Diagnostic agreement between prehospital and in-hospital diagnoses for dyspnea: a retrospective study in a French prehospital emergency medical service Authors: Malik El-Boudali, Prabakar Vaittinada Ayar, Geoffroy Rousseau, Dounia Hamdouche, Romain Jouffroy, Vincent Garrouste Published: 2026 ABSTRACT BackgroundDyspnea is a common, high-acuity presentation in emergency medicine and is associated with substantial short-term mortality. Early etiological diagnosis remains challenging in the prehospital setting, particularly in physician-staffed mobile intensive care unit (MICU) systems.MethodsWe conducted a retrospective, single-center study of adults managed by a physician-staffed MICU for dyspnea in Orléans, France, between 2023 and 2024. Agreement between the prehospital MICU working diagnosis and the first in-hospital coded diagnosis was assessed among patients with paired diagnostic data using percent agreement and Cohen’s kappa with bootstrap 95% confidence intervals. Prehospital management was retrospectively assessed against key principles of acute dyspnea management. Modified Poisson regression with robust variance was used to explore 28-day mortality.ResultsAmong 318 patients, 313 had paired diagnostic data. Overall agreement was 83.4% (261/313) with a Cohen’s kappa of 0.79 (95% bootstrap CI 0.74–0.84). The most frequent prehospital diagnostic categories were acute heart failure, acute COPD exacerbation, pneumonia, and asthma exacerbation. Among patients with paired diagnoses, 45 died within 28 days (14.4%). Mortality was higher in discordant than concordant cases (23.1% [12/52] vs. 12.6% [33/261]; crude RR: 1.83, 95% CI 1.01–3.29). In adjusted complete cases analysis (n = 269; 41 deaths), the association was attenuated aRR: 1.59 (95% CI 0.83–3.05). Prehospital management was judged consistent in 75.0% of discordant cases.ConclusionsIn this selected MICU population, agreement between prehospital working diagnoses and first in-hospital coded diagnoses was substantial. Most discordant cases still received consistent prehospital management. The association between discordance and 28-day mortality was attenuated after adjustment.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12873-026-01657-x. Background Dyspnea is one of the most common and potentially life-threatening symptoms encountered in prehospital and in-hospital emergency medicine [1, 2]. Dyspnea, a subjective sensation of uncomfortable or labored breathing, accounts for a substantial proportion of emergency medical service (EMS) activation and emergency department (ED) admissions worldwide [3]. In France, dyspnea represents up to 8% of the approximately 30 million annual emergency medical communication center (EMCC) calls [4, 5] and nearly 4% of prehospital mobile intensive care unit (MICU) interventions [6, 7]. Beyond its frequency, dyspnea is associated with significant morbidity, ED overcrowding, and high short-term mortality. In-hospital mortality among patients presenting with dyspnea ranges from 5% to 15%, and 30-day mortality is close to 10% [4, 8, 9]. Despite its clinical importance, dyspnea remains a daily diagnostic challenge in both prehospital and in-hospital settings. It is a highly sensitive but poorly specific symptom shared by a wide range of cardiac, pulmonary, neurological, metabolic, and psychiatric conditions [10–13]. Overlapping clinical presentations, time constraints, and limited access to diagnostic tools, especially in the out-of-hospital setting, complicate early etiological identification and may result in a delay in the administration of appropriate treatment or triage in time-sensitive conditions such as acute heart failure (AHF) or pulmonary embolism (PE). Several observational studies have reported only moderate agreement between prehospital and in-hospital diagnoses in dyspneic patients, with Cohen’s kappa values generally ranging from 0.4 to 0.6 [14–16]. More recent multicenter studies confirm that diagnostic concordance varies widely depending on the underlying etiology, with higher agreement for common cardiorespiratory conditions and lower performance for less frequent or atypical causes [17, 1, 18]. Notably, a recent prospective multicenter study by Veldhuis et al. highlighted the persistence of diagnostic discordance