Predicting Future Cognitive Impairment in De Novo Parkinson's Disease Using Clinical Data and Structural MRI

BACKGROUNDParkinsons disease is the second most common neurodegenerative condition and associated with increasing cognitive dysfunction as the disease progresses. However, subtle cognitive deficits can be detected at diagnosis in 42% of individuals, suggesting that damage may already be present. Our aim was to determine clinical and structural differences in those recently diagnosed with PD who later develop cognitive impairment, and whether these changes predict future cognitive decline. METHODSClinical and imaging data was acquired from the Parkinsons Progression Markers Initiative for 318 individuals with a diagnosis of Parkinsons disease and baseline 3T T1-weighted MRI. The cohort was divided according to cognitive status over follow-up, with 9 individuals developing Parkinsons disease dementia, 102 developing mild cognitive impairment and 207 remaining cognitively unaffected. FINDINGSAt baseline, those who went on to develop cognitive impairment (mild cognitive impairment or dementia) were older with more severe motor and non-motor symptoms (anosmia, rapid eye movement sleep behaviour disorder, depression). Grey matter loss was present in those destined for Parkinsons disease dementia in the precuneus, hippocampi, primary olfactory cortex, lingual gyrus, temporal cortex and cerebellum. Those who later developed mild cognitive impairment had an attenuated but similar pattern of grey matter loss in the temporal lobe, lingual gyrus and cerebellum. Using support vector machines with a feature selection step, future cognitive impairment could be predicted using 11 clinical variables (AUC = 0.81), structural imaging (AUC = 0.72) or a combination of these two modalities (AUC = 0.85). These models more accurately predicted those who developed dementia (subgroup sensitivity 100%). INTERPRETATIONSignificant abnormalities in cortical structure is present at least three years before dementia manifests in Parkinsons disease, with associated differences in clinical profiles. Combining this data provides a technique to accurately identify future cognitive impairment, providing a non-invasive way to stratify individuals early on.