Automatic de-identification of French electronic health records: a cost-effective approach exploiting distant supervision and deep learning models.

BACKGROUND: Electronic health records (EHRs) contain valuable information for clinical research; however, the sensitive nature of healthcare data presents security and confidentiality challenges. De-identification is therefore essential to protect personal data in EHRs and comply with government regulations. Named entity recognition (NER) methods have been proposed to remove personal identifiers, with deep learning-based models achieving better performance. However, manual annotation of training data is time-consuming and expensive. The aim of this study was to develop an automatic de-identification pipeline for all kinds of clinical documents based on a distant supervised method to significantly reduce the cost of manual annotations and to facilitate the transfer of the de-identification pipeline to other clinical centers. METHODS: We proposed an automated annotation process for French clinical de-identification, exploiting data from the eHOP clinical data warehouse (CDW) of the CHU de Rennes and national knowledge bases, as well as other features. In addition, this paper proposes an assisted data annotation solution using the Prodigy annotation tool. This approach aims to reduce the cost required to create a reference corpus for the evaluation of state-of-the-art NER models. Finally, we evaluated and compared the effectiveness of different NER methods. RESULTS: A French de-identification dataset was developed in this work, based on EHRs provided by the eHOP CDW at Rennes University Hospital, France. The dataset was rich in terms of personal information, and the distribution of entities was quite similar in the training and test datasets. We evaluated a Bi-LSTM + CRF sequence labeling architecture, combined with Flair + FastText word embeddings, on a test set of manually annotated clinical reports. The model outperformed the other tested models with a significant F1 score of 96,96%, demonstrating the effectiveness of our automatic approach for deidentifying sensitive information. CONCLUSIONS: This study provides an automatic de-identification pipeline for clinical notes, which can facilitate the reuse of EHRs for secondary purposes such as clinical research. Our study highlights the importance of using advanced NLP techniques for effective de-identification, as well as the need for innovative solutions such as distant supervision to overcome the challenge of limited annotated data in the medical domain.

Reliability of Drug-Drug Interaction Measurement on Real-Word Data: The ReMIAMes Project.

The ReMIAMes project proposes a methodological framework to provide a reliable and reproducible measurement of the frequency of drug-drug interactions (DDI) when performed on real-world data. This framework relies on (i) a fine-grained and contextualized definition of DDIs, (ii) a shared minimum information model to select the appropriate data for the correct interpretation of potential DDIs, (iii) an ontology-based inference module able to handle missing data to classify prescription lines with potential DDIs, (iv) a report generator giving the value of the measurement and explanations when potential false positive are detected due to a lack of available data. All the tools developed are intended to be publicly shared under open license.

eHOP Clinical Data Warehouse: From a Prototype to the Creation of an Inter-Regional Clinical Data Centers Network.

Creation of networks such as clinical data centers within the hospital enables efficient exploitation of clinical data from a local to an inter-regional scope. This work present the structuration of the French Western Clinical Data Center Network (FWCDCN) conducted between 2016 and 2018. As of November 2018, FWCDCD is compounded with 7 institutions. CDW of the combinded Clinical Data Centers (CDC) contains the data of over 4 million patients followed since 2000.

Integrating Biobank Data into a Clinical Data Research Network: The IBCB Project.

Development of biobanks is still hampered by difficulty to collect high quality sample annotations using patient clinical information. The IBCB project evaluated the feasibility of a nationwide clinical data research network for this purpose. METHOD: the infrastructure, based on eHOP and I2B2 technologies, was interfaced with the legacy IT components of 3 hospitals. The evaluation focused on the data management process and tested 5 expert queries in Hepatocarcinoma. RESULTS: the integration of biobank data was comprehensive and easy. Five out of 5 queries were successfully performed and shown consistent results with the data sources excepted one query which required to search in unstructured data. The platform was designed to be scalable and showed that with few effort biobank data and clinical data can be integrated and leveraged between hospitals. Clinical or phenotyping concepts extraction techniques from free text could significantly improve the samples annotation with fine granularity information.

An Integrated Workflow For Secondary Use of Patient Data for Clinical Research.

This work proposes an integrated workflow for secondary use of medical data to serve feasibility studies, and the prescreening and monitoring of research studies. All research issues are initially addressed by the Clinical Research Office through a research portal and subsequently redirected to relevant experts in the determined field of concentration. For secondary use of data, the workflow is then based on the clinical data warehouse of the hospital. A datamart with potentially eligible research candidates is constructed. Datamarts can either produce aggregated data, de-identified data, or identified data, according to the kind of study being treated. In conclusion, integrating the secondary use of data process into a general research workflow allows visibility of information technologies and improves the accessability of clinical data.

Semantic integration of medication data into the EHOP Clinical Data Warehouse.

UNLABELLED: Reusing medication data is crucial for many medical research domains. Semantic integration of such data in clinical data warehouse (CDW) is quite challenging. Our objective was to develop a reliable and scalable method for integrating prescription data into EHOP (a French CDW). METHOD: PN13/PHAST was used as the semantic interoperability standard during the ETL process, and to store the prescriptions as documents in the CDW. Theriaque was used as a drug knowledge database (DKDB), to annotate the prescription dataset with the finest granularity, and to provide semantic capabilities to the EHOP query workbench. RESULTS: the system was evaluated on a clinical data set. Depending on the use case, the precision ranged from 52% to 100%, Recall was always 100%. CONCLUSION: interoperability standards and DKDB, document approach, and the finest granularity approach are the key factors for successful drug data integration in CDW.

