Ontology-based, Real-time, Machine Learning Informatics System for Parkinson’s Disease


Currently, there are no definite laboratory or imaging tests available to diagnose Parkinson’s disease (PD), and it is clinically diagnosed by primary care physicians, general neurologists, and movement disorders neurologists based on presence of key movement symptoms and signs, such as slowness of movement (bradykinesia) with rest tremor and/or increase in muscle tone (rigidity), with or without non-motor symptoms. However, initial clinical diagnosis has been shown to be frequently inaccurate when compared to the gold standard neuropathological diagnosis in research studies, as the presenting clinical features can also be present in other conditions related to PD, such as atypical Parkinsonian disorders / syndromes (namely, Lewy body disease, multiple system Atrophy, progressive supranuclear palsy, corticobasal degeneration) or secondary parkinsonism (for example, drug-induced parkinsonism, vascular parkinsonism, etc.). The International Parkinson and Movement Disorders Society (MDS), a leading organization of PD specialists, has created criteria for diagnosing patients into two levels of diagnostic certainty, that is, “clinically established” and “clinically probable” PD, and these criteria have been shown to have high sensitivity and specificity. However, these criteria follow a complex algorithm and thus nearly impossible to implement for primary care clinicians and general neurologists, and relatively impractical in their current format even for movement disorders specialists.

Beyond the diagnosis, the most frequent question asked in the clinic by the PD patients and their caregivers is that of prognosis, that is, how will the patient’s disease progress in the future? To this end, the heterogeneity of PD prognosis is quite remarkable, sometimes referred as “many faces of Parkinson’s disease”, which has been formally recognized to represent distinct subtypes of PD, specifically, “mild-motor-predominant”, “intermediate”, and “diffuse malignant”. Such proposed subtypes have been linked with varying rates of progression, and linked with underlying complexity of molecular changes occurring in the nervous system in PD. For example, although alpha-synuclein (α-Synuclein) protein misfolding and accumulation in the form of Lewy bodies is considered a pathological hallmark of PD, a variety of genes and / or proteins are also implicated in PD, such as Leucine-rich repeat kinase 2 (LRRK2) and glucocerebrosidase (GBA), in addition to various genetic changes in the α-Synuclein gene.

With respect to PD in veterans, several studies have shown that exposure to neurotoxins, such as Agent Orange, are associated with increased risk of PD. Importantly, PD is s a presumptive condition for veterans who served in a certain place (Vietnam, Korean Demilitarized Zone, and Camp Lejeune) during a certain time period, under the assumptions that they might had been exposed to toxins (such as Agent Orange). However, there have been limited or no research on whether there are differences in PD associated with Agent Orange exposure in veterans versus PD not associated with Agent Orange exposure in non-veterans, especially with respect to the accuracy of diagnosis.

Above summarized issues are two of the most pressing challenges faced by clinicians, researchers, patients and their caregivers in the field of PD. Furthermore, there are no tools available to the clinicians or researchers at the point-of-care in clinic or in research studies, respectively, for classifying an individual patient on the MDS PD clinical criteria or forecast prognosis on relevant outcome measures.  As part of the ORMIS-PD research project, we are taking a novel approach to developing potential solutions for addressing these unmet needs. In aim 1, ORMIS-PD is being designed to enable capturing of the patient’s clinical information in a curated and intuitive fashion using the Parkinson & Movement Disorders Ontology (PMDO) and a touch-screen interface, respectively. Then, ORMIS-PD will automatically reconcile the collected information with the Movement Disorders Society criteria for computer-aided diagnosis of the patient. Subsequently, the ORMIS-PD platform will automatically calculate the prognosis score from the collected information, and the apply artificial intelligence method, driven by data of a large PD database Parkinson Progression Marker Initiative, for forecasting future changes in prognosis score. In aim 2, clinical information of PD patients will be collected using ORMIS-PD for two groups, one group with Agent Orange exposure and another group without Agent Orange exposure, at two sites, namely, University of Vermont Medical Center and Oregon Health & Science University/Portland Veterans Affairs Medical Center. Afterwards, the two groups will be compared with respect to their ORMIS-PD generated diagnosis and prognosis score.

The ORMIS-PD project is a research collaboration among University of Vermont (UVM), Case Western Reserve University (CWRU), and Portland Oregon Health & Sciences University (OHSU) / Portland VA Medical Center (PVAMC), and has been funded from a competitive $400,000, two-year grant starting September 2021 from the Department of Defense (DOD). Principal investigator Deepak K. Gupta, MD, Movement Disorders Neurologist at the Binter Center, and Assistant Professor of Neurological Sciences at UVM) is leading this collaborative, and multi-site project. In addition to Dr. Gupta, team investigators include James Boyd, MD, Director of the Binter Center, and Professor of Neurological Sciences at UVM; Satya S. Sahoo, PhD, Associate Professor of Computer Science and Director of Biomedical Health Informatics PhD program at CWRU; and Amie Hiller, MD, Associate Professor of Neurology at OHSU/PVAMC. Other members of the team include Katherine Chan, Research Coordinator in the department of neurology at UVMMC, Ms. Katrina Prantzalos, Postdoctoral Graduate Student at CWRU, Ms. Brenna Lobb, Research Coordinator at OHSU/PVAMC.

We are hopeful that successful completion of this research project will lay the groundwork for application of artificial intelligence approaches for improving PD diagnosis and predicting progression, and provide new information on potential differences in diagnostic classification and prognosis measures between neurotoxin-associated PD and idiopathic PD. There are several future directions of ORMIS-PD, such as, expansion to include context of atypical parkinsonian disorders, integration with patient-centered digital measures obtained using mobile health technologies, and predictive analytics for treatment responses and clinical trials for PD and related disorders.

Pictured above (from left to right and top to bottom): Ms. Lobb, Dr. Gupta, Dr. Sahoo, Dr. Hiller, Ms. Prantzalos, Ms. Chan, and Dr. Boyd from the ORMIS-PD project virtual kick-off meeting on September 3rd, 2021

Leave a Comment