Machine learning, a form of artificial intelligence (AI), has been able to define three distinct subtypes of Parkinson’s disease based on how quickly newly diagnosed patients’ symptoms progressed, potentially helping doctors more accurately detect the disease and predict its progression.
In the study by researchers at Cornell University, each subtype appears to have its own genetic signature, which could offer targets for more personalized treatment with new or existing drugs.
“We may need to consider personalized treatment strategies based on the patient’s disease subtype,” Fei Wang, PhD, founding director of the Institute of AI for Digital Health (AIDH) at Weill Cornell Medicine in New York, said in a university press article.
The study, “Identification of PACE subtypes of Parkinson’s disease and reorientation of treatments through integrative analyses of multimodal datawas published in npj Digital Medicine.
Parkinson’s disease progresses in different ways and at different rates in different patients.
The motor and non-motor symptoms of Parkinson’s disease are not experienced in the same way or at the same rate by everyone with Parkinson’s disease. This can make it difficult to diagnose the disease in its early stages or to decide which treatment is likely to be most effective.
“Parkinson’s disease is very heterogeneous, meaning that people with the same disease can have very different symptoms,” said Wang, a professor of medical informatics at Weill Cornell. “This indicates that there is probably no universal approach to treating it.”
Wang and his team used machine learning to recognize patterns in how symptoms manifest over five years or more, feeding computers data on disease progression from 406 adults newly diagnosed with Parkinson’s. The group’s average age at disease onset was 59.6 years, and the majority (65.5%) were men.
Machine learning is a branch of artificial intelligence that involves the use of algorithms that analyze and learn from existing data, allowing computers or machines to “learn” and make predictions.
“We used five-year longitudinal records of individuals across more than 140 items of diverse motor and non-motor assessments… (to build) a deep learning model,” the researchers wrote.
Three distinct subtypes were defined based on data from the Parkinson’s Progression Markers Initiative (PPMI) and validated with data from the National Institute of Neurological Disorders and Stroke (NINDS) Parkinson’s Disease Biomarkers Program (PDBP).
The first subtype, called Inching Pace (PD-I), involved 145 patients. All had mild motor and nonmotor symptoms at diagnosis, including milder sleep and cognitive disturbances, and had slow disease progression. Less than half (46.2%) were male.
The 207 people classified in the second subtype, moderate-pace (PD-M), started with mild symptoms but progressed at a moderate rate, with symptoms continuing to worsen from the second year of follow-up compared with the PD-I group. The majority of PD-M patients, 155, or 74.9%, were men.
Nearly 82% of patients in the fastest-progressing group were men
The PD-R (Rapid Pace) subtype, which included 54 patients, was characterized by rapid disease progression, with the fastest rate of symptom worsening. These patients were rated as having the most severe motor symptoms at the onset of Parkinson’s disease, and more significant nonmotor symptoms, particularly affecting cognition, the researchers noted. PD-R patients were on average older at disease onset than other patients—64.4 years—and 81.5% of this group were male.
To identify a genetic signature for each of the three subtypes, the team focused on about 90 known genetic variants linked to Parkinson’s disease, including APOEwhich has been associated with faster cognitive decline and a higher risk of death in patients.
For example, PD-R—the fastest-progressing subtype—could be identified by a set of 14 genes, most of which are related to inflammation and oxidative stress, two mechanisms known to affect the development and progression of Parkinson’s disease.
The researchers also looked for genes that were expressed differently, meaning those that became more active (upregulated) or less active (downregulated) for each subtype. Gene expression refers to the process by which the information encoded in a gene is transformed into a product, such as proteins.
In total, the PD-I subtype had 2,176 differentially expressed genes, the PD-M subtype 2,376, and the PD-R subtype 2,305 differentially expressed genes compared with healthy individuals. Different modules of approximately 210–240 genes were specified for each subtype.
Diabetes drug metformin could be used to treat Parkinson’s disease
Beyond defining the different subtypes, the researchers looked at drugs that could target the specific genetic signatures of each subtype. In addition to approved treatments for the disease like levodopa, they looked at drugs for other diseases that could be repurposed for these Parkinson’s subtypes.
Drawing on two real-world databases, the INSIGHT Clinical Research Network and the OneFlorida+ Clinical Research Consortium, they identified metformin, approved for type 2 diabetes, as particularly promising.
“We found that people taking the diabetes drug metformin appeared to have improved disease symptoms, particularly symptoms related to cognition and falls, compared with those not taking metformin,” said Chang Su, PhD, first author of the study.
This was particularly evident in people with the PD-R subtype, who are more likely to experience cognitive deficits in the early stages of the disease.
The safety of metformin in treating patients with Parkinson’s disease, however, still needs to be “rigorously evaluated in dedicated studies,” the scientists noted.
“This work provides a better understanding of the clinical and pathophysiological complexity of Parkinson’s disease progression and accelerates precision medicine,” the researchers conclude. Pathophysiology refers to the changes that cause or result from a disease.
Although further work is needed to validate this study, its results support “the existence of different pathophysiological mechanisms underlying different (Parkinson’s) subtypes,” leading to distinct “progression trajectories” of clinical importance, they added.