The performance of biomarkers for the diagnosis of Parkinson's disease: A systematic review.

Abstract

Background: Early diagnosis of Parkinson's disease remains challenging due to the current clinical diagnostic approach. With machine learning emerging as a powerful tool for biomarker discovery, we aim to determine whether biomarkers processed by machine learning models can enable earlier detection of Parkinson's disease.

Methods: We conducted a literature search with a 10-year limit that yielded 161 biomarkers derived from serum, cerebrospinal fluid and genes. Initially, biomarkers were classified into four groups according to the biological timeline of Parkinson's disease pathogenesis. Subsequently, we further organised the biomarkers into two categories: pre-motor phase and motor phase biomarkers. Two analyses were then conducted according to the aforementioned classifications, with the performance of biomarkers evaluated via their AUC values as derived from machine learning models.

Results: No significant differences were found in either analysis, suggesting that all biomarkers, regardless of their role in the biological sequences underpinning Parkinson's disease pathogenesis, nor their association with the pre-motor or motor phases of Parkinson's disease, have the potential to serve as equally valid diagnostic predictors. Additionally, we identified 26 top-performing biomarkers with high AUC values (>0.8).

Conclusion: The main finding in our analyses was that pre-motor phase biomarkers, which offer the advantage of enabling an earlier diagnosis compared to clinical methods, can achieve a comparably high level of diagnostic accuracy as motor phase biomarkers. Therefore, our foremost suggestion is further research into the clinical viability of pre-motor phase biomarkers that compose part of the aforementioned 26 top-performing biomarkers.