JMIR Publications

ABSTRACT

Background:

Pharmacoepidemiologic studies using real-world data often lack clinical context, much of which is embedded in free-text electronic health records (EHRs). However, sharing EHR narratives is restricted by privacy requirements, and conventional deidentification approaches may reduce analytic utility and limit auditability.

Objective:

To develop and evaluate a framework that (1) extracts structured cancer-related entities from Japanese EHR free text using BERT (Bidirectional Encoder Representations from Transformers)-based natural language processing (NLP) and (2) generates deidentified analytic text that preserves contextual information to support post-hoc auditing under privacy constraints.

Methods:

We conducted a retrospective observational study using the DATuM IDEA database, which integrates unstructured EHR narratives with structured records (claims, prescriptions, and procedure/surgery records) from two hospitals in Japan from January 1, 2019, through June 30, 2025. Deterministic linkage was performed within ICI (Integrated Clinical Care Informatics, Inc.) prior to deidentification. We quantified deidentification using masking and restoration rates and evaluated extraction success rates for TNM elements, internal logical consistency (M1 vs. Stage IV), and proxy-based external concordance using structured-data treatment proxies (ATC code L for systemic anticancer therapy and procedure-name keyword searches for cancer-directed interventions). A stratified manual audit of 100 narratives was conducted to assess entity retention (sanity check) and residual protected health information (PHI).

Results:

Among 51,876 patients with recorded diagnoses, 10,214 had neoplasms (ICD-10 C00–D48), and deidentified analytic text was available for 4,383 patients. The linkage yield was 99.2% (4,346/4,383). Median masking rates ranged from 9.9% to 15.0% across major cancer categories. TNM extraction success rates were 0.807 (T), 0.798 (N), and 0.805 (M). Concordance between extracted M1 and Stage IV was 0.853. Proxy-based validation showed a positive predictive value (PPV) of 0.700 for Stage IV (systemic therapy proxy) and a PPV of 0.031 for early-stage classification (procedure proxy). Manual audit demonstrated complete capture of TNM/stage information after normalization in generated texts and no residual PHI in the audited sample.

Conclusions:

This framework enables the generation of deidentified Japanese EHR narratives that preserve contextual structure for auditing while supporting structured entity extraction, thereby addressing the trade-off between privacy protection and reproducibility in real-world data research.