Better patient data improves personalized predictions of multiple sclerosis progression

June 6, 2021

There is now an AI model that can predict disease progression in people with multiple sclerosis. A more accurate prediction of the progression of MS can allow doctors to better prescribe a fitting treatment plan. (Shutterstock)

Researchers have developed a novel approach based on machine learning to project the likely progression of disability from multiple sclerosis, relying on the most complete patient history of any prior strategy, by utilizing information from more than 6,600 MS patients across five continents. The team can generate more precise predictions within milliseconds, potentially enabling better therapeutics for a serious and incurable disease.

The machine-learning innovation was built on data from 14 countries in Europe, Africa, the Middle East, Australia and North America. The group's new artificial intelligence approach was also substantially more accurate than its predecessors. A study detailing the researchers' contribution was published May 18 in Computer Methods and Programs in Biomedicine.

"In the last 10 years, there's been an absolute explosion in machine-learning algorithms," said Thijs Becker, a researcher at Hasselt University, in Belgium, in an interview with The Academic Times

"We focus on state-of-the-art machine-learning methods from the AI world to empower health care. We take off-the-shelves algorithms to address these challenges," added Edward De Brouwer, a Ph.D. student at KU Leuven, in Belgium.

In the case of this disease, AI can help the researchers zero in on unique cases. "We do individualized predictions. It's a step towards precision medicine, so it's not about the average impact of a drug, but how likely it is that the disease will progress in two years, given all your personal history," De Brouwer said.

MS is an autoimmune disease that affects the central nervous system. People with MS experience blurry vision, pain in their limbs, and other neurological effects that stem from a compromised immune system.

In a healthy person, the immune system is something like a personal protector that can ward off disease. In the body of a person with MS, this system is effectively reversed, as the immune system attacks the body. As with many other autoimmune diseases, including Type 1 diabetes and HIV/AIDS, there is no known cure. And though 2.8 million people across the world live with MS, neither doctors nor patients can predict the severity of symptoms — or even what symptoms will be experienced by the people who contract it.

Although people who are diagnosed with MS have dozens of medicines available to modify their disease course or address their symptoms, physicians still face the daunting task of picking the best treatments for a very unpredictable condition. A more accurate prediction of the progression of MS can allow doctors to prescribe a fitting course of therapy without wasting time and potentially causing irreversible damage. The key to effective treatment, then, is figuring out which medicine to provide at which point of the disease.

Recent MS research is mostly focused on clinical data sources, the co-authors note. Although this type of data is accurately reflective of real-world practice, the data is often irregular, because doctors do not always perform all the required tests during every office visit. 

"We looked at the added value of using the whole history of the patient," Becker said. Those patient histories gave the researchers a more holistic view of the disease that is otherwise hard to ascertain. "It's a kind of score that doesn't check something in your blood," Becker added. "It, rather, checks something in your history — something which is very intangible. It's hard to define."

The current study focused on longitudinal patient data from multiple countries to better represent disease progression over time and increase the predictive power of methods for projecting MS progression. All patients had an onset date of the disease of 1990 or later, as well as detailed clinical histories.

The team looked at data from the previous three years to predict the next two years of MS disability progression. Two years is a significant length of time for people living with the disease, who are trying to figure out how to conduct their lives — whether, for example, they should go back to school or buy a handicapped van, Becker noted. 

"Even if our models can't be used in clinical practice, they could be used by patients in order to plan their lives a little bit ahead," De Brouwer said.

De Brouwer sees the team's contribution as a tool to empower doctors by analyzing patient records. He considers it a powerful alternative to bogging down medical staff with a "mind-boggling" amount of data.

He and his colleagues were able to correctly predict the progression of MS in 41% of patients, compared with 30% in traditional models. "This increase in precision leads to a more efficient clinical care as the limited resources of neurologists can be focused on a smaller and more specific subset of patients requiring special attention," the authors noted in the study.

Specific treatments for MS continue to advance. In early 2021, researchers at BioNTech, which is making the Pfizer vaccine for COVID-19, achieved a groundbreaking result by successfully using mRNA technology for MS treatment. Though translating these promising results to humans is not necessarily easy, the team is optimistic about the future of MS research empowered by machine learning.

"What I hope to see more is clinical trial studies implementing AI models in the next five years," Becker said. "Some people think that doctors will be replaced by AI, which I think is a very strange idea. What you will hopefully see is that doctors will get 10% or 20% better at their job, which is already fantastic. That is already within reach, I think."

"Our research group has an initiative for trying to update clinical guidance for drugs in MS, especially in the COVID-19 pandemic, because MS is this immune disease," De Brouwer said. "We want to be as fast as possible in finding the right treatment, because currently, that's the whole weak point in the clinical treatment of MS."

The study, "Longitudinal machine learning modeling of MS patient trajectories improves predictions of disability progression," published May 18 in Computer Methods and Programs in Biomedicine, was authored by Edward De Brouwer and Yves Moreau, KU Leuven; Thijs Becker and Liesbet Peeters, Hasselt University; Eva Kubala Havrdova, Charles University in Prague and General University Hospital; Maria Trojano, University of Bari; Sara Eichau, Hospital Universitario Virgen Macarena; Serkan Ozakbas, Dokuz Eylul University; Marco Onofrj, University G. d'Annunzio; Pierre Grammond, CISSS Chaudire-Appalache; Jens Kuhle and Ludwig Kappos, University Hospital Basel; Patrizia Sola, Azienda Ospedaliera Universitaria; Elisabetta Cartechini, Azienda Sanitaria Unica Regionale Marche; Jeannette Lechner-Scott, University of Newcastle; Raed Alroughani, Amiri Hospital; Oliver Gerlach, Zuyderland Ziekenhuis; Tomas Kalincik, Royal Melbourne Hospital and University of Melbourne; Franco Granella, University of Parma; Francois GrandMaison, Neuro Rive-Sud; Roberto Bergamaschi, IRCCS Mondino Foundation; Maria José Sá, Centro Hospitalar Universitario de São João and University Fernando Pessoa; Bart Van Wijmeersch, Rehabilitation and MS-Centre Overpelt and Hasselt University; Aysun Soysal, Bakirkoy Education and Research Hospital for Psychiatric and Neurological Diseases; Jose Luis Sanchez-Menoyo, Hospital de Galdakao-Usansolo; Claudio Solaro, Mons. L. Novarese' Hospital; Cavit Boz, KTU Medical Faculty Farabi Hospital; Gerardo Iuliano, formerly Ospedali Riuniti di Salerno; Katherine Buzzard, Box Hill Hospital; Eduardo Aguera-Morales, University Hospital Reina Sofia; Murat Terzi, 19 Mayis University; Tamara Castillo Trivio, Hospital Universitario Donostia; Daniele Spitaleri, Azienda Ospedaliera di Rilievo Nazionale San Giuseppe Moscati Avellino; Vincent Van Pesch, Cliniques Universitaires Saint-Luc; Vahid Shaygannejad, Isfahan University of Medical Sciences; Fraser Moore, Jewish General Hospital; Celia Oreja-Guevara, Hospital Clinico San Carlos; Davide Maimone, Garibaldi Hospital; Riadh Gouider, Razi Hospital; Tunde Csepany, University of Debrecen; and Cristina Ramo-Tello, Hospital Germans Trias i Pujol.

We use cookies to improve your experience on our site and to show you relevant advertising.