- Research Article
- 10.1016/j.jval.2025.09.1137
EPH8 A Retrospective Study of Treatment Use in Metastatic Colorectal Cancer (mCRC) Based on French Hospital Discharge Database
- Dec 01, 2025
- Value in Health
- Elise Arnee + 3 more +3
Publications from 2021 to 2026
Showing 10 of 29 papers
EPH8 A Retrospective Study of Treatment Use in Metastatic Colorectal Cancer (mCRC) Based on French Hospital Discharge Database
EPH86 Estimating the Age- and Sex-Standardized Prevalence and Incidence of Depression and Anxiety in UK Adults Using the THIN Database
Vision-Language Models for Sick Leave Certificates: Beyond OCR in Real-World Form Understanding
The processing of sick leave certificates has become a critical challenge for health systems, with increasing administrative costs and more than 10 million documents handled annually in France. Although these certificates follow fixed templates, they are usually received as smartphone photos or noisy scans, often including handwritten notes and signatures that hinder reliable OCR-based extraction. While recent advances in document understanding suggest that foundation models can overcome these limitations, their comparative benefits for domain-specific, large-scale administrative tasks remain unclear. OCR-LLM pipelines offer interpretability but propagate recognition errors. Vision-Language Models (VLMs) promise end-to-end robustness, but they raise questions about accuracy, efficiency, and scalability. We evaluated these two approaches using a dataset of 183 real-world French sick leave certificates, encompassing typed, handwritten, scanned, and photographed documents. Results show that VLMs, particularly Qwen2.5 models, remarkably outperform OCR-LLM pipelines, achieving higher performance (up to 89.6% on typed scans), faster inference times (3s vs. 15s), and superior handling of complex fields such as employer details, social security numbers, and signatures. Theoretically, this study validates the advantage of joint visual-textual reasoning over sequential pipelines in multimodal document processing. It highlights the trade-offs between robustness, accuracy, and efficiency, offering new insights into the deployment of foundation models for administrative automation in healthcare.
Read moreFeasibility, user satisfaction, and knowledge improvement after a VR training program for healthcare professionals managing behavioral and psychological symptoms of dementia (BPSD): Protocol for the FORMSPC-REALVI single-arm pre-post study
BackgroundBehavioral and psychological symptoms are a common challenge for healthcare professionals when managing patients with dementia, and effective verbal and nonverbal communication skills are crucial in caring for such patients.ObjectivesThis article describes a research study protocol for investigating the effectiveness of a virtual reality (VR) training program for healthcare professionals in managing disruptive behavioral and psychological symptoms of dementia (BPSD), such as aggressiveness, agitation, and care refusal.MethodsThe training scenarios were co-designed with ten healthcare professionals and implemented using an immersive 3D VR platform. Forty geriatric healthcare professionals will participate in a 2-hour training session using VR movies and a Moodle-based theoretical reinforcement. Before and after the training, participants will complete self-assessment questionnaires and knowledge-based quizzes designed to evaluate their perceived competence and understanding of appropriate communication strategies with patients displaying BPSD. The primary outcome will be the change in quiz scores between the pre- and post-training evaluations. Secondary outcomes include training satisfaction, perceived competence, and system usability.HypothesesTrial registrationThis study is registered in the French General Data Protection Regulation (GDPR) registry of Assistance Publique – Hôpitaux de Paris (N° 2022 0518135339–18 May 2022). As the trial targets health-providers and measures effects only on them (and not on providers’ patients), clinical trial registration is not required (see ICMJE guidelines: https://www.icmje.org/about-icmje/faqs/clinical-trials-registration/).
Read moreCharacteristics of Patients Initiated on Budesonide/Glycopyrronium Bromide/Formoterol Fumarate Single Inhaler Triple Therapy for the Treatment of Chronic Obstructive Pulmonary Disease: A Population-Based Observational Study.
