Adrien Coulet
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Adrien Coulet has a total of 136 co-authors. Here are the top co-authors by number of shared publications:
- Pierre Monnin
Université Côte d’Azur, Inria, CNRS, I3S, Sophia-Antipolis, France
9 shared publications Malika Smaïl-Tabbone
Université de Lorraine, CNRS, Inria, LORIA, Nancy, France
8 shared publications- Amedeo Napoli
Université de Lorraine, CNRS, LORIA
7 shared publications Amedeo Napoli
Université de Lorraine, CNRS, Inria, LORIA
6 shared publications- Marie-Dominique Devignes
CNRS, Inria, LORIA, Université de Lorraine, Nancy, France
6 shared publications - Clement Jonquet
LIRMM University of Montpellier CNRS France
4 shared publications
Latest Publications
- Open Access
- 30 September 2026
- Published by Elsevier in SSRN Electronic Journal
MEDS-OWL: An Ontology-Driven Framework for Clinical Outcome Prediction with Graph Learning
Objective: To develop a formal semantic representation of the Medical Event Data Standard (MEDS), an event-centric standard for structured clinical data, using Semantic Web technologies and supporting reproducible graph-based machine learning. We further investigate how the semantic granularity of the resulting graph representation affects clinical outcome prediction.Methods: We present MEDS-OWL, a lightweight ontology that extends MEDS with a formal semantic representation, together with meds2rdf, a scalable Python library for transforming MEDS-compliant datasets into FAIR-aligned MEDS-RDF graphs. Round-trip fidelity and scalability were evaluated on four heterogeneous clinical datasets (NEUROVASC, MIMIC-IV, eICU, and NWICU), comprising more than 309,000 hospital stays. For downstream evaluation, 11 prediction tasks derived from MIMIC-IV were formulated as patient-level binary classification problems. A two-layer Relational Graph Convolutional Network with literal-aware encoding was evaluated across three MEDS-RDF schema variants differing in subject–event predicate granularity, enabling systematic assessment of the influence of representation granularity on graph-based learning.Results: The framework preserved structural completeness and the core MEDS fields with high fidelity across the four datasets. Its streaming conversion architecture enabled processing RDF graphs of up to 937 GB, achieving a 2.7× speedup over its predecessor. Across the 11 prediction tasks, increasing semantic granularity from MEDS-RDF-Base to Partial to Full consistently improved Macro-F1, narrowing the performance gap with tabular baselines.Conclusions: MEDS-OWL provides a scalable, interoperable, and FAIR-aligned framework for representing MEDS data as semantic clinical graphs. The results demonstrate that semantic representation extends beyond interoperability and therefore should be treated as an important determinant of graph-based clinical machine learning performance in predictive analytics for clinical outcomes.
- 30 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessMEDS-OWL: An Ontology-Driven Framework for Clinical Outcome Prediction with Graph Learning
Objective: To develop a formal semantic representation of the Medical Event Data Standard (MEDS), an event-centric standard for structured clinical data, using Semantic Web technologies and supporting reproducible graph-based machine learning. We further investigate how the semantic granularity of the resulting graph representation affects clinical outcome prediction.Methods: We present MEDS-OWL, a lightweight ontology that extends MEDS with a formal semantic representation, together with meds2rdf, a scalable Python library for transforming MEDS-compliant datasets into FAIR-aligned MEDS-RDF graphs. Round-trip fidelity and scalability were evaluated on four heterogeneous clinical datasets (NEUROVASC, MIMIC-IV, eICU, and NWICU), comprising more than 309,000 hospital stays. For downstream evaluation, 11 prediction tasks derived from MIMIC-IV were formulated as patient-level binary classification problems. A two-layer Relational Graph Convolutional Network with literal-aware encoding was evaluated across three MEDS-RDF schema variants differing in subject–event predicate granularity, enabling systematic assessment of the influence of representation granularity on graph-based learning.Results: The framework preserved structural completeness and the core MEDS fields with high fidelity across the four datasets. Its streaming conversion architecture enabled processing RDF graphs of up to 937 GB, achieving a 2.7× speedup over its predecessor. Across the 11 prediction tasks, increasing semantic granularity from MEDS-RDF-Base to Partial to Full consistently improved Macro-F1, narrowing the performance gap with tabular baselines.Conclusions: MEDS-OWL provides a scalable, interoperable, and FAIR-aligned framework for representing MEDS data as semantic clinical graphs. The results demonstrate that semantic representation extends beyond interoperability and therefore should be treated as an important determinant of graph-based clinical machine learning performance in predictive analytics for clinical outcomes.
