Publications by authors named "Marc Cuggia"
Monitoring Guideline Adherence in Severe Traumatic Brain Injury.
Stud Health Technol Inform
November 2024
Traumatic brain injuries (TBI) significantly impact global health, often resulting in death or long-term disability. We developed a quality dashboard to monitor adherence to severe TBI guidelines, leveraging data from Rennes University Hospital's clinical data warehouse collected between January 2020 and December 2023. We included 193 patients from the surgical ICU who were over 18 years old and excluded those without adequate intracranial pressure (ICP) monitoring data.
View Article and Find Full Text PDFBackground: Total hip, knee and shoulder arthroplasties (THKSA) are increasing due to expanding demands in ageing population. Material surveillance is important to prevent severe complications involving implantable medical devices (IMD) by taking appropriate preventive measures. Automating the analysis of patient and IMD features could benefit physicians and public health policies, allowing early issue detection and decision support.
View Article and Find Full Text PDFHow Trueness of Clinical Decision Support Systems Based on Machine Learning Is Assessed?
Stud Health Technol Inform
August 2024
The application of machine learning algorithms in clinical decision support systems (CDSS) holds great promise for advancing patient care, yet practical implementation faces significant evaluation challenges. Through a scoping review, we investigate the common definitions of ground truth to collect clinically relevant reference values, as well as the typical metrics and combinations employed for assessing trueness. Our analysis reveals that ground truth definition is mostly not in accordance with the standard ISO expectation and that used combination of metrics does not usually cover all aspects of CDSS trueness, particularly neglecting the negative class perspective.
View Article and Find Full Text PDFArticle Synopsis
- Electronic health data for implantable medical devices (IMD) allows real-time monitoring of risks, especially as joint surgeries like hip and knee replacements increase due to an aging population.
- A machine learning tool utilizing natural language processing (NLP) was created to automatically extract and analyze operation details from orthopedic medical reports, achieving excellent precision (97.0%) and recall (96.0%).
- By automating data extraction and monitoring of orthopedic devices through clinical data warehouses, the tool aims to enhance patient safety, support surgeons and policymakers with actionable insights, and improve compliance in medical reporting.
Secure Extraction of Personal Information from EHR by Federated Machine Learning.
Stud Health Technol Inform
August 2024
Article Synopsis
- Secure extraction of Personally Identifiable Information (PII) from Electronic Health Records (EHRs) is challenging due to privacy and security concerns, prompting a study on Federated Learning (FL) for French EHRs.
- The study used a multilingual BERT model and involved a simulation with 20 hospitals, comparing individual models (using only local data) and federated models (collaborative global model).
- Results showed that FL models maintain data confidentiality and achieved a competitive F1 score of 75.7%, highlighting FL's potential for improving health data analysis and privacy in EHRs.
Article Synopsis
- The paper presents a new method to improve access to clinical data warehouses (CDWs) for researchers and biomedical companies.
- It introduces a clinical data catalogue that answers key questions about data availability, quantity, and generation to aid project development.
- A prototype of the catalogue is demonstrated using visualization from the CDW of Rennes University Hospital.
The ONCO-FAIR project's initial experimentation aims to enhance data interoperability in oncology chemotherapy treatments, adhering to the FAIR principles. This study focuses on integrating the HL7 FHIR standard to address interoperability challenges within chemotherapy data exchange. Collaborating with healthcare institutions in Rennes, the research team assessed the limitations of current standards such as PN13, mCODE, and OSIRIS, leading to the customization of twelve FHIR resources complemented by two chemotherapy-specific extensions.
View Article and Find Full Text PDFArticle Synopsis
- * Researchers created machine learning models using a retrospective dataset to predict proper dosing based on anti-Xa results, with both random forest and XGB models achieving a mean AUROC of 0.80.
- * The study suggests that, after further validation, these machine learning models could be integrated into computerized physician order systems to assist doctors in making better dosing decisions.
