Introduction La maladie chronique du greffon contre l'hôte (cGVHD) est la complication la plus fréquente et la plus grave après une allogreffe de cellules souches hématopoïétiques (allo-HSCT).L’objectif était d’identifier les patients atteints de cGvHD, décrire leurs caractéristiques et les stratégies de traitements, via une méthode d’intelligence artificielle fondée sur le traitement automatique du langage naturel (IA–NLP) pour structurer les données issues des dossiers médicaux. Méthodes Une cohorte de patients atteints de cGVHD a été identifiée avec l'IA-NLP à tous les dossiers médicaux électroniques des patients, identification de ceux ayant un diagnostic de cGVHD et suivi au moins 12 mois après l'apparition de la cGVHD, puis extraction des variables d’intérêt.Un contrôle qualité en 2 étapes a été effectué :1. contrôle par un attaché de recherche clinique2. comparaison avec le registre de transplantation de moëlle osseuse de la SFGM-TC. Résultats Sur 513 patients, 235 ont développé une cGvHD : hommes 60%, âge moyen=49,6 ans. Les principales indications étaient : leucémie myéloïde aiguë (41,3%) et leucémie lymphoblastique aiguë (14,0%).Délai médian entre greffe et cGvHD: 174 jours.La cGVHD était sévère chez 103 patients, modérée chez 52 et légère chez 72, parmi les 227 patients ayant obtenu au moins un score.Les organes les plus souvent atteints étaient la peau (68,4%), les yeux (34,6%), le foie (33,3%), la bouche (29,8%), le tube digestif (25,9%) et les poumons (23,7%).84 patients ont reçu une deuxième ligne (ruxolitinib 42,9%, tacrolimus 13,1%, ciclosporine 7,1%, belumosudil 8,3%) et 35 une troisième ligne (ruxolitinib 37,1%, belumosudil 22,9%, ibrutinib 11,4%, tacrolimus 5,7%).La complétude des données était élevée pour les variables sociodémographiques et thérapeutiques (100%) et supérieure à 75% pour les caractéristiques de greffe. La concordance avec le registre EBMT dépassait 90% pour la majorité des variables communes. Discussion/Conclusion Cette étude utilise une méthode innovante d’extraction de données à l’aide de l’IA-NLP, et donne une première description multicentrique de la prise en charge de la cGVHD en France.

Introduction Les bases médico-administratives, telles que le SNDS, permettent d’identifier les soins précédant l’identification d’une maladie et de caractériser ainsi l’errance diagnostique. Cependant, les méthodes actuelles reposent encore sur des modèles statistiques (ex. régression logistique) basées sur la fréquence d’apparition des soins, sans prise en compte de leur ordre dans les parcours. Certains réseaux de neurones (récurrents, Transformers) offrent une alternative exploitant cette information séquentielle, mais leur adoption reste limitée par des interrogations sur leur gain prédictif réel, leur interprétabilité et leur temps d’entraînement nécessaire. L’objectif de cette étude était d’évaluer leur apport en capacité prédictive et de les confronter à ces limites. Méthodes Nous avons simulé des parcours de soins pour deux populations, 50 000 cas et 50 000 témoins, à partir de paramètres estimés séparément sur 22 000 patients atteints de sclérose latérale amyotrophique, identifiés par la délivrance de riluzole, et 22 000 témoins appariés sur le sexe et l’âge issus du SNDS. Les trajectoires simulées combinent des événements liés ou non à la pathologie d’intérêt. Nous avons évalué la capacité des modèles statistiques (régression logistique, SVM) et neuronaux (LSTM, Transformer) à discriminer ces deux populations à différents temps avant le diagnostic, en tenant compte ou non des délais entre événements. Les performances ont été mesurées par l’AUC et leur IC 95 %. Résultats Les approches statistiques ont obtenu un AUC moyen de 0,70 (IC95% : 0,68–0,72). Les réseaux LSTM et Transformers améliorent significativement l’AUC (+0.06), indiquant que l’ordre des évènements apporte une information discriminante. L’ajout des délais au modèle n’a pas apporté d’amélioration. Le temps d’entraînement reste modéré (≈15 minutes sur 100 000 séquences). Des méthodes d’interprétabilité (Integrated Gradients, SHAP) ont permis de retrouver les évènements qui discriminent les deux populations. Discussion/Conclusion Les modèles neuronaux apparaissent comme un compromis efficace entre performance, interprétabilité et coût computationnel. Ces résultats plaident pour une utilisation raisonnée de ces modèles dans les études SNDS qui visent à développer des modèles prédictifs.