- Open Access
- 14 September 2026
- Published by Springer Nature in Communications Medicine
- Vol. 6 (1) , 482
MAKAAO, an open access resource linking human autoantibodies and autoimmune diseases
- 14 September 2026
- Published by Springer Nature in Communications Medicine
- Vol. 6 (1) , 482
Open AccessMAKAAO, an open access resource linking human autoantibodies and autoimmune diseases
- Open Access
- 21 May 2026
- Published by IOS Press
- Vol. 336, 979-983
Build and Query Indexes of Clinical Documents with Easy-to-Reuse Pipelines
Electronic Health Records are a central source of healthcare data, containing structured data alongside unstructured clinical texts. The latter capture detailed reasoning, observations, treatment plans and clinical evolutions, which are crucial for phenotyping, and real-world evidence generation. Natural language processing enables the extraction, thus the subsequent use, of these crucial elements; however, these extractions remain one-off, study-specific efforts. This is detrimental as the extracted elements could be valuable for future research. We present medkit Seshat, an open-source Python pipeline that: (1) ingests free text, (2) recognizes relevant entities, (3) normalizes them with OMOP vocabularies, (4) builds an index that can either be searched by concept or by document. In addition, we share a flexible web UI to illustrate the interest of built indexes in terms of search, text analysis and export. Seshat aims at facilitating the reuse and adaptation of this prototypical pipeline to various purposes, with the main objective of enabling the secondary use of results of phenotyping campaigns.
- 21 May 2026
- Published by IOS Press
- Vol. 336, 979-983
Open AccessBuild and Query Indexes of Clinical Documents with Easy-to-Reuse Pipelines
Electronic Health Records are a central source of healthcare data, containing structured data alongside unstructured clinical texts. The latter capture detailed reasoning, observations, treatment plans and clinical evolutions, which are crucial for phenotyping, and real-world evidence generation. Natural language processing enables the extraction, thus the subsequent use, of these crucial elements; however, these extractions remain one-off, study-specific efforts. This is detrimental as the extracted elements could be valuable for future research. We present medkit Seshat, an open-source Python pipeline that: (1) ingests free text, (2) recognizes relevant entities, (3) normalizes them with OMOP vocabularies, (4) builds an index that can either be searched by concept or by document. In addition, we share a flexible web UI to illustrate the interest of built indexes in terms of search, text analysis and export. Seshat aims at facilitating the reuse and adaptation of this prototypical pipeline to various purposes, with the main objective of enabling the secondary use of results of phenotyping campaigns.
