Investigation of the Utility of Features in a Clinical De-identification Model: A Demonstration Using EHR Pathology Reports for Advanced NSCLC Patients.

Tanmoy Paul,Humayera Islam,Michael Barnes,Vasanthi Mandhadi,Nitesh Singh,Richard D Hammer,Teja Venkat Pavan Kotapati,Vishakha Sharma,Abu Saleh Mohammad Mosa,Yaswitha Jampani,Md Kamruz Zaman Rana,Preethi Aishwarya Tautam

doi:10.3389/fdgth.2022.728922

Abstract

BackgroundElectronic health record (EHR) systems contain a large volume of texts, including visit notes, discharge summaries, and various reports. To protect the confidentiality of patients, these records often need to be fully de-identified before circulating for secondary use. Machine learning (ML) based named entity recognition (NER) model has emerged as a popular technique of automatic de-identification.ObjectiveThe performance of a machine learning model highly depends on the selection of appropriate features. The objective of this study was to investigate the usability of multiple features in building a conditional random field (CRF) based clinical de-identification NER model.MethodsUsing open-source natural language processing (NLP) toolkits, we annotated protected health information (PHI) in 1,500 pathology reports and built supervised NER models using multiple features and their combinations. We further investigated the dependency of a model's performance on the size of training data.ResultsAmong the 10 feature extractors explored in this study, n-gram, prefix–suffix, word embedding, and word shape performed the best. A model using combination of these four feature sets yielded precision, recall, and F1-score for each PHI as follows: NAME (0.80; 0.79; 0.80), LOCATION (0.85; 0.83; 0.84), DATE (0.86; 0.79; 0.82), HOSPITAL (0.96; 0.93; 0.95), ID (0.99; 0.82; 0.90), and INITIALS (0.97; 0.49; 0.65). We also found that the model's performance becomes saturated when the training data size is beyond 200.ConclusionManual de-identification of large-scale data is an impractical procedure since it is time-consuming and subject to human errors. Analysis of the NER model's performance in this study sheds light on a semi-automatic clinical de-identification pipeline for enterprise-wide data warehousing.

Highlights

Clinical texts are vital components of electronic health records (EHR) and can be an enriched knowledge source for medical research
The key contributions of this study are twofold: (i) Firstly, our investigation will be helpful in choosing the proper set of features to build a de-identification named entity recognition (NER) model from a wide range of features at one’s disposal, and (ii) secondly, we propose a framework of a semi-automatic clinical deidentification pipeline which can play a significant role in an enterprise-wide data warehousing
Based on our analysis, it can be concluded that NG, PS, Word Embedding (WE), and Word Shape (WS) are the best features in building a clinical de-identification NER model

Summary

Introduction

Clinical texts are vital components of electronic health records (EHR) and can be an enriched knowledge source for medical research. Text-based medical records often contain potential patient identifiers and confidential information that must not be shared with third parties for ethical and legal reasons. The Health Insurance Portability and Accountability Act (HIPAA) in the United States requires removing patientprotected health information (PHI) from the medical records before sharing for secondary use [1]. The PHI items include name, geographic location, phone number, social security number, medical record number, etc. Electronic health record (EHR) systems contain a large volume of texts, including visit notes, discharge summaries, and various reports. To protect the confidentiality of patients, these records often need to be fully de-identified before circulating for secondary use. Machine learning (ML) based named entity recognition (NER) model has emerged as a popular technique of automatic de-identification

Methods

Results

Discussion

Conclusion