Principle-Based Approach for the De-Identification of Code-Mixed Electronic Health Records

Abstract

Code-mixing is a phenomenon when at least two languages combined in a hybrid way in the context of a single conversation. The use of mixed language is widespread in multilingual and multicultural countries and poses significant challenges for the development of automated language processing tools. In Taiwan's electronic health record (EHR) systems, the unstructured EHR texts are usually represented in the mixing of English and Chinese languages resulting in the difficulty for de-identification and synthetization of protected health information (PHI). We explored this problem by applied several state-of-the-art pretrained mono-and multilingual language models and proposed to apply the principle-based approach (PBA) for the tasks of PHI recognition and resynthesis on a code-mixed EHR corpus, which was annotated with 6 main categories and 25 subcategories of PHIs. In PBA, a hierarchical principle slot schema is defined to encode knowledge of code-mixed PHIs and the defined slots were learned from the training set to assemble into principles for recognizing PHI mentions and synthesizing surrogates at the same time. A semantic disambiguation process is developed used to disambiguate ambiguous PHI categories in the de-identification process and to dynamically extend the knowledge encoded in PBA during the knowledge augmentation process. The experimental results demonstrate that the proposed method can achieve the best micro-and macro-F-scores performance in comparison with the other mono-and multilingual language models finetuned on our code-mixed corpus.

Unlocking the Secrets Behind Advanced Artificial Intelligence Language Models in Deidentifying Chinese-English Mixed Clinical Text: Development and Validation Study

“…For example, Tang et al [15] fine-tuned M-BERT on their English-simplified Chinese social media data set for multi-label sentiment analysis, and obtained an F-score of 0.69, which was 15% higher than that for the model without M-BERT. While limited research exists on using BERT-based models or other LLMs for code-mixing deidentification in clinical data sets, our previous work [38] suggested the potential benefits of incorporating M-BERT to disambiguate PHI categories in code-mixed sentences. In this study, we focused on comprehending how BERT-family PLMs handle Chinese-English mixed issues that arise in actual clinical text and assessed the feasibility of using state-of-the-art LLMs to recognize PHI from sampled clinical text.…”

“…The test set comprised 200 discharge summaries along with 60,632 sentences. Finally, the principle-based resynthesis method proposed in our previous work [38] was used to generate surrogates, and the entire corpus was rechecked by one of the senior annotators (PTC) to ensure a high level of data consistency and correctness.…”