Parkinson’s disease (PD) presents a wide range of motor and non-motor symptoms, complicating its diagnosis and treatment. In a recent study led by Assistant Prof. Enrico Glaab from the Luxembourg Centre for Systems Biomedicine (LCSB) of University of Luxembourg as part of the European research project ERA PerMed DIGIPD, the researchers combined digital gait sensor with metabolomics and clinical data to assess their utility in classifying disease outcomes and comorbidities in people with Parkinson’s. This integrative approach using advanced machine learning methods could help reshape how we monitor and diagnose this complex disease.
Analysing motor symptoms with gait sensor analysis
The researchers analysed detailed gait measurements from 162 people with Parkinson’s and 129 healthy controls of the Luxembourg Parkinson’s Study, using shoe-attached sensors that captured motor function and gait variability. By testing both extracted gait parameters and raw time-series data, the team found that machine learning models using time-series data significantly improved diagnostic accuracy. Thereby, the team was able to distinguish people with Parkinson’s from controls with success rates between 83-92%, depending on the modeling approach. “Our findings demonstrate that digital gait biomarkers could serve as a non-invasive tool for diagnosing and monitoring motor impairments,” explains Dr. Cyril Brzenczek, postdoctoral researcher in the Biomedical Data Science group and first author of the study.
To classify motor symptom severity, the team also explored machine learning models based on gait sensor data compared to the MDS-Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) Part III motor examination score. “We could classify the severity of patients’ motor symptoms with success rates up to 75% and identify key predictors of motor severity,” explained Dr Quentin Klopfenstein, postdoctoral researcher and another first author of the study. “At-home gait assessments that capture similar changes could be developed as surrogate biomarkers to track motor impairment progression.” This could allow remote and continuous monitoring of motor symptoms, thereby facilitating timely adjustments in treatment, informed by current, real-time data rather than periodic clinical observations.
Classifying non-motor symptoms
The study also examined non-motor symptoms and comorbidities such as hallucinations, dopamine dysregulation syndrome, and depression. These comorbidities often make Parkinson’s disease harder to manage and reduce quality of life. The team found that integrating gait, metabolomics, and clinical data improved the detection of hallucinations, dyskinesia or freezing of gait with the latter achieving success rates of up to 91.7%.
This multimodal approach also revealed specific molecules as predictors associated with the outcomes studied. For example, cognitive performance as measured by the Montreal Cognitive Assessment (MoCA) score was strongly associated with changes in the abundance of specific metabolites, such as (S)-alpha-amino-omega-caprolactam, pointing to new potential molecular relationships. Further studies are needed to assess whether the changes of this metabolite occur as a result or cause of disease-associated processes.
Overall, the considered machine learning applications show that assessing the risk of developing certain symptoms or comorbidities associated with PD using a combination of simple sensors, clinical data and molecular analysis is possible and could allow clinicians to monitor such conditions better and earlier. Prof. Enrico Glaab, principal investigator of Biomedical Data Science group and senior author of the study, concludes: “Our research shows that integrating complementary data types, from digital gait measurements to molecular markers, could help develop more comprehensive approaches for tracking and detecting specific disease symptoms and comorbidities in people with Parkinson’s disease. This combined approach may contribute to better understanding individual differences in how the disease manifests and progresses.”
Original publication
Brzenczek, C., Klopfenstein, Q., Hähnel, T. et al. Integrating digital gait data with metabolomics and clinical data to predict outcomes in Parkinson’s disease. npj Digit. Med. 7, 235 (2024). https://doi.org/10.1038/s41746-024-01236-z
Funding
This study was conducted within the ERA PerMed project DIGIPD, as part of the European Union’s Horizon 2020 Programme for Research and Innovation. It has received financial support by the Luxembourg National Research Fund (FNR) and used data and samples from the National Centre of Excellence in Research on Parkinson’s disease (NCER-PD).