between prehospital and ED physicians despite advances in prehospital care and organization [19]. However, less is known about such diagnostic agreement in physician-staffed MICU systems and about whether diagnostic discordance is accompanied by inconsistent prehospital management or worse short-term outcomes. From an etiological perspective, large European cohorts reported that pulmonary infections, AHF, acute chronic obstructive pulmonary disease (COPD) and asthma were the leading causes of dyspnea in the ED, albeit with seasonal variations [9, 20]. Most previous studies reflect paramedic-led EMS systems. By contrast, the French SAMU/SMUR model relies on EMCC-based medical dispatch and physician-staffed MICUs, with an emphasis on on-scene medical assessment and early stabilization, and triage. This organization may influence both diagnostic reasoning and prehospital management. It is essential to acknowledge that diagnostic concordance should not be regarded exclusively as an indicator of physician performance. In the domain of prehospital emergency medicine, clinical decision-making frequently employs a syndromic approach, prioritizing the stabilization of vital functions over the establishment of a definitive etiological diagnosis. Diagnostic discordance may therefore reflect mixed, overlapping, or evolving presentations rather than inappropriate initial assessment. The extent to which diagnostic discordance adversely affects patient outcomes, as opposed to merely reflecting greater clinical complexity, remains insufficiently explored. Therefore, the objectives of this study were (i) to assess diagnostic agreement between the prehospital MICU diagnosis and the first in-hospital coded diagnosis among patients managed by a MICU for dyspnea, (ii) to perform a post hoc structured assessment of whether prehospital management was consistent with key principles of acute dyspnea management according to diagnostic agreement status, and (iii) to explore the association between diagnostic discordance and 28-day mortality. Methods Study design and setting This retrospective, single-center observational study used routinely collected prehospital and in-hospital clinical data and is reported in accordance with the STROBE and RECORD statements [21]. This study was conducted between January 1st, 2023, and December 31st, 2024, within the EMCC/MICU system of Orléans University Hospital, France. In the French emergency medical system, the EMCC provides medical dispatch and may send a physician-staffed mobile intensive care unit (MICU) for patients requiring on-scene medical assessment, stabilization, and triage. The study included adult patients managed by the MICU for dyspnea as the primary dispatch complaint who were subsequently admitted to Orléans University Hospital, including the emergency department (ED), intensive care unit (ICU), coronary care unit (CCU), or hospital wards. Inclusion and exclusion criteria Among adults, eligible cases were identified from the prehospital MICU activity records during the study period. Records were screened for dyspnea or respiratory distress as the primary dispatch complaint, and eligibility was confirmed by manual review of the MICU report to ensure that dyspnea was the main presenting symptom during prehospital assessment. Patients were excluded if they were not hospitalized at Orléans University Hospital, died before MICU arrival or during transport, or if no prehospital diagnosis was documented. Data collection Demographic data included age and medical history, with particular attention to cardiovascular risk factors, chronic respiratory diseases, chronic heart failure, chronic kidney disease, and active malignancy. Prehospital vital signs, including pulse oximetry (SpO2), heart rate, respiratory rate, non-invasive blood pressure, and core body temperature, were recorded upon arrival of first responders and/or the prehospital MICU team. Prehospital management data included oxygen supplementation, ventilatory support modalities, and relevant clinical examination findings. Adjustments to oxygen therapy and escalation of respiratory support during the prehospital phase were systematically recorded. Investigators had access to the full set of eligible MICU reports during the study period within