Cognitive impairment and Parkinson?s disease dementia (PDD) are well-established disorders in Parkinson?s disease (PD) which are debilitating and contribute to increased mortality. The course, severity, and rate of progression of cognitive symptoms in PD is variable and unpredictable. Some patients develop PDD within the first several years of diagnosis, while others remain cognitively intact or have a milder form of impairment for many years after diagnosis. This heterogeneity in impairment profile and risk to PDD likely reflects the diverse underlying pathophysiological mechanisms associated with PD progression and cognitive dysfunction. The combination of diverse pathological features and clinical phenotypes makes it challenging to inform patients what to expect during the course of disease, and is a substantial barrier to developing new drugs for cognitive impairment in PD. Therefore, developing prognostic markers of risk of cognitive progression in PD is important. Our overall hypothesis is that different combinations of biomarkers will be more informative in predicting cognitive progression compared to a single biomarker alone. Thus, our objective is to investigate multivariable data to identify unique clinical-molecular-imaging biomarker signatures that identify individuals with PD who are most likely to experience substantial cognitive changes that ultimately lead to PDD. To achieve our objective, we will first develop deep learning models that identify multivariable features that are prognostic of incident mild cognitive impairment in people with early-stage PD (aim 1). Next, we will develop deep learning models that identify multivariable features that are prognostic of either (a) conversion from mild cognitive impairment to PDD (aim 2); or, (b) reversion from mild cognitive impairment to cognitively intact PD (aim 3). Finally, we will develop deep learning models that identify prognostic markers of rapid cognitive deterioration in PD leading to PDD (aim 4). For the development of the proposed models, this project will take advantage of a large repository of clinical data available at the Center for Health + Technology (CHeT) at the University of Rochester. The modeling will be carried out at the University of Rochester Data Science Consortium (RDSC) by an experienced team of data scientists that will use multimodal temporal convolution network models. A multidisciplinary team that includes neurologists, pharmacologists, and data scientists will support model execution and the interpretation of the modeling results. The expected outcomes of the efforts from this research are comprehensive multivariable prognostic markers of cognitive progression in PD across different stages of disease, from mild impairment to PDD that will add novel insight into disease process.
Deep learning will be used to construct models of cognitive progression in Parkinson?s disease (PD) from multiple data modalities. The models will assess and evaluate the potential impact of clinical, biosamples, genetics, and imaging variables on the progression to mild cognitive impairment and dementia, as well as on rates of cognitive progression. These models will identify markers and their combinations that are prognostic to distinct cognitive progression trajectories in PD.
