Tool Shows Promise for Individualized Decision Support on Stopping Disease-Modifying Therapy for Multiple Sclerosis

A Cleveland Clinic-developed individualized decision support tool has been internally and externally validated for predicting risk of recurrent disease activity after discontinuation of disease-modifying therapy for multiple sclerosis (MS). The tool, developed using machine learning, was presented at ECTRIMS 2025 in Barcelona, Spain.

“The tool demonstrated excellent predictive agreement across validation in both our internal cohort and the population of the multicenter DISCO-MS clinical trial,” says Marisa McGinley, DO, a neurologist with Cleveland Clinic’s Mellen Center for Multiple Sclerosis Treatment and Research who presented the validation findings at the ECTRIMS congress. “We are now implementing the tool in the electronic health record at Cleveland Clinic and evaluating it in a clinical trial for use in supporting a personalized approach to discontinuing disease-modifying therapy.”

The instrument is the first known validated tool of its kind to predict risk of post-discontinuation disease recurrence across the spectrum of disease-modifying therapies (DMTs) for MS.

The case for personalizing DMT discontinuation

Despite the proliferation of DMTs for MS over the past three decades and their demonstrated efficacy in keeping MS disease activity at bay, DMTs do carry risks and can involve considerable economic costs.

“Many patients ask, ‘Do I have to be on my DMT forever? When might I come off my drug?’,” Dr. McGinley says. She notes that multiple observational studies have suggested that DMTs can be safely stopped at some point, particularly in older patients with stable disease. However, two randomized controlled trials examining DMT discontinuation — DISCO-MS and DOT-MS — either failed to show noninferiority versus continuation or showed signs of increased recurrence of disease activity.

“These mixed results left us with no clear guidance on when we can or should stop these medications,” Dr. McGinley explains. “The evidence to date suggested that discontinuation calls for a very personalized approach. That led us to explore developing a risk calculator that provides personalized decision support as opposed to a blanket recommendation for DMT discontinuation that applies to everyone.”

Study design and findings

She and Cleveland Clinic colleagues drew on the Mellen Center’s large patient volume to develop an algorithm to predict a patient’s risk of recurrent inflammatory disease activity — defined as relapse, a new T2 lesion or a gadolinium-enhancing lesion — if they discontinued their DMT compared with continuing the DMT.

The algorithm was built using a retrospective cohort of adults with MS on DMT who completed at least two MS office visits at Cleveland Clinic from January 2015 to July 2023. DMT continuers and discontinuers were matched 5:1 based on demographic and disease characteristics, yielding an internal cohort of 1,104 patients — 920 DMT continuers and 184 discontinuers.

Using patient outcomes after DMT discontinuation in this cohort, the team tested various algorithms to assess their predictive ability. The best-performing algorithm was developed using a machine learning method known as a random survival forest (RSF) model with repeated five-fold cross-validation with variable selection. Refinement of the algorithm resulted in a final model with 12 patient variables included for risk prediction.

The model was then assessed for predictive performance at two years after DMT continuation in this internal cohort and then in the dataset of the 259-patient DISCO-MS multicenter trial, which randomized patients 1:1 to continue or discontinue DMT.

The RSF model had the same area under the curve (AUC) at two years — i.e., 0.65 — in both the internal cohort and the DISCO-MS cohort. “These matching AUC values indicate excellent agreement between the two populations, which provides a reassuring level of validation for the model,” Dr. McGinley notes.

Across the cohorts, the RSF model predicted a 12.6% risk for recurrent MS inflammatory disease activity for DMT discontinuers. Although the variables that are most determinative for an individual’s specific risk score can vary from patient to patient, the variables that contributed most to risk level overall were as follows:

• Months since any inflammatory activity
• Duration of current DMT use
• Age
• Months since last relapse
• Months since last MRI activity
• DMT efficacy (low, moderate or high)

“What we have is a tool that can, from a mathematical standpoint, predict risk of recurrence after discontinuation for an individual patient that has been both internally and externally validated,” Dr. McGinley says. “Beyond predicting risk of inflammatory activity, it also gives the treating clinician a ranking of the factors that contributed most to the risk score at the individual patient level.”

Next steps and implications

The predictive tool based on the RSF model has just been integrated into Cleveland Clinic’s electronic health record for actual clinical use with patients. Dr. McGinley and her colleagues will be randomizing Mellen Center clinicians to either have access to the tool or not to have access so that they can evaluate whether and how it impacts decision-making around DMT discontinuation.

“We will be doing this in the context of a randomized trial to test our hypothesis that use of the tool will be associated with a higher proportion of patients being safely discontinued from DMT — that is, without recurrent disease activity,” she explains.

If that hypothesis is borne out, the Mellen Center team hopes to disseminate the tool to other health systems and via an online risk calculator. “We’d like to distribute this broadly,” she says, “particularly because we developed the tool to be very generalizable and useable for any MS provider. It’s based solely on information that is readily accessible — straightforward demographic factors and disease history. Although there may be an opportunity to include novel imaging and biomarkers later, we chose to use only information that’s currently available to ensure generalizability.”

And she believes interest in the tool will be high. “Many patients who’ve been on DMT for a while really want something like this,” Dr. McGinley says, citing a prior study she led to assess patients’ perspectives on discontinuing DMTs. “Patients in that study told us they want a recommendation about whether or not to discontinue that comes from their own clinician and that is concrete, with numbers behind it. This tool shows real potential on both counts.”

“These types of tools get us closer to personalized treatment decisions for discontinuation by using data to inform shared decision-making in the clinic,” adds co-investigator Daniel Ontaneda, MD, PhD, a neurologist with the Mellen Center. “Clinical trials might answer the question for a group, but not for an individual. This is where this type of tool is most useful.”