A preliminary analysis posted to the medRxiv preprint server on 6 March 2026, and explicitly not certified by peer review, reports that how people with Parkinson's disease performed on baseline vision tests predicted later cognitive decline more clearly than the genotypes most often raised in prognostic conversations. The prospective study, run by Matthew Solomons, Naomi Hannaway, Oliver Fox and colleagues at UCL Queen Square Institute of Neurology with the Vision in Parkinson's Study team, recruited people with Parkinson's and no dementia at baseline between 2017 and 2020 at 35 UK Parkinson's clinics. Of 450 participants (mean age 71.7 years, 68 per cent male, mean baseline Montreal Cognitive Assessment score 27.7), 263 were classed as poor-vision on a web-based battery of visual tests. Over a mean follow-up of 32.7 months (SD 12.3), the poor-vision group progressed to Parkinson's mild cognitive impairment at more than twice the rate of the rest (hazard ratio 2.34, 95 per cent confidence interval 1.58 to 3.48, FDR-corrected p 0.00062, adjusted for age and sex).
The genetic results were weaker and, in places, absent. Of the 450, 110 carried at least one APOE4 allele; carrying it was not associated with progression to mild cognitive impairment (hazard ratio 1.178, CI 0.75 to 1.85, corrected p 0.72) or to dementia (hazard ratio 1.17, CI 0.39 to 3.49, corrected p 0.97), and the picture did not change when participants were split by vision status. Twenty-four participants carried one of two GBA1 risk variants (10 E365K, 14 T408M); across the whole cohort this was not significantly associated with progression to mild cognitive impairment (hazard ratio 1.68, CI 0.84 to 3.36), and there was a single dementia conversion among carriers, too few for the authors to model. Only within the good-vision group did GBA1 carriage carry a clear signal (hazard ratio 4.61, CI 1.73 to 12.28). Polygenic risk scores, calculated in 325 participants, missed significance for Parkinson's and Alzheimer's overall after correction (1.64, corrected p 0.058; 1.65, corrected p 0.057), and only reached it when combined with vision status: high Parkinson's score in good-vision participants (2.66, CI 1.21 to 5.81), high Alzheimer's score in poor-vision participants (1.91, CI 1.10 to 3.32), and both scores high in good-vision participants (6.14, CI 1.36 to 27.83). Two other loci, LRP1B and RIMS1, showed no significant effect. Rare SNCA mutations are cited by the authors as a known contributor to dementia in Parkinson's, but were not tested in this analysis.
Several boundaries sit around those numbers. Progression was defined by sustained MoCA scores, below 25 for mild cognitive impairment and 21 or lower for dementia, not by a clinician's diagnosis; most events were mild cognitive impairment rather than dementia, and the dementia analyses rest on small numbers with wide confidence intervals. The poor-vision group was on average about five years older than the good-vision group, although the models were corrected for age and sex. The authors' claim that stratifying by vision and genetics could cut a trial from 705 to 160 participants comes from simulation, not from a trial that has been run.
The null APOE4 finding sits against larger and longer work. A pooled analysis of 1,002 newly diagnosed patients from six European population-based cohorts, published in Movement Disorders in February 2022 and followed for up to 10 years, reported that carriers of a single APOE e4 allele had roughly three times the risk of Parkinson's disease dementia, with no significant effect for the SNCA rs356219 or MAPT variants tested. A separate pooled cohort analysis published in npj Parkinson's Disease in February 2026 found that APOE genotype and cerebrospinal fluid markers of Alzheimer's pathology track cognitive decline differently in sporadic, GBA1 and LRRK2 Parkinson's. On that evidence, the new preprint's flat APOE result is more plausibly a function of a clinic-based sample followed for under three years than a closing of the question.