ACTRIMS 2025: Combining risk scores may accurately predict MS
Using genetic plus electronic health data found to have accuracy over 90%
A new model that combines genetic and symptom-based risk scores to predict the development of multiple sclerosis (MS) could help to accelerate the disease’s diagnosis, and allow patients to receive earlier treatment, a team of U.S. researchers noted in a study.
In a presentation detailing this work at this year’s Americas Committee for Treatment and Research in Multiple Sclerosis (ACTRIMS) Forum, Alyks Linerud, a research assistant at Brigham Young University in Utah, noted “a pressing need for early, accurate, and minimally invasive tools for disease prediction” in MS.
The researchers noted, in their study, that a symptom-based risk score derived from electronic health records, which is already in use, is quite good at predicting MS in patients of different ancestries. However, adding a genetic risk score adjusting specifically for a person’s ancestry was found to further refine the model, and helped to identify patients with an accuracy of 91% to 99%.
According to the researchers, the addition of the genetic risk score also may allow MS to be detected in people who are younger, and those who have not yet experienced many symptoms or accumulated doctor visits.
“Our goal is to create a prediction model that could be used at the primary care stage to identify and screen individuals at high risk for multiple sclerosis, and thereby refer appropriate individuals to neurology clinics earlier,” Linerud said in the presentation at the ACTRIMS Forum 2025, held Feb. 27-March 1 in West Palm Beach, Florida, and virtually.
Linerud’s talk was titled “Predicting Multiple Sclerosis (MS): a Genetic and Phenotypic Risk Score Model for Disease.”
Researchers tested utility of risk scores in people of different ancestry
In individuals with MS, there’s frequently a long gap between the onset of the disease’s first symptom and a multiple sclerosis diagnosis. While this delay seems to have become shorter in more recent years, it’s not uncommon for patients to wait several months to years before getting a final diagnosis.
Evidence suggests that starting treatment as early as possible can improve the long-term outcomes of patients.
“Even two years’ delay in diagnosis and treatment can greatly impact disability decades later,” Linerud said.
Here, a team led by scientists at Brigham Young aimed to assess whether genetic and clinical data could be leveraged to identify individuals with MS earlier and shorten the time to diagnosis. The team used genetic data and electronic health records from the All of Us research platform, which stores health data from people across the U.S. with an aim toward improving individualized care.
The study included about 1,400 people with MS, alongside 117,000 controls. Most participants were of European ancestry, followed by those with predominantly African and mixed American ancestry. The last group included people with different proportions of European, Sub-Saharan African, and Native American ancestry.
All participants had whole-genome sequencing data, which provides information about the entire set of genes in an individual, and helps identify gene variants, or mutations, that may be disease-related.
These data were compared with a recent genome-wide association study by the International Multiple Sclerosis Genetics Consortium, performed among individuals of predominantly European ancestry. That study confirmed more than 200 variants as risk factors for MS, with 400 additional suggestive variants.
Now, the researchers specifically tested whether these variants could be used to build a genetic risk model to identify MS patients in their population. The team found that a model that used the 200 confirmed variants was better at predicting MS than a model including all 600 established and suggestive variants.
Still, Linerud said, both models were found to “fail to perform at desirable standards, showing the need for factoring more than genetics into risk score models.”
Combining genetic and clinical data improved model’s predictive ability
In a first attempt at improving the model, the team developed a genetic risk score using the 200 variants that took into account individual ancestry. For instance, the HLA DRB1*1501 variant is known to confer a strong risk for MS in a European population, but not in an African American population — and the model adjusted the impact of each variant in each genetic ancestry.
“This local ancestry GRS [genetic risk score] is the first of its kind in any disease,” Linerud said.
While the adjusted model worked slightly better for the predominantly African and European genetic ancestry populations, its overall performance did not improve compared with the nonadjusted models, the researchers noted.
The team then developed a risk model based on clinical data. The researchers looked at diagnostic codes in electronic health records that matched symptoms published on the National MS Society website. Each symptom was given a score based on its prevalence in the MS population, and this was used to derive a phenotypic risk score for each patient.
These results showed that the clinician-derived risk scores were highly predictive of MS, and were also more equitable across groups of different ancestries.
This local ancestry GRS [genetic risk score] is the first of its kind in any disease. … This result is exciting for MS prediction.
In a model that included 140 different signs and symptoms across multiple organs, particularly the nervous system, the ability to correctly discriminate patients from controls ranged from 84% in people of European ancestry to 95% in people of mixed American ancestry.
“This result is exciting for MS prediction, as well as the ability to create PheRS [phenotypic risk scores] for other complex diseases without labor-intensive manual review of symptoms,” Linerud said.
The team is now evaluating the benefits of combining genetic scores with symptom-based risk scores. A first analysis has already demonstrated that the model’s predictive ability was further improved when the genetic and clinical data were combined, reaching an accuracy of 99% in people of mixed American ancestry.
“The combined model shows great promise for predictive ability in diverse populations,” Linerud said.
Overall, according to the researchers, “combination genotype-phenotype risk models have the potential to aid in early screening and diagnosis of MS.”
About the Author