Fibrodysplasia ossificans progressiva (FOP) is an ultra-rare genetic bone disorder associated with considerable clinical, economic, and societal impacts on patients, their caregivers, and the healthcare system. The aim of this study was to update the estimated prevalence of FOP in France, characterize the FOP patient population and impact of the disease, assess associated mortality rates, and quantify healthcare resource utilization and socioeconomic costs for individuals with FOP. An observational, retrospective case-control study was conducted using linked data for individuals with FOP from the French National Rare Diseases Registry and the French National Healthcare System Database (SNDS). Results were compared with the SNDS for a control group of individuals with any disease other than FOP. Prevalence of FOP was estimated at 1.39 per million. Data were available for 77 individuals with FOP and 769 controls without FOP. Of 70 individuals with FOP and a social security number at the index date, 50.0% (n = 35/70) were female, and the mean (SD) age was 25.3 (15.7) yr. Comorbidities were consistently more prevalent in individuals with FOP than in people without FOP. Consultations with general practitioners and non-physician healthcare providers, hospitalizations, and use of medical devices and treatments were significantly greater in individuals with vs without FOP (p < .05 for all). Mean total annual healthcare expenditure was more than 10 times greater in individuals with FOP than those without FOP from a societal perspective (largely driven by outpatient costs) and 14 times greater from a payer perspective. Overall survival was significantly worse in individuals with FOP than in individuals without (p < .0001). Findings from this study reinforce the clinical, economic, and societal impacts associated with FOP. Improving disease awareness and developing new interventions to support a reduction in these impacts should be considered key priorities for the FOP community and healthcare providers.

Background: Severe viral lower respiratory tract diseases (LRTD) are a major cause of hospitalisation and death, particularly among older adults and patients with chronic comorbidities. Their hospital burden, post-discharge outcomes, and associated costs remain poorly characterised in the post-pandemic period. Methods: The exhaustive French hospital discharge database (PMSI) was used to identify hospital stays involving severe viral LRTD, defined by ICD-10-coded diagnoses, between September 1, 2022 and August 31, 2024. Patients were followed for 90 days after discharge. Outcomes included hospitalisation incidence, in-hospital outcomes, readmissions, mortality, rehabilitation use, and direct hospital costs. Findings: Among 911,818 overnight hospital stays for all-cause LRTD, 345,050 (37.8%) met the definition of severe viral LRTD, corresponding to 324,449 unique patients. The hospitalisation incidence rate was 2.61 per 1,000 person-years in 2022/2023 and 2.21 in 2023/2024. Mean age was 56.0 (±35.1) years and 52.9% were males. Overall, 106,821 (31.0%) stays involved intensive care admission, 24,803 (7.2%) ended in in-hospital death, and mean length of stay was 9.8 days. The most frequently identified virus were SARS-CoV-2 (62.3%), RSV (18.0%) and influenza (12.2%). Among 317,553 patients with 90-day follow-up, 101,113 (31.8%) were readmitted for any cause, while 43,760 (13.8%) required rehabilitation care. Overall mortality during the index stay or within 90 days of discharge was 9.8%. Mean cost of the index hospitalisation was €6,785. Interpretation: Severe viral LRTD imposed a significant burden, with high mortality, frequent readmissions, substantial rehabilitation and costs beyond the initial hospitalisation, highlighting the need for strengthened prevention and management of high-risk patients.Funding: This study was sponsored by AstraZeneca S.A.S.

