Classifying the risk of cognitive impairment in Parkinson's disease using serum bile acid profiles and machine learning.

Cognitive impairment (CI) is a common and early non-motor manifestation of Parkinson's disease (PD), yet its biochemical basis remains poorly understood. Given the emerging link between bile acids (BAs) and neurodegeneration, we investigated whether serum BA profiles differ by cognitive status in PD and whether they can classify CI. A total of 363 participants were enrolled, including 63 healthy controls, 154 PD patients with normal cognition, and 146 with CI. Serum BA concentrations were quantified by ultra-performance liquid chromatography-tandem mass spectrometry, and multivariate as well as machine learning analyses were applied. Compared with cognitively normal PD patients, those with CI exhibited distinct BA alterations, characterized by elevated deoxycholic and cholic acids and reduced glyco- and tauro-conjugated species. Deoxycholic acid showed the strongest negative correlations with cognitive scores. Machine learning models based on combined BA profiles, particularly the random forest classifier, achieved robust discrimination between PD-CI and PD-NC groups (AUC up to 0.90). These findings indicate that BA dysregulation is closely linked to cognitive impairment in PD and may serve as a promising metabolic biomarker for early detection. Clinical trial number. Not applicable.