The Ouest Data Hub (ODH) a project lead by GCS HUGO which is a cooperation group of University Hospitals in the French Grand Ouest region represents a groundbreaking initiative in this territory, advancing health data sharing and reuse to support research driven by real-world health data. Central to its structure are the Clinical Data Warehouses (CDWs) and Clinical Data Centers (CDCs), essential for analytics and as the linchpin of the ODH's status as an interregional Learning Health System. Aimed at fostering innovation and research, the ODH's collaborative and multi-institutional model effectively utilizes both local and shared resources.
View Article and Find Full Text PDFArticle Synopsis
- - The paper discusses a new visualization dashboard designed to display quality indicators for intensive care units (ICUs) utilizing the OMOP Common Data Model (CDM).
- - This dashboard allows users to visualize important quality data through various formats like histograms, pie charts, and tables.
- - Future plans include adding more quality indicators to the dashboard and assessing feedback from clinicians on its usefulness.
Article Synopsis
- - This study evaluates how well the OMOP common data model standardizes Continuous Renal Replacement Therapy (CRRT) data from ICU patients in two French hospitals.
- - Researchers successfully extracted and transformed data from 1,696 ICU stays into the OMOP format, aligning 46 standard concepts related to CRRT, despite facing issues with data variability.
- - The findings indicate that while the OMOP model shows promise for harmonizing ICU data, further improvements are needed to enhance clinical decision-making and patient outcomes in critical care.
Article Synopsis
- * It highlights real-world events, such as new equipment or changes in formulas, that can disrupt the consistency of lab result values over time, which necessitates monitoring for data quality.
- * The authors propose an automated dashboard using change point detection methods to track and visualize these disruptions, allowing for better understanding and explanation of changes in lab assays by biologists.
Key Research Areas (KRAs) were identified to establish a semantic interoperability framework for intensive medicine data in Europe. These include assessing common data model value, ensuring smooth data interoperability, supporting data standardization for efficient dataset use, and defining anonymization requirements to balance data protection and innovation.
View Article and Find Full Text PDFArticle Synopsis
- Electronic health records (EHRs) hold essential data for clinical research, but their sensitive nature requires effective de-identification methods to ensure privacy and adhere to regulations.
- The study introduces an automated de-identification pipeline utilizing a distant supervised method to lower costs and simplify the adaptation of this technology to various clinical settings.
- A French dataset was created for testing the pipeline, and a Bi-LSTM + CRF model achieved a high F1 score of 96.96%, indicating strong performance in identifying and removing personal identifiers from clinical documents.
Article Synopsis
- Researchers aimed to classify heart failure with preserved ejection fraction (HFpEF) patients into distinct phenotypes due to their varied prognoses and treatment needs.
- The study utilized machine learning to analyze clinical data and echocardiography results from a large hospital database, identifying four unique phenotypic clusters.
- The findings suggest that AI-driven phenotypes can help doctors better assess patient risk and tailor treatment plans to improve health outcomes.
In France and in other countries, we observed a significant growth in human polyvalent immunoglobulins (PvIg) usage. PvIg is manufactured from plasma collected from numeral donors, and its production is complex. Supply tensions have been observed for several years, and it is necessary to limit their consumption.
View Article and Find Full Text PDFArticle Synopsis
- The study focuses on creating an automated algorithm to quickly identify patients who might qualify for specific anti-cancer treatments, reducing the lengthy prescreening process for clinical trials.
- It analyzed 640 anonymized reports from multidisciplinary team meetings related to lung cancer, using regular expressions to extract relevant eligibility criteria, achieving impressive metrics: an average F1-score of 93%, 98% precision, and 92% recall.
- Despite these successes, there were significant inconsistencies in the completeness of patient and tumor information, with genetic mutations being particularly underreported and challenging to extract automatically.
Article Synopsis
- Disease surveillance systems are essential for public health officials, as they help design timely interventions to tackle disease outbreaks, but current systems in France lag by 1 to 3 weeks in reporting gastroenteritis activities.