- 27 March 2026
- Published by Agence Bibliographique de l'Enseignement Supérieur
Construction et utilisation d'une base de connaissances pharmacogénomique pour l'intégration de données et la découverte de connaissances
Cette thèse porte sur l’utilisation d’ontologies et de bases de connaissances pour guider différentes étapes du processus d’extraction de connaissances à partir de bases de données (ECBD) et une application dans le domaine de la pharmacogénomique. Les données relatives à ce domaine sont hétérogènes, complexes, et distribuées dans diverses bases de données, ce qui rend cruciale l’étape préliminaire de préparation et d’intégration des données à fouiller. Je propose pour guider cette étape une approche originale d’intégration de données qui s’appuie sur une représentation des connaissances du domaine sous forme de deux ontologies en logiques de description : SNP-Ontology et SO-Pharm. Cette approche a été implémentée grâce aux technologies du Web sémantique et conduit au peuplement d’une base de connaissances pharmacogénomique. Le fait que les données à fouiller soient alors disponibles dans une base de connaissances entraîne de nouvelles potentialités pour le processus d’extraction de connaissances. Je me suis d’abord intéressé au problème de la sélection des données les plus pertinentes à fouiller en montrant comment la base de connaissances peut être exploitée dans ce but. Ensuite j’ai décrit et appliqué à la pharmacogénomique, une méthode qui permet l’extraction de connaissances directement à partir d’une base de connaissances. Cette méthode appelée Analyse des Assertions de Rôles (ou AAR) permet d’utiliser des algorithmes de fouille de données sur un ensemble d’assertions de la base de connaissances pharmacogénomique et d’expliciter des connaissances nouvelles et pertinentes qui y étaient enfouies.
- 27 March 2026
- Published by Agence Bibliographique de l'Enseignement Supérieur
Construction et utilisation d'une base de connaissances pharmacogénomique pour l'intégration de données et la découverte de connaissances
Cette thèse porte sur l’utilisation d’ontologies et de bases de connaissances pour guider différentes étapes du processus d’extraction de connaissances à partir de bases de données (ECBD) et une application dans le domaine de la pharmacogénomique. Les données relatives à ce domaine sont hétérogènes, complexes, et distribuées dans diverses bases de données, ce qui rend cruciale l’étape préliminaire de préparation et d’intégration des données à fouiller. Je propose pour guider cette étape une approche originale d’intégration de données qui s’appuie sur une représentation des connaissances du domaine sous forme de deux ontologies en logiques de description : SNP-Ontology et SO-Pharm. Cette approche a été implémentée grâce aux technologies du Web sémantique et conduit au peuplement d’une base de connaissances pharmacogénomique. Le fait que les données à fouiller soient alors disponibles dans une base de connaissances entraîne de nouvelles potentialités pour le processus d’extraction de connaissances. Je me suis d’abord intéressé au problème de la sélection des données les plus pertinentes à fouiller en montrant comment la base de connaissances peut être exploitée dans ce but. Ensuite j’ai décrit et appliqué à la pharmacogénomique, une méthode qui permet l’extraction de connaissances directement à partir d’une base de connaissances. Cette méthode appelée Analyse des Assertions de Rôles (ou AAR) permet d’utiliser des algorithmes de fouille de données sur un ensemble d’assertions de la base de connaissances pharmacogénomique et d’expliciter des connaissances nouvelles et pertinentes qui y étaient enfouies.
- 11 October 2025
- Published by Oxford University Press (OUP) in Journal of the American Medical Informatics Association
- Vol. 32 (12) , 1855-1864
Efficient extraction of medication information from clinical notes: an evaluation in 2 languages
Objective: To evaluate the accuracy, computational cost, and portability of a new natural language processing (NLP) method for extracting medication information from clinical narratives. Materials and Methods: We propose an original transformer-based architecture for the extraction of entities and their relations pertaining to patients’ medication regimen. First, we used this approach to train and evaluate a model on French clinical notes, using a newly annotated corpus from Hôpitaux Universitaires de Strasbourg. Second, the portability of the approach was assessed by conducting an evaluation on clinical documents in English from the 2018 n2c2 shared task. Information extraction accuracy and computational cost were assessed by comparison with an available method using transformers. Results: The proposed architecture achieves on the task of relation extraction itself performance that are competitive with the state-of-the-art on both French and English (F-measures 0.82 and 0.96 vs 0.81 and 0.95), but reduces the computational cost by 10. End-to-end (Named Entity recognition and Relation Extraction) F1 performance is 0.69 and 0.82 for French and English corpus. Discussion: While an existing system developed for English notes was deployed in a French hospital setting with reasonable effort, we found that an alternative architecture offered end-to-end drug information extraction with comparable extraction performance and lower computational impact for both French and English clinical text processing, respectively. Conclusion: The proposed architecture can be used to extract medication information from clinical text with high performance and low computational cost and consequently suits with usually limited hospital IT resources.
