AI exposies what MS disability scales have missed for decades: AI reveals the hidden realities inside identical MS disability scores
The Problem with a Single Number
For decades, multiple sclerosis has been measured by a number.
The Expanded Disability Status Scale (EDSS) has served as the dominant yardstick for evaluating disease progression, determining trial eligibility, and shaping treatment decisions. Its simplicity made it powerful. Its narrowness made it flawed.
Two patients could share the same EDSS score and yet live radically different lives, one struggling with balance and coordination, another battling sensory loss or muscle weakness that quietly erodes daily independence. Clinicians have long known this. Patients have lived it. But the system lacked the tools to quantify that difference.
Now, artificial intelligence is beginning to expose what the EDSS has concealed.
A large-scale analysis published in the Multiple Sclerosis Journal signals a pivotal shift in how disability in MS may be understood, measured, and ultimately treated. By applying AI-based clustering algorithms to thousands of patient assessments, researchers demonstrated that identical EDSS scores can mask profoundly different patterns of functional impairment.
This is not a technical footnote. It is a challenge to the very architecture of modern neurological care.
What the Study Reveals and Why It Matters
Led by Martina Greslin, a neurologist and doctoral researcher at the University Hospital Basel, the study analyzed more than 13,000 clinical assessments from 1,636 patients with secondary progressive multiple sclerosis who participated in the phase 3 EXPAND trial.
The researchers focused on patients with EDSS scores between 4.0 and 6.5—a range where walking ability heavily dominates the score. They then stripped the data down to 15 granular subscores drawn from three functional systems: pyramidal, cerebellar, and sensory.
Using machine learning algorithms, the team identified four distinct disability patterns within the same EDSS scores, patterns that traditional scoring systems treat as clinically equivalent.
In other words, AI revealed that the EDSS, long treated as a clinical truth, is closer to a blunt instrument.
Four Patients, One Score, Four Realities
The AI-generated clusters exposed four recurring patterns:
• Pattern A: Patients with widespread impairment affecting more than half of daily living activities—muscle weakness, tremor, truncal ataxia, gait instability.
• Pattern B: Dominated by spasticity and impaired tandem walking.
• Pattern C: Marked primarily by sensory deficits or abnormal balance testing.
• Pattern D: Minimal impact across these domains, despite identical EDSS scores.
These patterns matter because they reflect how patients actually function, not just how far they can walk.
Higher EDSS scores were more likely to fall into Pattern A, suggesting that these AI-derived clusters may also carry prognostic value, hinting at future disease trajectories that a single number cannot predict.
Why This Changes Clinical Trials
The implications extend far beyond academic classification.
Clinical trials in multiple sclerosis often fail not because treatments are ineffective, but because outcome measures are too crude to detect meaningful change. When ambulation dominates disability scoring, improvements in upper-limb function, balance, or sensory symptoms may go unnoticed.
By introducing granularity into disability measurement, AI-based clustering offers a way to:
• Increase trial sensitivity
• Reduce noise in outcome data
• Identify responders more accurately
• Align endpoints with patient-relevant outcomes
As the study authors note, this approach could “increase the power of clinical trials to detect treatment effects on deficits that are relevant to patients and their activities of daily living.”
That sentence alone should command the attention of every pharmaceutical developer and regulator.
Precision Medicine, Finally Applied to MS
For years, precision medicine has promised individualized care. In multiple sclerosis, progress has been uneven.
AI-driven disability patterning brings that promise closer to reality.
Rather than treating all patients with the same EDSS score as interchangeable, clinicians could begin to:
• Match therapies to dominant impairment patterns
• Stratify patients more intelligently
• Identify who is likely to benefit from symptomatic versus disease-modifying interventions
• Design rehabilitation strategies that reflect real functional needs
This is personalization not at the molecular level, but at the lived-experience level, where medicine ultimately matters most.
A Broader Shift in Neurology
This study does not stand alone. It reflects a broader transformation underway in neurology.
Machine learning is increasingly used to:
• Interpret complex imaging data
• Identify subtypes within heterogeneous diseases
• Predict progression and treatment response
• Integrate clinical, imaging, and biological data into unified models
In early 2025, discussions among MS experts at the ACTRIMS Forum highlighted this very trend: AI is moving from experimental novelty to clinical utility. The question is no longer whether AI belongs in neurology, but how responsibly and rigorously it will be integrated.
The Caution Beneath the Promise
Yet enthusiasm must be tempered with realism.
AI does not replace clinical judgment. It reframes it.
Clustering algorithms reflect the data they are trained on. Their outputs demand interpretation, validation, and ethical oversight. Over-reliance without understanding could introduce new biases even as old ones are corrected.
The Basel study itself acknowledges that further research is needed to explore prognostic value and clinical implementation. That humility is essential.
The future of AI in medicine will not be built on automation alone, but on collaboration between clinicians, data scientists, and patients.
The End of “One-Score Medicine”
What this study ultimately exposes is a deeper truth: medicine has outgrown single-number thinking.
Complex diseases require multidimensional understanding. Artificial intelligence, when applied thoughtfully, offers a way to see what averages erase and categories obscure.
In multiple sclerosis, AI is not merely refining measurement. It is restoring visibility to patient experience.
And that may be its most important contribution of all.