Objectif Évaluer l’impact potentiel de l’initiation du solriamfetol indiqué pour une somnolence résiduelle sous pression positive continue sur l’observance du traitement par PPC chez des adultes atteints de syndrome d’apnée obstructive du sommeil (SAHOS). Méthodes Cohorte rétrospective issue de la base de données du SNDS. Tous les patients ayant reçu au moins une administration de solriamfetol pour SAHOS entre le 01/09/2021 et le 30/06/2023 et ayant reçu au moins un remboursement de CPAP dans les 12 mois précédant le début de l’administration de solriamfetol ont été inclus. Un score d’observance à la PPC a été calculé à partir des niveaux de remboursement de ce traitement sur des durées de 4 semaines : • Adhésion ≥112heures : score=3 (plus de 4heures) • Adhésion ≥56heures et <112heures : score=2 (entre 2 et 4heures) • Observance <56heures : score=1 (moins de 2heures) • Patient n’étant plus équipé de PPC : score=0 Pour chaque patient, nous avons calculé la moyenne du score sur 6 séquences de 4 semaines précédant et suivant la date d’initiation du solriamfétol. Un groupe contrôle issu du SNDS traité par PPC mais non traité par solriamfetol a été apparié par score de propension aux sujets traités par solriamfetol. Résultats 930 patients ont été inclus et appariés à 8920 témoins (Fig. 1). Les caractéristiques des cas et des contrôles étaient très similaires (âge=56,4 et 57,1 ans, 64,5 % et 64,6 % d’hommes et proportions des comorbidités très similaires). Au cours de la période de 6 mois précédant et suivant le début du traitement par le solriamfetol, le score moyen d’adhésion à la CPAP était légèrement mais significativement plus élevé dans le groupe solriamfetol que dans le groupe contrôle (avant : 2,8 vs 2,6, p<0,0001, après : 2,8 vs 2,5, p<0,0001) (Fig. 1). Le score est resté stable ou a augmenté chez 79,9 % des patients traités par solriamfetol versus 75,4 % dans le groupe contrôle. La distribution du score avant et après initiation du solriamfetol est comparable entre les 2 groupes. Conclusion L’observance au traitement primaire du SAHOS par PPC se maintient sous solriamfetol, montrant un léger avantage par rapport aux témoins appariés, ce qui confirme son utilisation appropriée conformément aux conditions de remboursement.

OBJECTIVE:The analysis of care trajectories derived from electronic health records and claims data has become increasingly common in biomedical informatics. This has enabled large-scale studies of care processes, yet widely used binary code representations result in high-dimensional, sparse data that fail to capture semantic relationships between medical concepts. Learning dense vector representations (embeddings) has emerged as a promising approach to address these limitations. We aimed to construct and share joint embeddings for the International Classification of Diseases (ICD-10) and the Anatomical Therapeutic Chemical (ATC) classification system, providing reusable semantic representations of diagnoses and treatments from real-world claims data. MATERIALS AND METHODS:Using claims records from 1.5 million patients, we defined code co-occurrences within temporal windows and constructed a Positive Pointwise Mutual Information (PPMI) matrix spanning ICD-10 and ATC codes. Singular Value Decomposition (SVD) was applied to derive a low-dimensional embedding space. Evaluation combined UMAP visualization, nearest-neighbor retrieval, and a code-level classification task based on ICD chapters and ATC classes. RESULTS:The embeddings reflected the hierarchical organization of ICD-10 and ATC and revealed associations across coding systems, including clinically relevant diagnosis-treatment relationships. The classification task achieved mean AUCs of 0.93 for ICD-10 and 0.90 for ATC, indicating strong grouping of semantically related codes. DISCUSSION:The embeddings provide a reusable, code-level semantic representation that can support code retrieval, reduce manual code grouping, and be aggregated into patient-level features without training a task-specific model. CONCLUSION:We release the first openly available joint ICD-10-ATC embedding space derived from real-world claims data, providing a reusable resource for biomedical informatics research.