Orléans University Hospital. Access to in-hospital coded diagnoses was obtained through the institutional coding software (EASILY®) for patients admitted to Orléans University Hospital. Person-level linkage between prehospital MICU reports and in-hospital coded diagnoses was performed within the institution. Extracted variables underwent basic consistency and plausibility checks; when discrepancies were identified, source documents were reviewed and unresolved values were coded as missing. Diagnostic definitions and classification The prehospital MICU diagnosis corresponded to the working diagnosis documented by the MICU physician at the end of the prehospital intervention. The in-hospital diagnosis corresponded to the first coded diagnosis recorded during initial hospital management and extracted from EASILY®. For the present analysis, recorded diagnoses were grouped into predefined analytical categories before statistical analysis, using the same classification framework for prehospital and in-hospital diagnoses. When several diagnoses were documented, the primary analytical diagnosis was assigned according to the condition considered to be the main driver of acute dyspnea; if no predominant mechanism could be identified, the case was classified as mixed/other. The mapping between recorded diagnostic labels and analytical categories is provided in Supplementary Table S2. Post hoc structured assessment of prehospital management Prehospital management was assessed retrospectively using the 2025 French guidance on acute dyspnea as a structured interpretive framework rather than as a strict audit of contemporaneous compliance [22]. Two independent emergency physicians reviewed each case based on the MICU report and the first in-hospital coded diagnosis, blinded to the identity of the prehospital physician and to outcomes. Each case was classified as having prehospital management judged consistent with key principles of acute dyspnea management only when both reviewers independently agreed. For readability, this classification is referred to below as “consistent management”. Outcomes The primary outcome was diagnostic agreement between the prehospital MICU diagnosis and the first in-hospital coded diagnosis, assessed using overall percent agreement and Cohen’s kappa. Secondary outcomes were: (1) the etiological distribution of dyspnea among patients managed by the MICU; (2) patient characteristics and prehospital management according to concordance status; (3) the proportion of cases with prehospital management judged consistent, overall and according to concordance status; and (4) an exploratory analysis of the association between diagnostic discordance and 28-day mortality. Statistical analysis Quantitative variables were presented as means (SD) or medians (IQR 25th–75th) as appropriate; categorical variables were presented as n (%). Diagnostic agreement was assessed among patients with both prehospital and in-hospital diagnoses available. Because prehospital and in-hospital diagnoses were paired within the same patient, agreement was described using paired cross-tabulation, overall percent agreement and unweighted Cohen’s kappa. 95% confidence intervals for kappa were estimated using percentile bootstrap with 5,000 replicates [23]. Baseline comparisons between concordant and discordant groups were exploratory and unadjusted. These comparisons were performed at the patient level, with concordant and discordant cases considered as mutually exclusive groups; corresponding p values are reported descriptively and should not be overinterpreted. Concordant and discordant cases were compared using χ² or Fisher’s exact tests for categorical variables and Student’s t-test or Wilcoxon rank-sum tests for continuous variables, as appropriate. The proportion of cases with prehospital management judged consistent with key principles of acute dyspnea management was compared between groups using the same approach. To estimate crude and adjusted risk ratios for 28-day mortality, we used a modified Poisson regression model with robust variance estimation [24], with diagnostic discordance as the exposure and 28-day mortality as the outcome. This approach was chosen to estimate relative risks rather than odds ratios, because relative risks