IntroductionIdentifying Parkinson's disease (PD) patients at risk of cognitive decline is crucial for enhancing clinical interventions. While several models predicting cognitive decline in PD exist, a new machine learning framework called disease progression models (DPMs) offers a data-driven approach to understand disease evolution. MethodsWe enrolled 423 PD patients and 196 healthy controls from the Parkinson's Progression Markers Initiative (PPMI). Our study encompassed a range of biomarkers, including motor, neurocognitive, and neuroimaging evaluations at baseline and annually. A methodology was employed to select optimal combinations of biomarkers for constructing DPMs with superior predictive capabilities for both diagnosing and estimating conversion times toward cognitive decline. ResultsAt baseline, the approach showed excellent performance in identifying individuals at high risk of cognitive decline within the first five years. Furthermore, the proposed timeline from cognitive impairment to dementia was also used to explore clinical events such as the onset of cognitive impairment, the development of dementia or amyloid pathology. The presence of amyloid pathology did not alter the progression of cognitive impairment among PD patients. ConclusionsNeuropsychological measures and certain biomarkers, including cerebrospinal fluid (CSF) amyloid beta 42 (Aβ42) and dopamine transporter deficits, can be used to predict cognitive decline and estimate a timeline from cognitive impairment to dementia, with amyloid pathology preceding the onset of dementia in many cases. Our DPMs suggested that the profiles of CSF Aβ42 and phosphorylated tau in PD patients may differ from those in aging patients and those with Alzheimer's disease.