Objective: To develop a model that predicts alpha synuclein (a-syn) status of patients at risk of Parkinson’s disease (PD) using non-motor and cognitive tests that can be utilised in primary health care settings with limited time and resources.

Background: Alpha-synuclein seeding amplification assay (α-synuclein SAA) is emerging as a key biomarker for the biological definition of PD [1]. A-synuclein SAA has been found to be positive in 86% of people with prodromal PD, particularly those with hyposmia or isolated REM sleep behaviour disorder (iRBD) [2-4]. However, confirming these prodromal features requires specialised tests, such as polysomnography for iRBD and smell testing for hyposmia which are not widely available. Developing a prediction model using simple non-motor and cognitive tests could facilitate early screening in primary care, identifying at-risk individuals who warrant closer follow-up.

Method: We used data from Parkinson’s Progression Maker Initiative (PPMI) cohort. We trained machine learning models (XGBoost and TabPFN) to predict a-syn status using data from UPDRS- part I activities of daily living, Scales for Outcomes in Parkinson’s disease (SCOPA-autonomic symptoms), Montreal Cognitive Assessment MoCA, blood pressure, Benton Judgement of Line Orientation and Hopkins Verbal Learning Test [5,6]. We employed SHhapley Additive exPlanations (SHAP) to investigate the features with the highest impact on the model’s prediction [7].

Results: 1057 prodromal patients were identified with a known a-syn status. 580 SAA+’ve (63.1% male, mean age 63.37 ± 7.15 years), and 477 SAA-’ve (37.75% male, mean age 63.37 ± 7.15 years). The best model reached an accuracy of 68%, and Area Under the Curve of 0.70 (59% sensitivity,76% specificity, 95% CI 0.67 to 0.73), SHAP analysis demonstrated that orthostatic hypotension, reduced verbal fluency (from Hopkins and MoCA), and higher gastrointestinal scores in SCOPA-aut were correlated with SAA positivity (Fig. 1).

Conclusion: A machine learning model using non-motor features beyond hyposmia and iRBD moderately predicted α-synuclein status in people at risk. Orthostatic hypotension, reduced verbal fluency, and gastrointestinal dysfunction might be promising clinical markers but further refinement is needed to enhance clinical utility in primary care screening.

Data on the right, associated with SAA+, left:SAA–

References: [1] Zheng Y, Li S, Yang C, Yu Z, Jiang Y, Feng T. Comparison of biospecimens for α‐synuclein seed amplification assays in Parkinson’s disease: a systematic review and network meta‐analysis. European Journal of Neurology. 2023 Dec;30(12):3949-67.

[2] Mollenhauer B, Caspell‐Garcia CJ, Coffey CS, Taylor P, Singleton A, Shaw LM, Trojanowski JQ, Frasier M, Simuni T, Iranzo A, Oertel W. Longitudinal analyses of cerebrospinal fluid α‐Synuclein in prodromal and early Parkinson’s disease. Movement disorders. 2019 Sep;34(9):1354-64.

[3] Marek K, Russell DS, Concha-Marambio L, Choi SH, Jennings D, Brumm MC, Coffey CS, Brown E, Seibyl J, Stern M, Soto C. Evidence for alpha-synuclein aggregation in older individuals with hyposmia: a cross-sectional study. EBioMedicine. 2025 Feb 1;112.

[4] Antelmi E, Donadio V, Incensi A, Plazzi G, Liguori R. Skin nerve phosphorylated α-synuclein deposits in idiopathic REM sleep behavior disorder. Neurology. 2017 May 30;88(22):2128-31.

[5] Chen T, Guestrin C. XGBoost: A scalable tree boosting system. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining; 2016 Aug 13-17; San Francisco, CA, USA. New York: ACM; 2016. p. 785-94.

[6] Hollmann N, Müller S, Purucker L, Krishnakumar A, Körfer M, Hoo SB, Schirrmeister RT, Hutter F. Accurate predictions on small data with a tabular foundation model. Nature. 2025 Jan 9;637(8045):319-26.

[7] Lundberg SM, Lee SI. A unified approach to interpreting model predictions. Adv Neural Inf Process Syst. 2017;30:4765-74.

« Back to 2025 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/early-indicators-of-alpha-synuclein-aggregation-can-autonomic-and-cognitive-markers-predict-prodromal-parkinsons-disease/