Sequences of healthcare events from claims data are increasingly used for predictive modeling. Despite the rise in popularity of neural networks, the benefit of modeling event order and timing remains underexplored. We simulated patient trajectories using parameters estimated from claims data including 22,000 amyotrophic lateral sclerosis (ALS) cases and 22,000 age and gender-matched controls. The simulation reproduced irregular event timing and condition-related activity that increased near an incident diagnosis. We compared three model families for population classification: frequency-based models, sequential models (long short-term memory and Transformer architectures) capturing event order, and sequential models incorporating temporal information. Performance was evaluated using the area under the ROC curve (AUC) at multiple time points before diagnosis. Sequential models outperformed frequency-based baselines by about +0.06 AUC on average, confirming the benefit of modeling event order. Adding time information provided no noticeable improvement, and results were stable across different time encoding and scaling methods. These findings suggest that, in claims-based classification tasks, event order captures most of the useful temporal signal, while explicit time encoding offers limited additional benefit under similar conditions.

Although the use of direct oral anticoagulants increases in parallel with the increase in atrial fibrillation (AF) with age, none of the bleeding risk scores (HAS-BLED, HEMORR2HAGES, ATRIA nor RE-LY) have been developed in a geriatric population. Our study aimed to develop a bleeding risk score adapted to this specific population and this therapeutic class. Multicentre, longitudinal, prospective, observational, pharmacoepidemiologic study conducted in 60 French cardiologic and geriatric centres included consecutive patients aged ≥80 years with AF, treated with rivaroxaban and followed for at least 1 year. All thromboembolic, bleeding and clinical events, including falls, hospitalisations or deaths, were recorded every 3 months for 1 year. A predictive risk score based on the clinical variables most associated with bleeding events was developed from the total sample, randomly divided into a training sample and a validation sample. Among the 839 patients included (mean age = 86 year old, 62% women), there were 78 (9.3%) major haemorrhagic events. Variables associated with bleeding were age, anaemia, low albuminemia, amiodarone use and low creatinine clearance estimated with Cockcroft formula, grouped together as the A4C score. The A4C score better identifies bleeding risk in subjects ≥80 years than the scores already validated in younger populations, as its area under the curve in the training sample and in the validation sample was 0.73 and 0.66, respectively, whereas it was 0.49/0.55 for the HAS-BLED score, 0.53/0.50 for the HEMORR2HAGES score, 0.58/0.61 for the ATRIA score and 0.57/0.54 for the RE-LY score. A threshold of the A4C score ≥ 2 identifies subjects ≥80 years at high risk of bleeding.

Background Oral glucocorticoids (OCS) remain one of the most important treatments for SLE but are associated with damage. Evidence regarding the real-world use of OCS in nationwide SLE populations is currently lacking. The aim of this study was to analyse OCS use and SLE treatments in French patients with SLE at the national level.Methods The nationwide French health insurance claims database, which contains pseudonymised data for ≈66 million people, was used. Prevalent patients with SLE (International Classification of Diseases, 10th Revision code M32, recorded as a chronic condition or associated with hospital stay) were identified over the year 2019. SLE treatments were captured through actual drug deliveries by pharmacies and mean daily OCS doses (prednisone equivalent) were calculated for the year 2019.Results The 2019 French prevalent SLE population comprised 31 852 patients (86.3% of women, with a mean age of 49.7 (±15.9) years) with a mean disease duration of 7.1 (±6.2) years. Among these, 48.3% were treated with OCS. The mean daily OCS dose was ≤5 mg/day in 35.9%, more than 5 mg but <7.5 mg/day in 6.4% and ≥7.5 mg/day in 6.0%. The use of other SLE treatments was significantly increased in patients with higher doses of OCS (p<0.0001). Potential complications of OCS, including cardiovascular diseases, infections and osteoporosis, were significantly increased in patients with SLE receiving more than 5 mg of OCS per day (p<0.0001, for all). Strikingly, 13.6% of patients receiving mean daily OCS doses >5 mg/day were not treated with antimalarial, immunosuppressant or biologic drugs for SLE.Conclusions In total, 48.2% of French patients with SLE were treated with OCS in 2019, including 12.4% at a mean dose >5 mg/day, with an increased risk of OCS complications and a limited use of antimalarials, immunosuppressants or biologics. These results highlight the urgent need for the implementation of more robust OCS-sparing strategies in SLE.

