Abstract
Cognitive impairment is a common non-motor sym.ptom of neurodegenerative Parkinson's disease (PD). However, the onset and severity of cognitive impairment is heterogenous and currently cannot be predicted. This makes it challenging for informing patients in advance, and for the application of early therapeutic interventions. Moreover, the biological pathways that underlie cognitive impairment in PD remain to be fully determined. To better predict and understand the biological pathways involved in cognitive impairment, we have integrated multimodal datasets including clinical, demographic, biomarker and high throughout proteomic data from the longitudinal Parkinson's progression markers initiative (PPMI). The integrated dataset was used to develop three machine learning models that could predict the development of cognitive impairment nine years in the future, compared to patients that remained cognitively normal over the same duration. The best model could predict future cognitive impairment with 88% area under the curve. Permutation feature importance revealed that cytokine, interleukin-2 (IL-2) and insulin like growth factor (IGF) signalling were significant biological pathways contributing to the prediction of future cognitive impairment.