One diagnosis, many diseases

If subtyping is going to matter for trials, one framework has to actually beat the others at flagging rapid progressors. So all three were put head-to-head over ten years of PPMI follow-up: the classic clinical motor split (tremor-dominant vs. postural-instability/gait-difficulty vs. indeterminate), the pathological brain-first vs. body-first model, and the data-driven DM/IM/MMP model — scored against 25 predefined milestones spanning six functional domains.

The data-driven model won. Diffuse-malignant patients reached the most milestones and carried the highest hazard of progression after adjustment for age. The clinical payoff is concrete: trial simulations showed that enrolling DM patients could cut required sample sizes by roughly half at standard trial durations versus unstratified cohorts — a direct lever on the feasibility of disease-modification trials.

A subtype worth stratifying on should have biology behind it, not just statistics. Across 1,390 PPMI patients genotyped for seven PD-associated genes, the question was which subtyping frameworks map onto genetic architecture. The clinical tremor/gait split showed no genetic signal that survived multiple-comparison correction. The biologically grounded frameworks did: LRRK2 variants were strongly enriched among α-synuclein seed-amplification–negative patients (Cramér's V = 0.25), and GBA1 variants tracked the body-first and diffuse-malignant subtypes.

The message: the data-driven and pathological subtypes carve PD along lines that genetics recognizes — the clinical motor labels largely do not.

The original data-driven model had a practical flaw: it used fixed baseline thresholds. That works if you have a drug-naive patient at diagnosis — but it cannot classify the patient sitting in clinic three years in, whose scores have already drifted with the disease. To subtype prospectively, the yardstick has to move with disease duration.

So the thresholds were rebuilt year by year. Across 1,030 de novo idiopathic PD patients from the April-2026 PPMI freeze, within-disease-year percentiles were computed for each motor and non-motor scale, so a patient is compared against peers at the same disease duration. The result is a stable, prospective classifier: subtype prevalence held steady across years, and the survival separation was decisive — DM patients progressed three times faster than MMP.

Mild cognitive impairment in PD (PD–MCI) is prognostically messy — some patients convert to dementia quickly, others never do — and the clinical, pathological, and data-driven frameworks don't capture which is which. This line proposes a different axis entirely: imaging-defined subtypes based on cholinergic degeneration — beginning with the nucleus basalis of Meynert (cholinergic nucleus 4, Ch4), the brain's principal source of cortical acetylcholine.

In baseline MRI from 162 PD–MCI participants in PPMI, Ch4 grey-matter density was measured with voxel-based morphometry and probabilistic maps. Patients with low Ch4 density had worse motor and autonomic burden, poorer olfaction, and — critically — progressed to cognitive-decline milestones significantly faster. The proposal: PD–MCI with Ch4 degeneration is a distinct, more malignant subtype — a candidate for enriching trials of cholinergic therapies.