Improving Interoperability and Generalizability through Technology in Parkinson’s Disease

Abstract

This cumulative dissertation investigates how the opportunities presented by rapidly expanding health data can be effectively realized in clinical research and care, using Parkinson’s disease as its main focus. While the volume and diversity of health data have grown exponentially, driven by digitization, novel data sources, and patient participation, these developments pose new challenges in terms of data quality, integration into existing systems, and the generation of actionable clinical knowledge.

Within the included publications, multiple approaches to interoperability are explored, including the prospective design of digital tools for standardized collection of patient-reported outcomes, retrospective structuring of clinical documents using vision-language models, and the harmonization of wearable sensor data. The implementation of a research platform further demonstrates how secure, standards-based secondary data use enables hypothesis-driven analysis on large clinical cohorts while ensuring privacy. Building on the interoperable data, the thesis explores analytic strategies to improve the robustness and generalizability of machine learning models. Both approaches, grounded in clinical knowledge and transfer learning techniques, are evaluated for their capacity to enable effective reuse of data and models across diverse datasets and tasks.

In summary, the dissertation provides concrete evidence that robust digital infrastructures, a systematic focus on interoperability, and the integration of clinical expertise with machine learning methods are crucial for developing reliable and adaptable models in healthcare. Although ongoing challenges remain, particularly in integrating multimodal data and advancing data standards, the solutions outlined here offer practical pathways toward scalable, reproducible, and clinically meaningful informatics within contemporary hospital settings.