- This study aimed to assess the feasibility of using internet search trends and electronic health records for near real-time predictions of acute gastroenteritis incidence in France.
- The findings indicate that combining different data sources can enhance gastroenteritis surveillance and allow for forecasting activity spikes up to 10 weeks in advance, potentially mitigating the disease's impact.
Article Synopsis
- Traditional dengue control relies on vector control and case reporting, which can be slow and often underestimates the disease burden, especially in smaller regions like Martinique.
- This study investigates the use of diverse data sources, including hospitalization records and Google Trends, to improve dengue tracking and reduce reporting delays on the island.
- Findings indicate that real-world data can enhance dengue surveillance, as certain indicators, like hospitalization rates, show strong correlations with dengue outbreaks, allowing for earlier detection of rising case numbers.
Article Synopsis
- Vital status after hospital discharge is crucial for biomedical data warehouses but often lacks reliable data; the French National Mortality Database (FNMD) provides a potential solution, although matching BDW records with FNMD presents various challenges like name changes and clerical errors.
- The study developed a deterministic algorithm using advanced data cleaning techniques and the Damerau-Levenshtein distance to improve the matching process and assessed its performance across records from three university hospitals in France.
- Results showed that the new algorithm had a significantly higher sensitivity (93.3%) compared to a simpler direct matching method (82.7%), especially for men and patients born in France, demonstrating its effectiveness in identifying deceased patients.
Article Synopsis
- * The primary goal is to standardize the process and transformations involved in feature extraction, including storage solutions within a data warehouse framework.
- * Results from interviews with researchers identified two key concepts, "track" and "feature," and proposed the creation of "TRACK" and "FEATURE" tables for better organization of extracted data.
Article Synopsis
- Direct oral anticoagulants and vitamin K antagonists are identified as potentially inappropriate medications (PIMs) in older adults, with specific drug-drug interactions (DDIs) that can increase the risk of harmful effects like bleeding.
- The study aimed to assess the occurrence of these PIMs, DDIs, and PIM-DDIs among elderly patients in both primary care and hospitals, as well as predict bleeding events related to them using machine learning techniques.
- Findings showed similar PIM prevalence in both care settings, but higher rates of DDIs and PIM-DDIs in hospitals; while they weren't major predictors of bleeding, their optimization by healthcare professionals is important for patient safety.
The Challenges of Implementing Comprehensive Clinical Data Warehouses in Hospitals.
Int J Environ Res Public Health
June 2022
Digital health, e-health, telemedicine-this abundance of terms illustrates the scientific and technical revolution at work, made possible by high-speed processing of health data, artificial intelligence (AI), and the profound upheavals currently taking place and yet to come in health systems [...
View Article and Find Full Text PDFArticle Synopsis
- Book music is a form of music representation used in street organs, made of thick cardboard with holes that indicate musical notes, which serves as inspiration for a new way to visualize clinical time-dependent data.
- In this proposed method, each row signifies a binary time-dependent variable, while each hole represents the actual value, allowing for organized data from various healthcare sources like demographics, lab results, and patient flow.
- This representation is particularly useful for survival analysis and facilitates the extraction of meaningful insights from healthcare data, improving the understanding of how time impacts medical outcomes.
Article Synopsis
- - Clinical image data analysis is evolving, requiring integration of imaging data into Clinical Data Warehouses (CDWs) while addressing challenges in interoperability and semantics, which led to the development of a web service called I4DW for querying pixel data.
- - The implementation of I4DW was evaluated using a prostate cancer cohort, demonstrating efficient DICOM data transfer with average retrieval times of around 5.94 seconds for series and 0.9 seconds for metadata, achieving high precision (0.95) and complete recall (1).
- - Future improvements for I4DW will focus on enhancing performance and ensuring patient data de-identification, while its design ensures scalability and can be applied across different clinical domains.