were considered more clinically interpretable. The adjusted model included prespecified markers of baseline severity available during prehospital assessment: age, initial SpO2, systolic blood pressure, altered mental status, long-term oxygen therapy, and signs of respiratory struggle. No imputation was performed. Crude analyses used all available data among patients with both diagnoses available. The adjusted modified Poisson model was conducted as a complete-case analysis. Covariates for adjustment were prespecified before model fitting, based on clinical relevance and availability at the time of prehospital assessment, rather than selected according to univariable p values. Because complete time-to-follow-up information was not available for survivors, no time-to-event model was fitted. Analyses were performed using R (version 4.5.1) [25]. Use of AI-assisted language tools An AI-assisted language tool was used for language editing (clarity and grammar) only. All scientific content, analyses, and interpretation were produced and verified by the authors. Ethical considerations This retrospective study used routinely collected health data and did not involve any additional intervention. According to French regulations, the study was classified as research not involving human participants (“recherche n’impliquant pas la personne humaine”) and was conducted in compliance with the French MR-004 reference methodology for research using existing healthcare data. The Research Ethics Committee of Orléans University Hospital (Comité d’Ethique de la Recherche du CHU d’Orléans) issued a favorable opinion for this study on April 3, 2026 (CERO 2602-09). Individual informed consent was not required for this retrospective research on existing data. In accordance with French regulations and the institutional information procedure, patients were informed of the secondary use of their data and could object to such use. Results Participant selection Between January 1st, 2023, and December 31st, 2024, a total of 344 patients were initially screened. After applying exclusion criteria, 318 (90%) patients were included in the study population (Fig. 1). Among them, 313 patients had both a prehospital MICU diagnosis and a first in-hospital coded diagnosis available and constituted the diagnostic agreement population. Twenty eight day mortality was available for all patients in this population. The adjusted mortality analysis was restricted to 269 complete cases. Fig. 1Study flowchart. The flowchart shows patient selection from screened MICU interventions to the final study population. Diagnostic agreement analyses were restricted to patients with both a prehospital MICU diagnosis and a first in-hospital coded diagnosis available. The diagnostic categories displayed in the final row correspond to the initial prehospital MICU working diagnosis. Legend: MICU= Mobile Intensive Care Unit; AHF= Acute Heart Failure; COPD= Chronic Obstructive Pulmonary Disease Study flowchart. The flowchart shows patient selection from screened MICU interventions to the final study population. Diagnostic agreement analyses were restricted to patients with both a prehospital MICU diagnosis and a first in-hospital coded diagnosis available. The diagnostic categories displayed in the final row correspond to the initial prehospital MICU working diagnosis. Legend: MICU= Mobile Intensive Care Unit; AHF= Acute Heart Failure; COPD= Chronic Obstructive Pulmonary Disease Baseline characteristics The study population (n = 318) had a median age of 73 years [IQR: 63–80] and 139 patients (43.7%) were female (Table 1). The most common cardiovascular risk factors were hypertension (68.2%, n = 217), dyslipidemia (44.0%, n = 140), and active smoking (17.3%, n = 55). COPD (30.8%, n = 98) and chronic heart failure with reduced ejection fraction (19.2%, n = 61) were the most frequent comorbidities. Chronic kidney disease was present in 12.0% (n = 38). Additionally, 11.6% (n = 37) of patients required long-term oxygen therapy. Table 1Baseline characteristics of the study populationOverall paired-diagnosis population (n = 313)Concordant diagnostic(n = 261)Discordant diagnostic(n = 52)P valuesAge —Years72.0 [62.0-80.0]72.0 [62.0–80.0]73.0 [66.0–80.0]0.43Gender — Female139 (44.4%)117 (44.8%)22 (42.3%)0.86Cardiovascular risk factors Age > 50 274 (87.5%)225 (86.2%)49 (94.2%)0.17 Dyslipidemia 137 (43.8%)121 (46.4%)16 (30.8%)0.05 Hypertension 214 (68.4%)180 (69.0%)34 (65.4%)0.73 Smoking 55 (17.6%)45 (17.2%)10 (19.2%)0.89 Former smoking 87 (27.8%)73 (28.0%)14 (26.9%)1.00Medical history Chronic kidney disease 38 (12.1%)30 (11.5%)8 (15.4%)0.58 COPD 97 (31.0%)82 (31.4%)15 (28.8%)0.84 Asthma 44 (14.1%)38 (14.6%)6 (11.5%)0.72 Chronic heart failure (LVEF < 50%) 59 (18.8%)49 (18.8%)10 (19.2%)1.00 Supraventricular rhythm disorder 66 (21.1%)54 (20.7%)12 (23.1%)0.84 Active cancer 59 (18.8%)48 (18.4%)11 (21.2%)0.79 History of PE 21 (6.7%)15 (5.7%)6 (11.5%)0.13 History of deep vein thrombosis 19 (6.1%)15 (5.7%)4 (7.7%)0.53Long-term oxygen therapy37 (11.8%)36 (13.8%)1 (1.9%)< 0.05Pulse oximetry (SpO₂ %) On arrival of first responders 83.0 [72.0–91.0]83.0 [73.0–91.0]80.0 [66.0–91.0]0.23 On arrival of MICU 96.0 [92.0–99.0]96.0 [92.0–99.0]97.0 [94.0–98.0]0.43 On arrival at ED 97.0 [95.0–99.0]97.0 [95.0–99.0]97.0 [95.0–99.0]0.60Oxygen Therapy Initiation by First Responders263 (84.0%)218 (83.5%)45 (86.5%)0.74Vital parameters Heart rate (bpm) 103.0 [89.0-120.0]104.0 [89.0–120.0]102.0 [88.0–113.8]0.56 Respiratory rate (rpm) 32.0 [28.0–40.0]32.0 [28.0–40.0]31.5 [28.0–38.0]0.53 Systolic blood pressure (mmHg) 151.5 [128.0-174.0]153.5 [126.8–177.0]142.0 [130.0–161.5]0.27 Pathological temperature (< 36 / >38.5 °C) 51.0 (16.3)42 (16.2%)9 (17.3%)1.00Clinical signs Impaired consciousness 47 (15.0%)38 (14.6%)9 (17.3%)0.77 Cyanosis 22 (7.0%)16 (6.1%)6 (11.5%)0.23 Peripheral oedema 78 (25.2)64 (24.6%)14 (28.0%)0.74Normal cardiac auscultation158 (50.6%)130 (50.0%)28 (53.8%)0.72Normal pulmonary auscultation19 (6.1%)13 (5.0%)6 (11.8%)0.10Abnormal pulmonary auscultation292 (93.9%)247 (95.0%)45 (88.2%)0.1 Unilateral crackles 41 (13.2%)35 (13.5%)6 (11.8%)0.92 Bilateral crackles 121 (38.9%)98 (37.7%)23 (45.1%)0.4 Wheezing 120 (38.6%)109 (41.9%)11 (21.6%)0.01 Ronchi 37 (11.9%)28 (10.8%)9 (17.6%)0.25 Silent auscultation 10 (3.2%)9 (3.5%)1 (2.0%)1 Silent auscultation and dullness on percussion 6 (1.9%)5 (1.9%)1 (2.0%)1 Signs of respiratory distress 186 (59.6%)164 (63.1%)22 (42.3%)0.01Normal abdominal examination263 (92.9%)222 (94.1%)41 (87.2%)0.12Respiratory support escalation by MICU103 (32.9%)87 (33.3%)16 (30.8%)0.84 Initiation of non-invasive ventilation 86 (27.5%)76 (29.1%)10 (19.2%)0.2 High-flow oxygen therapy 9 (2.9%)6 (2.3%)3 (5.8%)0.17 Orotracheal intubation 8 (2.6%)5 (1.9%)3 (5.8%)0.13Decrease in oxygen therapy55 (17.6%)45 (17.2%)10 (19.2%)0.89Oxygen therapy weaning21 (6.7%)18 (6.9%)3 (5.8%)1.00Hospitalization195 (62.3%)155 (59.4%)40 (76.9%)0.03Consistent prehospital management284 (90.7%)245 (93.9%)39 (75.0%)< 0.001Death within 28 days45 (14.4%)33 (12.6%)12 (23.1%)0.08Legend: IQR= Interquartile Range; COPD= Chronic Obstructive Pulmonary Disease; LVEF= Left Ventricular Ejection Fraction; PE= Pulmonary Embolism; MICU= Mobile Intensive Care Unit; ED= Emergency DepartmentValues are median [interquartile range] or n (%). This table is restricted to patients with both prehospital MICU and first in-hospital coded diagnoses available. Percentages were calculated using available data for each variable; denominators may therefore vary because of missing data. P values compare concordant and discordant cases and are reported descriptively Table 1Baseline characteristics of the study populationOverall paired-diagnosis population (n = 313)Concordant diagnostic(n = 261)Discordant diagnostic(n = 52)P valuesAge —Years72.0 [62.0-80.0]72.0 [62.0–80.0]73.0 [66.0–80.0]0.43Gender — Female139 (44.4%)117 (44.8%)22 (42.3%)0.86Cardiovascular risk factors Age > 50 274 (87.5%)225 (86.2%)49 (94.2%)0.17 Dyslipidemia 137 (43.8%)121 (46.4%)16 (30.8%)0.05 Hypertension 214 (68.4%)180 (69.0%)34 (65.4%)0.73 Smoking 55 (17.6%)45 (17.2%)10 (19.2%)0.89 Former smoking 87 (27.8%)73 (28.0%)14 (26.9%)1.00Medical history Chronic kidney disease 38 (12.1%)30 (11.5%)8 (15.4%)0.58 COPD 97 (31.0%)82 (31.4%)15 (28.8%)0.84 Asthma 44 (14.1%)38 (14.6%)6 (11.5%)0.72 Chronic heart failure (LVEF < 50%) 59 (18.8%)49 (18.8%)10 (19.2%)1.00 Supraventricular rhythm disorder 66 (21.1%)54 (20.7%)12 (23.1%)0.84 Active cancer 59 (18.8%)48 (18.4%)11 (21.2%)0.79 History of PE 21 (6.7%)15 (5.7%)6 (11.5%)0.13 