INTRODUCTION:There are few data on healthcare resource use and related costs of French haemophilia A (HA) and B (HB) patients. AIMS:This study aimed to describe the profile of HA and HB patients, current disease management, clinical burden and costs. METHODS:Data related to haemophilia patients of all ages alive on 1/1/2022 were extracted from the nationwide French claims database (SNDS). Patients were divided into four treatment groups: on-demand or prophylaxis with or without inhibitors. Haemophilia patients were compared with a control group (ratio 1:3) matched for age, gender and region using risk ratios (RR [95% confidence interval]). The annual direct health care costs per person were estimated. RESULTS:A total of 5,577 (HA) and 1,332 (HB) patients were included (mean age: 36.4 years). Most patients were treated on-demand (HA: 72.8%; HB: 76.6%) and a few had inhibitors (HA: 3.6%; HB: 1.1%). Overall, haemophilia clinical burden was significantly higher than among controls, in particular, mortality (RR:1.42 [1.04-1.92]), work disability (RR: 2.71 [2.22-3.30]), hospitalisation for major bleeding (RR:12.06 [8.67-16.80]), orthopaedic surgery (RR: 2.97 [2.65-3.32]) and hospitalisation all causes (RR: 2.44 [2.31-2.58]). This burden was more important in patients with inhibitors or treated in prophylaxis and was close for HA and HB patients. The annual per-person costs were €282,560 and €181,566 for HA and HB in prophylaxis without inhibitors, respectively. The population with inhibitors, although limited, had even much higher costs. CONCLUSION:The clinical burden and costs of haemophilia treatments may be very high especially in patients in prophylaxis and/or with inhibitors.