- 11 October 2025
- Published by Oxford University Press (OUP) in Journal of the American Medical Informatics Association
- Vol. 32 (12) , 1855-1864
Efficient extraction of medication information from clinical notes: an evaluation in 2 languages
Objective: To evaluate the accuracy, computational cost, and portability of a new natural language processing (NLP) method for extracting medication information from clinical narratives. Materials and Methods: We propose an original transformer-based architecture for the extraction of entities and their relations pertaining to patients’ medication regimen. First, we used this approach to train and evaluate a model on French clinical notes, using a newly annotated corpus from Hôpitaux Universitaires de Strasbourg. Second, the portability of the approach was assessed by conducting an evaluation on clinical documents in English from the 2018 n2c2 shared task. Information extraction accuracy and computational cost were assessed by comparison with an available method using transformers. Results: The proposed architecture achieves on the task of relation extraction itself performance that are competitive with the state-of-the-art on both French and English (F-measures 0.82 and 0.96 vs 0.81 and 0.95), but reduces the computational cost by 10. End-to-end (Named Entity recognition and Relation Extraction) F1 performance is 0.69 and 0.82 for French and English corpus. Discussion: While an existing system developed for English notes was deployed in a French hospital setting with reasonable effort, we found that an alternative architecture offered end-to-end drug information extraction with comparable extraction performance and lower computational impact for both French and English clinical text processing, respectively. Conclusion: The proposed architecture can be used to extract medication information from clinical text with high performance and low computational cost and consequently suits with usually limited hospital IT resources.
- 1 June 2025
- Published by Springer Nature
- p. 282-300
Predicting Clinical Outcomes from Patient Care Pathways Represented with Temporal Knowledge Graphs
- 1 June 2025
- Published by Springer Nature
- p. 282-300
Predicting Clinical Outcomes from Patient Care Pathways Represented with Temporal Knowledge Graphs
- Open Access
ProtoDrift: Evaluating the impact of small-to-big adjustments to chemotherapy protocols
Adjustments to chemotherapy protocols are common to adapt treatments to individual patient needs, yet consensus on the impact of such adjustements is lacking. We introduce ProtoDrift, a novel metric that quantifies time and dose adjustments in chemotherapies. This weighted distance enables an assessment of their impact on patient outcomes, offering more detailed analyses than the traditional Relative Dose Intensity (RDI). We compared ProtoDrift and RDI prediction performances through survival analyses on 20,808 patients across 38 groups, categorised by cancer location and treatment line at two hospitals. Without optimisation, ProtoDrift achieves either comparable or better prediction results in 71% of patient groups (27 out of 38). Once optimised, ProtoDrift surpasses the RDI C-index predictions in 89% (16 out of 18) of patient groups from the first hospital. This study confirms ProtoDrift as an advanced tool for refining chemotherapy regimen design, highlighting the critical role of time adjustments in patient outcomes.
Open AccessProtoDrift: Evaluating the impact of small-to-big adjustments to chemotherapy protocols
Adjustments to chemotherapy protocols are common to adapt treatments to individual patient needs, yet consensus on the impact of such adjustements is lacking. We introduce ProtoDrift, a novel metric that quantifies time and dose adjustments in chemotherapies. This weighted distance enables an assessment of their impact on patient outcomes, offering more detailed analyses than the traditional Relative Dose Intensity (RDI). We compared ProtoDrift and RDI prediction performances through survival analyses on 20,808 patients across 38 groups, categorised by cancer location and treatment line at two hospitals. Without optimisation, ProtoDrift achieves either comparable or better prediction results in 71% of patient groups (27 out of 38). Once optimised, ProtoDrift surpasses the RDI C-index predictions in 89% (16 out of 18) of patient groups from the first hospital. This study confirms ProtoDrift as an advanced tool for refining chemotherapy regimen design, highlighting the critical role of time adjustments in patient outcomes.
