A recent study published in Neurology suggests that individuals at high risk of developing Parkinson’s disease could be identified years before symptoms arise. By using machine learning to analyze proteins found in blood samples and combining this data with simple clinical information, researchers developed a model capable of predicting Parkinson’s disease risk up to 15 years in advance. This early detection could help prevent or delay the progression of this neurodegenerative disorder, offering new hope for disease management and treatment.
Parkinson’s disease is the second most common neurodegenerative disorder after Alzheimer’s. It primarily affects movement, causing tremors, stiffness, and balance problems. However, by the time these symptoms appear, significant and irreversible brain damage has often occurred. Parkinson’s has a long “prodromal phase” that can last for decades before typical motor symptoms become evident. During this period, non-motor symptoms such as sleep disorders, depression, and loss of smell may emerge, but they are often not recognized as early warning signs of Parkinson’s.
The challenge in treating Parkinson’s lies in its late diagnosis, after extensive brain damage has already occurred. Current treatments focus on managing symptoms rather than halting disease progression. Scientists believe that identifying the disease at earlier stages—before noticeable motor symptoms—could allow for interventions that prevent or delay the onset of more severe symptoms.
The new study, led by Jian-Feng Feng, Dean of the Institute of Science and Technology for Brain-Inspired Intelligence at Fudan University, along with Wei Cheng, Principal Investigator at the same institute, was driven by the need for an accessible and non-invasive way to identify those at high risk for developing Parkinson’s years in advance.
The study team also included Lin-Bo Wang, a Young Associate Principal Investigator at Fudan University, and Jia You, a former postdoctoral researcher who has since been promoted to Young Principal Investigator. Their collective efforts focused on combining machine learning with blood biomarkers to create a predictive model that could detect Parkinson’s disease risk long before clinical symptoms appear.
“Parkinson’s disease is characterized by the irreversible loss of dopaminergic neurons, which is caused by α-synuclein aggregates,” the researchers told PsyPost. “In 2022, upon discovering that the brain volume of newly diagnosed patients correlates with future clinical progression, we developed an interest in predicting Parkinson’s disease years prior to clinical diagnosis. Early detection holds utmost significance for the development of treatments aimed at decelerating brain atrophy and postponing disease progression in the very early stage of the disease.”
To develop a model capable of predicting Parkinson’s risk, the researchers analyzed data from over 50,000 participants in the UK Biobank, a large health resource in the United Kingdom that collects data on genetic, clinical, and lifestyle factors. The study focused on the levels of 1,463 different proteins found in the blood and how these proteins might be linked to future Parkinson’s diagnoses.
The study included 52,503 participants who did not have Parkinson’s at the start of the study. These participants had their blood plasma analyzed to measure protein levels. The research team used machine learning, a form of artificial intelligence that identifies patterns in data, to analyze the blood protein levels alongside clinical and demographic information. This information included factors like age, education, history of head injuries, and other health markers.
During a median follow-up period of 14 years, 751 participants developed Parkinson’s disease. The researchers used this data to train a machine learning model that could predict who was at risk of developing Parkinson’s based on their protein levels and clinical characteristics. They then validated the model using a separate dataset from the Parkinson’s Progression Markers Initiative, a project that includes people diagnosed with Parkinson’s, those at high risk of developing the disease, and healthy individuals.
The model developed by the researchers achieved a high level of accuracy, correctly identifying individuals at risk for Parkinson’s disease in both the UK Biobank and the validation dataset. The analysis revealed that 22 specific proteins found in blood plasma were significantly associated with Parkinson’s risk. Some of the most important proteins identified included neurofilament light (NfL), a protein linked to brain cell damage, and several proteins involved in inflammation and muscle function.
By integrating clinical information such as age, history of traumatic brain injury, and blood creatinine levels (a marker of muscle mass and kidney function), the researchers were able to improve the accuracy of the model. The final version of the model, which included both protein and clinical data, was able to predict Parkinson’s risk with a high degree of accuracy, even up to 15 years before diagnosis.
“A combination of plasma proteins and clinical-demographic measures is capable of identifying individuals at high risk of developing Parkinson’s disease up to 15 years prior to the clinical diagnosis,” Feng and his colleagues said. “Our model can be integrated into routine health examinations to detect high-risk individuals of developing Parkinson’s disease, thereby presenting opportunities to explore and assess neuroprotective treatments.”
The study also revealed that certain proteins showed distinct changes over time in individuals who eventually developed Parkinson’s. For example, levels of the protein NfL began to rise about 12 years before diagnosis, while other proteins linked to inflammation and muscle function showed changes several years before diagnosis. These findings suggest that monitoring protein levels over time could provide valuable insights into an individual’s risk of developing Parkinson’s.
“We were surprised to discover that changes in several proteins could be observed more than a decade before clinical diagnosis,” the researchers explained. “For example, NfL, a marker of neuronal damage and the most significant predictive protein, showed increased levels 12 years before diagnosis, indicating early neuroaxonal damage.”
While the findings of this study are promising, there are several limitations that need to be addressed in future research. One limitation is the lack of diversity in the study population. Most participants in the UK Biobank and the Parkinson’s Progression Markers Initiative were of European descent, which may limit the generalizability of the findings to other populations. Future studies will need to validate the model in more diverse populations to ensure it is effective for everyone.
Another limitation is that the diagnosis of Parkinson’s in the UK Biobank was based on medical records, which may not always be accurate. Some participants could have been misdiagnosed, particularly in cases where specialists were not involved in the diagnosis. More accurate diagnostic methods, such as brain imaging, could help improve the reliability of future studies.
In addition, while the study identified several proteins that are associated with Parkinson’s risk, many of these proteins are also linked to other neurodegenerative diseases. For example, elevated levels of NfL have been found in Alzheimer’s disease and other conditions that involve brain cell damage. Therefore, these proteins may not be specific enough to Parkinson’s, and additional biomarkers may be needed to distinguish Parkinson’s from other diseases.
The study also used a semi-quantitative method to measure protein levels, which may limit the accuracy of the findings. Future studies that use more precise measurement techniques could help refine the model and improve its predictive power.
Finally, the model was trained using data collected at a single point in time, which may not capture biological fluctuations in protein levels. Repeated measurements of protein levels over time could provide more accurate predictions and help identify the most reliable biomarkers for early detection of Parkinson’s.
Despite these limitations, this study represents a significant step forward in the early detection of Parkinson’s disease. The model developed by the researchers offers a non-invasive, cost-effective way to identify individuals at high risk of developing Parkinson’s, potentially allowing for earlier interventions that could slow or prevent the progression of the disease. While further research is needed to validate the findings and refine the model, the results suggest that blood-based biomarkers, combined with clinical information, could be a valuable tool for predicting Parkinson’s risk in the general population.
“Our long-term objective is to develop a series of predictive models applicable in community sets,” the researchers said. “These models will utilize non-invasive, cost-efficient, and readily accessible features to detect PD and other neurodegenerative disorders years prior to clinical diagnosis, with the aim of slowing down or preventing their progression.”
“In addition, the testing fee for plasma proteins is currently high in high-throughput proteomics. We are working on collaborating with companies to conduct blood tests specifically targeting these biomarkers, which would significantly reduce application costs.”
The study, “Prediction of Future Parkinson Disease Using Plasma Proteins Combined With Clinical-Demographic Measures,” was authored by Jia You, Linbo Wang, Yujia Wang, Jujiao Kang, Jintai Yu, Wei Cheng, and Jianfeng Feng.
