Scientists are using machine learning to analyse data collected from the world’s largest study of people with Huntington’s disease.
Huntington’s disease is a complex condition to treat. There is no cure and no treatments that can alter its course. Its progression is hard to predict, but a new retrospective study shows how machine learning could play an important role in predicting its trajectory.
The study was conducted by scientists at the University of Iowa Health Care, US biotech company Genentech and Roche Products, the company behind the development of a potential new therapy for HD, Tominersen.
“Advancement and application of data-driven machine learning models have the potential to be used in real-world clinical settings to shorten times to diagnosis and understand whether patients are responding positively to treatments relative to natural history,” a spokesperson from Roche and Genentech told NR Times.
“Scientists are trying to understand why and how the disease progresses in different ways in individuals. These results provide useful additions to the HD natural history literature and describe how applying new analysis methods could uncover a deeper understanding of how the disease evolves and progresses.”
The study applies k-means clustering to existing data which, according to the study’s authors, has not been applied to the natural progression of HD before. The researchers used data from Enroll-HD, the largest ongoing observational study on Huntington’s disease in the world. The patient registry is made up of information from 20,000 people and includes data about their symptoms, genetics and other clinical factors.
The machine learning model allowed the researchers to identify patterns in the data that could indicate how the disease progresses. They started by clustering patients into three categories: rapid, moderate and slow progressors. Slow progressors accounted for 29.2% of the almost 5000-strong cohort, while 45.5% were moderate progressors and 25.3% were categorised as rapid progressors.
A further algorithm delved deeper into individual patients’ data to identify a set of cross-sectional characteristics that predict the future trajectory of the condition for each patient. The researchers found that the most useful data for predicting HD progression were patients’ CAG-age product (CAP) score, a mathematical formula that is used to estimate a person’s age of onset. The number of years since the onset of symptoms also played a significant role in predicting progression.
“We found that by using a type of algorithm called clustering, we could group patients into different clusters based on how their disease progressed,” the study spokesperson said. “We then used another algorithm to predict how the disease would progress in individual patients based on their clinical characteristics.”
The machine learning method made it possible to identify patterns of progression in Huntington’s disease and predict how the disease will progress in individual patients. In a real-world setting, this could enable doctors to personalise treatment plans based on these predictions.
A Roche spokesperson said: “Machine learning approaches such as this could be considered for application in real-world clinical practice to support the treating physician in assessing whether patients are improving or worsening on future disease-modifying treatment (DMT) compared with patients of a similar clinical profile receiving standard of care.”
In reality, a clinical visit would include a series of core assessments such as those seen in the enrollment stage of this study. Clinicians would then use a model that is trained on Enroll-HD data to predict the trajectory a patient is on compared with others with a similar profile.
The researchers said that, over time, this prediction may support a change in clinical decision-making and, as a result, personalise each patient’s healthcare journey.
The spokesperson added: “These findings support the notion that providing clinicians with the ability to monitor an individual’s progression against key cognitive, behavioural and motor symptoms in real-time will support enhanced decision-making and identify those eligible for clinical trials of DMTs earlier.
“The article shows that using machine learning to analyse large and complex datasets, like those from the Enroll-HD study, can help scientists better understand how neurological diseases like Huntington’s progress and develop personalised treatment plans for patients.”
The study builds on previous research that has also aimed to capture the progression of multiple outcomes simultaneously. One clinically validated measure, known as cUHDRS, weights scores of cognition, motor and functional ability into a single component score.
The method has proven to be useful as a primary outcome measure in clinical trials. The research team aimed to build on these existing efforts by acknowledging the behavioural and psychiatric aspects of Huntington’s Disease, visualising how each domain evolves over time.
“Holistically capturing disease progression can be helpful for understanding the global pattern of symptoms over the lifespan of individuals with HD,” the researchers said in a joint response to NR Times.
“In our study, we selected a priority: a set of clinically meaningful motor, cognitive and behavioural outcomes that clinicians were likely to be familiar with in clinical practice.
“Despite the individual heterogeneity in the rate of decline between domains, our approach suggests that it may be possible to classify patients into rapid, moderate and slow progressors, depending on their overall function.”
Concluding the article, the authors of the paper said further work is needed to “develop prognostic models of HD progression as these could help clinicians with individualised clinical care planning and thus improve disease management.”
