Cognitive impairment is an important symptom of Parkinson’s disease (PD) and predicting future cognitive decline is crucial for clinical practice. Here, we aim to identify latent sub-groups of longitudinal trajectories of cognitive change in PD patients, and explore predictors of differences in cognitive change. Longitudinal cognitive performance data from 349 newly diagnosed PD patients and 145 healthy controls from the Parkinson Progression Marker Initiative were modeled using a multivariate latent class linear mixed model. Resultant latent classes were compared on a number of baseline demographics, and clinical variables, as well as cerebrospinal fluid (CSF) biomarkers and striatal dopamine transporter (DAT) density markers of neuropathology. Trajectories of cognitive change in PD were best described by two latent classes. A large subgroup (90%), which showed a subtle impairment in cognitive performance compared to controls but remained stable over the course of the study, and a small subgroup (10%) which rapidly declined in all cognitive performance measures. Rapid decliners did not differ significantly from the larger group in terms of disease duration, severity or motor symptoms at baseline. However, rapid decliners had lower CSF amyloidß42 levels, a higher prevalence of sleep disorder and pronounced loss of caudate DAT density at baseline. These data suggest the existence of a distinct minority sub-type of PD in which rapid cognitive change in PD can occur uncoupled from motor symptoms or disease severity, likely reflecting early pathological change that extends from motor areas of the striatum into associative compartments and cortex.