Development and clinical translation of smartphone-based digital biomarkers for Parkinson's disease: a structured review.
TL;DR
Smartphone-based digital biomarkers offer a promising approach to improving diagnostic accuracy in PD and are emerging as a frontier in precision neurology and scalable PD care, however, these findings have several limitations and should be interpreted with caution.
Abstract
Diagnosis of Parkinson's disease (PD) remains a major clinical challenge, particularly in the prodromal and early clinically evident stages, when symptoms are subtle and phenotypically overlap with other mimicking disorders. The widespread availability of smartphones equipped with inertial, acoustic, touchscreen, and geolocation sensors has enabled the emergence of smartphone-based digital biomarkers that capture continuous, real-world motor and non-motor manifestations of PD. To critically review the validation state, construct validity, clinical applications, and potential for longitudinal monitoring of smartphone-derived digital biomarkers for early PD detection and for differential diagnosis from mimicking disorders, while addressing methodological, regulatory, and implementation challenges. This manuscript reviews the current evidence on smartphone-based active and passive digital assessments for the early diagnosis and screening of PD. We examine clinical validation and implementation, as well as advances in technological processing, multimodal digital phenotyping, and machine learning models, for the early detection and screening of PD. Smartphone-derived digital markers demonstrate promising capability for detecting several symptoms and support early diagnosis of PD. A total of 30 studies showed significant correlations of such markers with established clinical scales and suggested utility in differentiating PD from other Parkinsonian syndromes and prodromal PD patients. Six major categories based on the clinical assessment were identified: 8 studies with voice-based assessment, 6 with touchscreen-based assessment, 4 with video-recording assessment, 4 with inertial measurement unit-based assessment, 7 with multimodal assessment approaches, and 1 with questionnaire-based smartphone assessment. However, substantial limitations persist, including device and protocol heterogeneity, limited external validation, small datasets, algorithmic bias, inadequate explainability, and inconsistent reporting standards. Smartphone-based digital biomarkers offer a promising approach to improving diagnostic accuracy in PD and are emerging as a frontier in precision neurology and scalable PD care. However, these findings have several limitations and should be interpreted with caution. Future research should prioritise multicentre longitudinal validation, multimodal integration, explainable AI, and real-world deployment to establish clinically actionable digital biomarkers for early PD detection and differential diagnosis.