Search papers, labs, and topics across Lattice.
To unlock privacy-restricted clinical records without institutional data leakage, the authors developed MedDeID, an on-premises framework that trains compact transformer models for clinical text de-identification using either real hospital data or purely synthetic notes. Evaluated on an adjudicated Dutch hospital benchmark, the hospital-trained model achieved 98.9% identifier recall with just 0.24% over-redaction, while the synthetic-only model reached 96.1% recall. Crucially, the synthetic-trained model outperformed its real-data counterpart on out-of-domain primary-care notes (90.3% versus 87.0% recall) while exhibiting higher robustness to identifier-format perturbations.
Compact transformers trained purely on synthetic clinical text can match or exceed real-data de-identification performance while proving more robust to format perturbations across out-of-distribution primary care domains.
Clinical notes contain personally identifiable information (PII), restricting reuse for research and medical AI, especially when data cannot leave an institution. We developed MedDeID, an on-premises framework combining in-house annotation and synthetic-note generation with model training, inference, pseudonymisation and evaluation. On an independently annotated, adjudicated 300-note Dutch hospital benchmark, a hospital-trained compact transformer detected 98.9% of identifying text while redacting 0.24% of text outside annotated identifiers; a synthetic-only counterpart detected 96.1%. On 100 primary-care notes, the synthetic-trained model achieved higher recall than the hospital-trained model (90.3% versus 87.0%) and greater robustness to identifier-format perturbations. An English instantiation trained without real text detected 99.7% and 98.9% of annotated identifier characters on two external synthetic benchmarks. These results demonstrate transfer of the workflow to another language, but not clinical English performance. MedDeID provides a route to locally governed de-identification using real or synthetic training data.
