Occurrence in Multiple Sclerosis Patients” at the Artificial Intelligence and Computational Methods for Public Health and the Environment workshop organised as part of the IEEE International Conference on Bioinformatics and Biomedicine (BIBM) 2024.
This study, conducted as part of the H2020 BRAINTEASER project, investigates how environmental factors, including pollution and weather patterns, contribute to predicting imminent relapses in multiple sclerosis. By leveraging clinical and environmental data from the week preceding potential relapse events, this research highlights the role of machine learning in advancing personalised medicine for MS patients.

Last week, our post-doc Alessandro Guazzo attended the 10th International Conference on Machine Learning, Optimization, and Data Science held in Castiglione della Pescaia, Italy. 

Alessandro gave a presentation on his work “Predicting Multiple Sclerosis Worsening Using Stratification-Based and Time-Dependent Variables Extracted from Routine Visits Data” developed within the BRAINTEASER project, focusing on improving the prediction of multiple sclerosis worsening. His talk highlighted innovative stratification-based variable extraction approaches, which he had the opportunity to study during his recent visit to Carnegie Mellon University. 

The 8th plenary meeting of the BRAINTEASER project is currently taking place in Turin (19-20 September 2024). During these two days, all project partners are presenting their updates across the 12 work packages. Our group leads Work Package 7 (WP7), focused on “AI models for disease progression.” The objectives of WP7 include developing models to predict disease progression in amyotrophic lateral sclerosis (ALS) and multiple sclerosis (MS), incorporating new variables at various individual levels that are currently underutilised in the literature, validating these models on external datasets, and using them to create in-silico patient simulators.

In recent months, our team has been working on a task aimed at developing an in-silico patient simulator capable of modelling and simulating disease evolution, specifically focusing on changes in patients’ independence in domains affected by ALS and MS. Today, our post-doc Erica Tavazzi presented the results of these efforts. The models, based on dynamic Bayesian networks and developed using dynamic clinical data from ALS and MS patients, have successfully captured the relationships between key variables over time, allowing for a detailed simulation of disease progression. Moreover, we demonstrated the models’ effectiveness in evaluating the impact of specific biomarkers on disease progression and the efficacy of preventive treatments.

We extend our sincere thanks to the Department of Neuroscience at the University of Turin for hosting and organising this meeting. We are proud to be part of the BRAINTEASER consortium, which aims to harness advanced AI technologies to improve the management and understanding of neurodegenerative diseases, ultimately enhancing patients’ quality of life.

Last week, we had the pleasure of participating in the 15th Conference and Labs of the Evaluation Forum (CLEF), held from 9-12 September in Grenoble, France. As part of the iDPP Challenge, we presented our work titled “Using wearable and environmental data to improve the prediction of amyotrophic lateral sclerosis (ALS) and multiple sclerosis (MS) progression: an exploratory study” (E. Marinello, A. Guazzo, E. Longato, E. Tavazzi, I. Trescato, M. Vettoretti, B. Di Camillo), developed within the BRAINTEASER H2020 project. The paper, included in the CLEF 2024 Working Notes, can be accessed here: https://ceur-ws.org/Vol-3740/paper-125.pdf.

Our study focuses on leveraging sensor data to enhance predictive models for ALS and MS progression, with the aim of assisting clinicians in making more informed therapeutic decisions. We developed machine learning approaches to predict ALSFRS-R scores in ALS patients and relapses in MS patients using wearable and environmental data. The results were promising, demonstrating improved prediction accuracy when sensor data were incorporated.

Participating in the iDPPChallenge provided a great opportunity to share our findings on how data science can enhance clinical decision-making in neurodegenerative diseases

On Sunday, 8th September 2024, the 2nd Stakeholder Meeting of the REDDIE project took place at the IFEMA Convention Centre in Madrid. During the event, our tenure-track researcher and PI, Martina Vettoretti, presented the activities related to Work Package 5 (WP5), Advanced Methodological Frameworks for Enhancing Clinical Trials with Real-World and Virtual Evidence, for which our group is the leader.

The presentation focused on the progress made in two key methodological tasks. The first is the enhancement of methodologies for conducting retrospective observational studies, where we have performed a systematic review to identify and evaluate available methods on a set of artificial datasets. This work aims to define guidelines for selecting appropriate methods in various contexts and develop a fully-automated analysis pipeline to improve reproducibility and scalability.
The second task involves advancing tools for building data-driven simulation models. These models are becoming increasingly important as the simulated data they can create provide a complementary perspective to real-world data and clinical trial data. Simulated data can be generated using computational models, such as dynamic Bayesian networks (DBNs), a technique in which our group has both methodological expertise and practical experience. Within the REDDIE project, we are advancing the use of DBNs to simulate long-term outcomes in patients with type 1 and type 2 diabetes, leveraging RWD sourced from national diabetes registries to build more accurate models.
These efforts will ultimately contribute to more personalised approaches in diabetes management, while also improving the scalability and robustness of the simulation models used in healthcare research.

We extend our gratitude to the organisers for a well-executed and insightful meeting, and for the valuable feedback received from key stakeholders, whose input is essential to the success of this project.