- Open Access
Comparing representations of long clinical texts for the task of patient note-identification
In this paper, we address the challenge of patient-note identification, which involves accurately matching an anonymized clinical note to its corresponding patient, represented by a set of related notes. This task has broad applications, including duplicate records detection and patient similarity analysis, which require robust patient-level representations. We explore various embedding methods, including Hierarchical Attention Networks (HAN), three-level Hierarchical Transformer Networks (HTN), LongFormer, and advanced BERT-based models, focusing on their ability to process mediumto-long clinical texts effectively. Additionally, we evaluate different pooling strategies (mean, max, and mean_max) for aggregating wordlevel embeddings into patient-level representations and we examine the impact of sliding windows on model performance. Our results indicate that BERT-based embeddings outperform traditional and hierarchical models, particularly in processing lengthy clinical notes and capturing nuanced patient representations. Among the pooling strategies, mean_max pooling consistently yields the best results, highlighting its ability to capture critical features from clinical notes. Furthermore, the reproduction of our results on both MIMIC dataset and Necker hospital data warehouse illustrates the generalizability of these approaches to real-world applications, emphasizing the importance of both embedding methods and aggregation strategies in optimizing patient-note identification and enhancing patient-level modeling.
Open AccessComparing representations of long clinical texts for the task of patient note-identification
In this paper, we address the challenge of patient-note identification, which involves accurately matching an anonymized clinical note to its corresponding patient, represented by a set of related notes. This task has broad applications, including duplicate records detection and patient similarity analysis, which require robust patient-level representations. We explore various embedding methods, including Hierarchical Attention Networks (HAN), three-level Hierarchical Transformer Networks (HTN), LongFormer, and advanced BERT-based models, focusing on their ability to process mediumto-long clinical texts effectively. Additionally, we evaluate different pooling strategies (mean, max, and mean_max) for aggregating wordlevel embeddings into patient-level representations and we examine the impact of sliding windows on model performance. Our results indicate that BERT-based embeddings outperform traditional and hierarchical models, particularly in processing lengthy clinical notes and capturing nuanced patient representations. Among the pooling strategies, mean_max pooling consistently yields the best results, highlighting its ability to capture critical features from clinical notes. Furthermore, the reproduction of our results on both MIMIC dataset and Necker hospital data warehouse illustrates the generalizability of these approaches to real-world applications, emphasizing the importance of both embedding methods and aggregation strategies in optimizing patient-note identification and enhancing patient-level modeling.
- Open Access
- 12 February 2025
- Published by Elsevier in Journal of Biomedical Informatics
- Vol. 163, 104788
Enhancing clinical data warehousing with provenance data to support longitudinal analyses and large file management: The gitOmmix approach for genomic and image data
- 12 February 2025
- Published by Elsevier in Journal of Biomedical Informatics
- Vol. 163, 104788
Open AccessEnhancing clinical data warehousing with provenance data to support longitudinal analyses and large file management: The gitOmmix approach for genomic and image data
- Open Access
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 109-123
Comparing representations of long clinical texts for the task of patient-note identification
Safa Alsaidi, Marc Vincent, Olivia Boyer, Nicolas Garcelon, Miguel Couceiro, Adrien Coulet. Proceedings of the Second Workshop on Patient-Oriented Language Processing (CL4Health). 2025.
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 109-123
Open AccessComparing representations of long clinical texts for the task of patient-note identification
Safa Alsaidi, Marc Vincent, Olivia Boyer, Nicolas Garcelon, Miguel Couceiro, Adrien Coulet. Proceedings of the Second Workshop on Patient-Oriented Language Processing (CL4Health). 2025.