By Diagnostics World Staff
June 18, 2026 | Researchers from Japan have developed a machine learning framework called DiSPAH (Disease-progression Speed and Pathway Analysis based on a Hidden Markov model) to estimate both the pathway and speed of disease progression in individual patients. Their work is in pre-print at npj Digital Medicine (DOI: 10.1038/s41746-026-02665-8)
Progression for chronic diseases can be frustratingly heterogeneous between patients observed Yuichiro Yada and Honda Naoki, coauthors and researchers at Nagoya University’s Graduate School of Medicine.
This variability is not only clinically frustrating but also impedes effective treatment planning, clinical trial design, and patient counseling, they write in the paper. Using ALS (Amyotrophic Lateral Sclerosis) as a use case, the authors sought to model disease progression in two orthogonal dimensions: the progression pathway (the order in which functions deteriorate) and progression speed (how quickly a patient moves through the stages of disease).
Modeling Disease Progression
The result was the DiSPAH framework. DiSPAH uses a continuous-time hidden Markov model in which clinical features observed at irregular hospital visits are used to probabilistically infer underlying disease states. Transitions between these latent disease states capture the diversity of progression pathways. To model differences in progression speed, the researchers introduced a patient-specific speed parameter that time-scales transition rates.
The researchers built their tool from two datasets of patients with limb-onset ALS, a form of the disease where symptoms begin in the arms or legs rather than in the muscles controlling speech and swallowing (bulbar-onset ALS). From the AnswerALS dataset, 264 patients were used to train the model. From the PRO-ACT cohort, 2,565 patients were used to validate the results.
The system identified six distinct patterns of disease progression among the patients. Some showed slow decline in motor function, with little effect on speech or breathing, while others experienced rapid deterioration.
“Subtle differences between patients also emerged. For example, in some patients gross motor functions such as walking declined before fine motor skills such as writing or buttoning a shirt, while in others the opposite was true,” said Yada in a press release about the work. “These six patterns were identified in one patient dataset and largely reproduced in a second, larger dataset, suggesting that they capture common progression patterns in limb-onset ALS.”
Importantly, speed and decline patterns were found to be independent of each other. A patient could follow a severe pattern at a slow speed, or a milder one at a fast speed. Previous tools could not measure both dimensions at once.
One of the most important findings was that DiSPAH could, to some extent, predict a patient’s progression speed and broad progression pattern from information available at the first medical visit based only on basic functional assessments and the presence of certain gene mutations. These early predictions have important potential for patient care. Doctors could use them to plan treatment, prepare patients and families, and design better clinical trials by grouping participants according to how their disease advances.
The researchers also identified a genetic mutation of note. “Patients with the C9orf72 mutation exhibit faster disease progression compared to those without it. The finding is consistent with the previous report linking C9orf72 mutation to shorter survival duration in ALS,” the authors wrote. When they analyzed data from motor neurons grown in the laboratory from patients' own stem cells, the results showed that faster ALS progression may be linked to problems in how cells produce and manage proteins, as well as signs of cellular stress.
This points to a possible biological explanation for why some patients decline faster than others and gives scientists a new target for future research into ALS treatments.
“Our results highlight that jointly modeling progression pathway and speed improves prediction of heterogeneous disease courses, offering a powerful tool for personalized care and research in ALS and other chronic conditions,” the authors wrote.
They aim to extend the tool to all ALS patient types, improve its reliability, and ultimately apply it to other chronic diseases such as Alzheimer's and Parkinson's disease.