Budesonide/glycopyrronium bromide/formoterol fumarate (BUD/GLY/FOR) single inhaler triple therapy became available to prescribe to patients with severe chronic obstructive pulmonary disease (COPD) in France in 2021. The characteristics of patients prescribed BUD/GLY/FOR triple therapy and guideline adherence have not been previously described in France. To describe the characteristics of COPD patients initiated on BUD/GLY/FOR triple therapy, assess adherence to COPD management guidelines, and explore any differences by prescribing physician. A cross-sectional study using data from The Health Improvement Network (THIN®) France database was conducted. Patients with ≥2 recorded diagnostic codes for COPD were included. Demographic characteristics, comorbidities, management, COPD-related characteristics, and guideline adherence (Société de Pneumologie de Langue Française (SPLF); Haute Autorité de Santé (HAS)), stratified by initiating physician speciality (general practitioner (GP) or pulmonologist) were described. A total of 263 patients initiating BUD/GLY/FOR triple therapy were included. Mean (SD) age was 68.8 (11.8) years; 53.6% were male. Mean (SD) COPD duration was 6.4 (5.5) years. Comorbidities were common, with slightly more cardiometabolic and mental health conditions recorded in the GP-initiated group, and more comorbid respiratory conditions recorded in the pulmonologist-initiated group. About 77.2% (n=203) of patients had at least one moderate or severe exacerbation in the 12 months before initiation. About 86.3% had a previous record of dual (n=117, 44.5%) or triple (n=110, 41.8%) therapy. About 68.8% had been initiated on BUD/GLY/FOR triple therapy in line with SPLF guidelines (62.4% and 72.4% in the GP- and pulmonologist-initiated groups, respectively); among those with a record of COPD severity, 75.2% were initiated in line with HAS guidelines (69.2% and 76.3% in the GP- and pulmonologist-initiated groups, respectively). The majority of COPD patients are prescribed BUD/GLY/FOR triple therapy in accordance with current treatment guidelines, irrespective of whether the therapy is prescribed by a general practitioner or a pulmonologist.
Read moreEPH69 Analysis of Influenza Vaccination Coverage and Vaccine Effectiveness in Population Aged 65 and Over in Ambulatory Medicine in France From 2016 to 2019: A Study Based on Real-World Data
EPH133 AI Using Real-Life Data, an Additional Diagnostic Tool for Doctors to Identify Patient at Very High Risk of Renal Failure
IEcons: A New Consensus Approach Using Multi-Text Representations for Clustering Task
Today we are able to generate a large set of text representations from the simple Bag-of-word (BOW) to the recent transformers capturing the semantic and the contextual text meaning. It was proven that there is no best text representation for text clustering task. Consequently, some works combined text representations using a consensus clustering approach. Two consensus approach types exist, namely explicit and implicit consensus. In the explicit consensus, also known asensemble clustering, the consensus function is applied a posterior after obtaining cluster labels from each text representation clustering allowing to capture global mutual information between the partitions of all text representations. On the other hand, implicit consensus uses tensor clustering to optimize the clustering consensus partition that deals with similarity matrices of text representations.
Read moreAntibiotic prescriptions associated with a diagnosis of acute nasopharyngitis by general GPs in France: a retrospective study
BackgroundNasopharyngitis is a common viral infection that has led to an overuse of prescription drugs, in particular antibiotics, which are not indicated for this condition.AimThe purpose of this study was to describe drug prescriptions for patients with a diagnosis of acute rhinopharyngitis in general practices in France.Design & settingRetrospective study of 1 067 403 prescriptions for a diagnosis of nasopharyngitis issued by 2637 physicians to 754 476 patients living in metropolitan France.MethodThe data were sourced from the prescription software, Cegedim, for the period 1 January 2018 to 31 December 2021 and analysed according to patients’ and physicians’ ages.ResultsA total of 2 591 584 medications were prescribed by GPs, with a median of three medications per patient. A total of 171 540 courses of antibiotics were prescribed (16% prescription rates), with amoxicillin being the most frequently prescribed (102 089 prescriptions; 59.5% of antibiotic prescriptions). Amoxicillin prescription increased in extreme age groups (18.2% of visits in those aged 9 years and under, and 10.0% of visits in those aged over 80 years, while patients aged 20–29-years were prescribed amoxicillin in just 2.9% of visits), and more prescriptions are issued by older doctors (GPs older than 70 years prescribed antibiotics in 26.4% of visits versus 3.2% of visits by GPs aged under 29 years).ConclusionNasopharyngitis is frequently a cause of therapeutic over-prescriptions including antibiotics, with an antibiotic prescription rate of 16%. Additional research is required to enhance our understanding of factors linked to drug prescriptions.
Read moreHierarchical Tensor Clustering for Multiple Graphs Representation
Graph clustering is a challenging task, especially when there is a hierarchical structure. The availability of multiple graphs (or relational graphs), in the multi-graph setting, provides additional information that can be leveraged to improve clustering results. This paper aims to develop a new hierarchical clustering algorithm for multi-graphs, the HTGM algorithm. This algorithm represents the set of graphs in the multi-graph as a 3-way tensor, and maximizes a modularity measure, extending the modularity-based graph clustering algorithm to multi-graphs and tensor structures. We evaluate the proposed algorithm over synthetic and real-world datasets and show the effectiveness of the proposed algorithm by benchmarking it to alternative clustering algorithms.
Read more