Background The daily dose of oral glucocorticoids (OCS) is associated with damage in Systemic Lupus Erythematosus (SLE), and OCS reduction is a major goal of SLE care. However, evidence regarding the real-world use of OCS in nation-wide populations are currently lacking. Objectives The aim of this study was to analyze the use of OCS in French patients with SLE, at the national level, using medico-administrative data. Methods This study used the French health-insurance claims database (SNDS), which contains pseudonymized individual data for over 66 million people. SLE patients were identified with the ICD-10 diagnosis code for SLE (M32), documented as a chronic condition or associated to hospital stay. Included SLE patients were defined as patients alive on January 1st 2020 whenever the SLE diagnosis occurred. SLE comorbidities and OCS complications were identified through validated algorithms. Real-world use of treatments was identified through drug deliveries in pharmacies and daily OCS doses (expressed in prednisone-equivalent) were calculated for the year 2019. Comparisons were made to age- and gender-matched controls without SLE from the general population. Results A total of 32,173 patients with SLE (mean age 49.9 ± 16.0 years; 86.1% women) were alive on January 1st 2020, with a mean disease duration of 7.1 ± 6.2 years.Among these prevalent SLE patients, 48.2% were treated with OCS. The mean daily dose of OCS was below 5 mg/day in 35.7%, between 5 and 7.5 mg/day in 6.4% and above 7.5 mg/day in 6.1%. OCS use was significantly more frequent in women, in patients with CMUc (specific health coverage for low income patients) and decreased with age (p<0.0001, for all). SLE-specific comorbidities, including glomerular disease, skin involvement, polyarthritis, pleurisy, pericarditis, and thrombocytopenia, were significantly increased in patients with higher doses of OCS (p<0.0001, for all). Use of SLE treatments other than OCS was significantly increased in patients with higher doses of OCS (p<0.0001, for all) (Table 1). Strikingly, 14.6% of patients receiving 5 to 7.5 mg of OCS per day and 14.2% of those receiving more than 7.5 mg per day were not treated with antimalarial drugs, immunosuppressives or other biologic treatments for SLE. Complications of OCS, including cardiovascular diseases, infections, osteoporosis, and obesity, were significantly increased in SLE patients receiving ≥ 5 mg per day of OCS (p<0.0001 for all). The overall annual mean cost of healthcare consumptions from a societal perspective in 2019 was 6,048€ for prevalent SLE patients and 2,601€ for matched controls (p<0.0001). Among prevalent SLE patients, the cost increased significantly according to the OCS daily dose: 4,633€ for patients without OCS, 6,383€ (daily dose of 0-5 mg/day), 9,815€ (5-7.5 mg/day) and 13,861€ (above 7.5 mg/day). Conclusion To the best of our knowledge, this is the first nation-wide study reporting on real-life use of OCS in patients with SLE. The proportion of patients treated with high-dose OCS ≥ 7.5mg/day remains unacceptably high and associated with increased comorbidities, OCS complications and significantly increased healthcare costs. Also, over 14% of patients receiving OCS doses ≥ 5 mg/day were not treated with antimalarial drugs, immunosuppressives or other biologic treatments for SLE. These results highlight the need for tight disease control and implementation of robust OCS-sparing strategies in SLE. REFERENCES: NIL. Acknowledgements: NIL. Disclosure of Interests Laurent Arnaud Consultant of: Abbvie, Alexion, Alpine, Amgen, AstraZeneca, Biogen, BMS, Boehringer-Ingelheim, GSK, Grifols, Janssen, LFB, Lilly, Kezar, Medac, Novartis, Oséus, Pfizer, Roche-Chugaï, Sêmeia, UCB, Caroline Fabry-Vendrand Employee of: AstraZeneca, Remus Todea Employee of: AstraZeneca, Blandine VIDAL Employee of: AstraZeneca, Cottin Juliette Grant/research support from: I am employed by the company CEMKA who receive grants by pharmaceutical companies to conduct epidemiological and economic studies, Julien Robert Grant/research support from: I am employed by the company CEMKA who receive grants by pharmaceutical companies to conduct epidemiological and economic studies, Stéphane Bouée Grant/research support from: I am employed by the company CEMKA who receive grants by pharmaceutical companies to conduct epidemiological and economic studies, Gabriel Thabut Employee of: AstraZeneca.Table 1Proportion of patients treated with SLE treatments other than corticosteroids in 2019No OCSOCS0-5 mg/dayOCS5-7.5 mg/dayOCS≥ 7.5 mg/dayp-valueAntimalarials(Hydroxychloroquine/ Chloroquine)8,826 (53.0%)7,286 (63.5%)1,505 (72.9%)1,416 (71.9%)<0.0001Methotrexate958 (5.7%)1,405 (12.2%)361 (17.5%)358 (18.2%)<0.0001Mycophenolate mofetil337 (2.0%)878 (7.6%)350 (17.0%)496 (25.2%)<0.0001Azathioprine264 (1.6%)549 (4.8%)175 (8.5%)206 (10.5%)<0.0001Cyclophosphamide207 (1.2%)367 (3.2%)147 (7.1%)266 (13.5%)<0.0001

OBJECTIVES:Currently, two classes of oral anticoagulants are available in nursing home residents: vitamin K antagonists (VKA) and direct oral anticoagulants (DOAC). DOACs have a higher net clinical benefit than VKAs but DOACs are about 10 times more expensive than VKAs. The objective of our study was to assess and compare the overall costs of anti-coagulant strategy (VKA or DOAC), i.e., including drugs, laboratory costs and time spent in human capital (nurses and medical time) in nursing homes in France. METHODS:This was an observational, multicenter, prospective study including nine nursing homes in France. Among these nursing homes, 241 patients aged 75 years and older and treated with VKA (n = 140) or DOAC (n = 101) therapy accepted to participate in the study. RESULTS:During the 3-month follow-up period, the adjusted mean costs per patient were higher for VKA than DOACs for nurse care (€327 (57) vs. €154 (56), p<.0001) for general practitioner care (€297 (91) vs. €204 (91), p = 0.02), for coordinating physicians care (€13 (7) vs. €5 (7), p < 0.07), for laboratory tests (€23 (5) vs. €5 (5), p<.0001), but were lower for drug costs (€8 (3) vs. €165 (3), p<.0001). The average overall cost for 3 months per patient was €668 (140) with VKA vs. €533 (139) with DOAC (p = 0.02). CONCLUSION:Our study showed that in nursing homes despite a higher drug cost, DOAC therapy is associated with a lower total cost and less time used by nurses and physicians for drug monitoring when compared to VKA.