De-identification of health data for open sharing: A scoping review

Open science and data sharing initiatives are increasingly shaping health research by promoting transparency and collaboration. However, sharing health data introduces privacy risks, requiring robust de-identification methods. Previous reviews have largely focused on specific data types, with limited synthesis across diverse and multimodal health data. This scoping review mapped de-identification methods across health data modalities and identified common techniques, tools, challenges, and gaps.

We conducted a scoping review guided by the Arksey and O'Malley framework, with enhancements from Levac et al. and the Joanna Briggs Institute. Searches of MEDLINE, EMBASE, and Web of Science identified 2434 documents. After duplicate removal, 2384 documents underwent full-text screening. Data was charted using an iteratively refined extraction tool capturing study characteristics, de-identification methods, data modalities, evaluation approaches, and reported outcomes. Quantitative and inductive content analyses were conducted.

A total of 309 documents met inclusion criteria. Most focused on unstructured data (n = 91, 29.4 %), structured data (n = 86, 27.8 %), or medical images (n = 67, 21.7 %). Single-modality data were addressed in 88.7 % of documents (n = 274), while 11.3 % (n = 35) addressed multimodal data. Hybrid approaches combining multiple techniques were more common (n = 173, 56.0 %) than single-method approaches (n = 136, 44.0 %). Most methods were developed in academic or research settings (n = 262, 84.8 %), with limited clinical implementation (n = 19, 6.1 %). Patient/public involvement was rare (n = 14, 4.5 %).

Common author-reported challenges included preserving data utility, scalability, and inconsistent reporting standards. Findings highlight the need for modality-specific evaluation frameworks and standardized reporting to support trustworthy, privacy-preserving health data sharing.