Documentation in pharmacovigilance: using an ontology to extend and normalize Pubmed queries.

OBJECTIVES: To assess and understand adverse drug reactions (ADRs), a systematic review of reference databases like Pubmed is a necessary and mandatory step in Pharmacovigilance. In order to assist pharmacovigilance team with a computerized tool, we performed a comparative study of 4 different approaches to query Pubmed through ADR-drug terms. The aim of this study is to assess how an ontology of adverse effects, used to normalize and extend queries, could improve this search. MATERIAL AND METHOD: The ontological resource OntoEIM contains 58,000 classes and integrates MedDRA terminology. The entry point is a ADR-Drug term and the four methods are (i) a direct search on Pubmed (ii) a search with a normalized query enhanced with domain-specific Mesh Heading criteria, (iii) a search with the same elaborated query extended to the MeSH sub-hierarchy of the adverse effect entry and (iv) a search with a set of MedDRA terms grouped by subsomption in the OntoEIM ontology. For each of the 16 queries performed and analysed, relevant publications are selected "manually" by two pharmacovigilant experts. RESULTS: The recall is respectively of 63%, 50%, 67% and 74%, the precision of 13%, 26%, 29% and 4%. The best recall is provided by the ontology-based method, for 4 cases out of 16 this method returns relevant publications when the others return no results. CONCLUSION: Results show that an ontology-based search tool improves the recall performance, but other tools and methods are needed to raise the precision.

A knowledge management platform for documentation of case reports in pharmacovigilance.

Most countries have developed information systems to report drug adverse effects. However, as in other domains where systematic reviews are needed, there is little guidance on how systematic documentation of drug adverse effects should be performed. The objective of the VigiTermes project is to develop a platform to improve documentation of pharmacovigilance case reports for the pharmaceutical industry and regulatory authorities. In order to improve systematic reviews of adverse drug reactions, we developed a prototype that first reproduces and standardizes search strategies, then extracts information from the Medline abstracts which were retrieved and annotates them. The platform aims at providing transparent access and analysis tools to pharmacovigilance experts investigating relevance of safety signals related to drugs. The platform's architecture consists in the integration of two vendor tools ITM and Luxid and one academic web service for knowledge extraction from medical literature. Whereas a manual search performed by a pharmacovigilance expert retrieved 578 publications, the system proposed a list of 229 publications thus decreasing time required for review by 60%. Recall was 70% and additional developments are required in order to improve exhaustivity.

A contextual coding system for transplantation and end stage diseases.

The Establissement français des Greffes (EfG) is a state agency dealing with Public Health issues related to organ, tissue and cell transplantation in France. EfG maintains a national information system (EfG-IS) for the evaluation of organ transplantation activities. The EfG-IS is moving toward a new n-tier architecture comprising a terminological server. Because this terminological server is shared by various kind of transplant teams and dialysis centers to record patients data at different time point, contextualisation of terms appeared as a functional requirement. We report in this paper various contexts for medical terms and how they have been taken into account.

Medical pedagogical resources management.

The main objective of this work is to help the management of training resources for students using a pedagogical network available at the Medical School of Rennes. With the increase of the number of connections and the number of medical documents available on this network, the management of new contents requires a lot of efforts for the webmaster. In order to improve the management of the resources, we implemented an automatic web engine for teachers, able to manage the links for the most interesting resources for their practice.

Developing the ontological foundations of a terminological system for end-stage diseases, organ failure, dialysis and transplantation.

The Etablissement français des Greffes (EfG) is a state agency dealing with Public Health issues related to organ, tissue and cell transplantation in France. The evaluation of organ retrieval and transplantation activities, one of its missions, is supported by a national information system (EfG-IS). The EfG-IS is moving towards a new n-tier architecture comprising a terminology server for end-stage diseases, organ failure, dialysis and transplantation (EfG-TS). Following a preliminary audit of the existing coding system and in order to facilitate data recording, to improve the quality of information, to assume compatibility with terminological existing standards and to allow semantic interoperability with other local, national or international registries, a specific work has been conducted on the thesauri to integrate within the EfG-TS. In this paper focusing on the server's content rather than the container, we report first the functional and cognitive requirements that resulted from the preliminary audit. We then describe the methodological approach used to build the terminological server on "sound ontological foundations". We performed the semantic analysis of existing medical terms to set up disease description frame-like structures. These diseases description frames consist of a limited set of nosological discriminating slots such as etiology, semiology, pathology, evolution and associated diseases. Each relevant medical term is thus associated to a concept defined and inserted within a hierarchy according to disease description frame resulting from the semantic analysis. Last, because this terminological server is shared by various transplant and dialysis centers to record patient data at different time point, contextualization of terms appeared as one of the functional requirements. We will also point out various contexts for medical terms and how they have been taken into account.