Multiple Sclerosis: using machine learning models to predict the course of the disease

A multidisciplinary study, involving four different departments of Sapienza University of Rome, identified a new paradigm aimed at predicting the medium-term course of multiple sclerosis by using machine learning algorithms and data already available in clinical routine. Results have been published on the journal PLoS ONE

Multiple Sclerosis (MS) represents the main cause of progressive neurological disability among young people, primarily affecting individuals between 20 and 50 years of age, with a very high human and social cost. The disease usually starts with a Relapsing-Remitting form, that alternating between phases of symptom worsening and remission, which gradually develops into a secondary progressive form in which disability accumulates. 

The natural course of the disease is extremely variable depending on the individual and no reliable prediction model is currently available. This inability to predict the course of the disease is very limiting since there are several therapies that may prevent or delay relapses also for a long time. In general, the more effective the therapy, the more severe the undesired effects. Reliable prognostic predictors would be of great importance. They would help differentiate the therapies based on the predicted aggressiveness of the illness, reserving high-impact therapies only for those patients with an increased risk of progression of the disease.

As in other medical fields, the study on multiple sclerosis has also started to exploit the ability of AI approaches, trying to investigate the variables that best predict disease evolution in time. Researchers are using approaches based on machine learning algorithms, but, to date, none of the proposed prognostic methods have achieved reliable levels of “clinical grade”. Moreover, several of these works rely on high-technical data, rarely used in the average clinical routine. Therefore, even if the predictive efficacy may be improved this will remain a factor likely to limit the widespread application of this method.

A multidisciplinary team of Sapienza University, by adopting an interesting change of perspective, performed a study aimed at predicting the medium-term course of multiple sclerosis by using data usually available in clinical routine. For this project, physicists, engineers, and experts in machine learning and decision support systems collaborated with neurologists and neurophysiologists from four different departments of Sapienza.

Following an initial stage, aimed at finding a common language able to enable an efficient level of communication between people coming from different fields, the team operated using clinical records of patients of Sant’Andrea university hospital, Rome, Italy.

The data obtained were then made usable for the machine learning models and analysed adopting two different methods: one based on the data of the most recent available visit (Visit-Oriented) and the other on the full clinical history of each patient (History-Oriented). Outcomes show how clinical data may be enough to reliably predict the course of multiple sclerosis in each individual. Consequently, this work can be the first step in using machine learning models in any hospital. Even though the models' levels of prediction need further improvement, they indicate the path to follow to improve the analysis outcome, by also maintaining an active collaboration between those who collect data and those who analyse them.

The following departments and researchers took part in the study: Department of Computer, Control and Management Engineering Antonio Ruperti, Department of Neurosciences, Mental Health and Sensory Organs, Department of Physics, Istituto dei Sistemi Complessi (ISC-CNR), and Francesca Grassi of the Department of Physiology and Pharmacology Vittorio Erspamer.

 

References: 

Considering patient clinical history impacts performance of machine learning models in predicting course of multiple sclerosis - Ruggiero Seccia, Daniele Gammelli, Fabio Dominici, Silvia Romano, Anna Chiara Landi, Marco Salvetti, Andrea Tacchella, Andrea Zaccaria, Andrea Crisanti, Francesca Grassi, Laura Palagi - Plos One, Published: March 20, 2020 https://doi.org/10.1371/journal.pone.0230219

 

Further Information

Francesca Grassi 
Dipartimento di Fisiologia e farmacologia Vittorio Erspamer
francesca.grassi@uniroma1.it

 

Tuesday, 24 March 2020

© Sapienza Università di Roma - Piazzale Aldo Moro 5, 00185 Roma - (+39) 06 49911 - CF 80209930587 PI 02133771002