History of deep vein thrombosis 19 (6.1%)15 (5.7%)4 (7.7%)0.53Long-term oxygen therapy37 (11.8%)36 (13.8%)1 (1.9%)< 0.05Pulse oximetry (SpO₂ %) On arrival of first responders 83.0 [72.0–91.0]83.0 [73.0–91.0]80.0 [66.0–91.0]0.23 On arrival of MICU 96.0 [92.0–99.0]96.0 [92.0–99.0]97.0 [94.0–98.0]0.43 On arrival at ED 97.0 [95.0–99.0]97.0 [95.0–99.0]97.0 [95.0–99.0]0.60Oxygen Therapy Initiation by First Responders263 (84.0%)218 (83.5%)45 (86.5%)0.74Vital parameters Heart rate (bpm) 103.0 [89.0-120.0]104.0 [89.0–120.0]102.0 [88.0–113.8]0.56 Respiratory rate (rpm) 32.0 [28.0–40.0]32.0 [28.0–40.0]31.5 [28.0–38.0]0.53 Systolic blood pressure (mmHg) 151.5 [128.0-174.0]153.5 [126.8–177.0]142.0 [130.0–161.5]0.27 Pathological temperature (< 36 / >38.5 °C) 51.0 (16.3)42 (16.2%)9 (17.3%)1.00Clinical signs Impaired consciousness 47 (15.0%)38 (14.6%)9 (17.3%)0.77 Cyanosis 22 (7.0%)16 (6.1%)6 (11.5%)0.23 Peripheral oedema 78 (25.2)64 (24.6%)14 (28.0%)0.74Normal cardiac auscultation158 (50.6%)130 (50.0%)28 (53.8%)0.72Normal pulmonary auscultation19 (6.1%)13 (5.0%)6 (11.8%)0.10Abnormal pulmonary auscultation292 (93.9%)247 (95.0%)45 (88.2%)0.1 Unilateral crackles 41 (13.2%)35 (13.5%)6 (11.8%)0.92 Bilateral crackles 121 (38.9%)98 (37.7%)23 (45.1%)0.4 Wheezing 120 (38.6%)109 (41.9%)11 (21.6%)0.01 Ronchi 37 (11.9%)28 (10.8%)9 (17.6%)0.25 Silent auscultation 10 (3.2%)9 (3.5%)1 (2.0%)1 Silent auscultation and dullness on percussion 6 (1.9%)5 (1.9%)1 (2.0%)1 Signs of respiratory distress 186 (59.6%)164 (63.1%)22 (42.3%)0.01Normal abdominal examination263 (92.9%)222 (94.1%)41 (87.2%)0.12Respiratory support escalation by MICU103 (32.9%)87 (33.3%)16 (30.8%)0.84 Initiation of non-invasive ventilation 86 (27.5%)76 (29.1%)10 (19.2%)0.2 High-flow oxygen therapy 9 (2.9%)6 (2.3%)3 (5.8%)0.17 Orotracheal intubation 8 (2.6%)5 (1.9%)3 (5.8%)0.13Decrease in oxygen therapy55 (17.6%)45 (17.2%)10 (19.2%)0.89Oxygen therapy weaning21 (6.7%)18 (6.9%)3 (5.8%)1.00Hospitalization195 (62.3%)155 (59.4%)40 (76.9%)0.03Consistent prehospital management284 (90.7%)245 (93.9%)39 (75.0%)< 0.001Death within 28 days45 (14.4%)33 (12.6%)12 (23.1%)0.08Legend: IQR= Interquartile Range; COPD= Chronic Obstructive Pulmonary Disease; LVEF= Left Ventricular Ejection Fraction; PE= Pulmonary Embolism; MICU= Mobile Intensive Care Unit; ED= Emergency DepartmentValues are median [interquartile range] or n (%). This table is restricted to patients with both prehospital MICU and first in-hospital coded diagnoses available. Percentages were calculated using available data for each variable; denominators may therefore vary because of missing data. P values compare concordant and discordant cases and are reported descriptively Baseline characteristics of the study population Legend: IQR= Interquartile Range; COPD= Chronic Obstructive Pulmonary Disease; LVEF= Left Ventricular Ejection Fraction; PE= Pulmonary Embolism; MICU= Mobile Intensive Care Unit; ED= Emergency Department Values are median [interquartile range] or n (%). This table is restricted to patients with both prehospital MICU and first in-hospital coded diagnoses available. Percentages were calculated using available data for each variable; denominators may therefore vary because of missing data. P values compare concordant and discordant cases and are reported descriptively At first responder arrival, median SpO2 was 83% [IQR: 72–91], respiratory rate 32 breaths/min [IQR: 28–40], systolic blood pressure 151 mmHg [IQR: 128–174], and heart rate 103 beats/min [IQR: 89–120]. Abnormal pulmonary auscultation was observed in 94.0% (n = 299) of cases, mainly bilateral crackles (39.0%, n = 124) and wheezing (38.4%, n = 122). Signs of respiratory distress were observed in 59.4% (n = 189) of patients. Oxygen therapy was initiated by first responders in 84.3% (n = 268). MICU escalation of respiratory support occurred in 33.3% (n = 106) of patients; among these, non-invasive ventilation was initiated in 84.0% (n = 89), high-flow oxygen therapy in 8.5% (n = 9), and orotracheal intubation in 7.6% (n = 8). Oxygen therapy was decreased in 17.0% (n = 55) and discontinued in 6.6% (n = 21) of patients. Main result: diagnostic agreement Among the 313 patients with paired prehospital and in-hospital diagnoses, overall agreement of 83.4% (261/313) and a Cohen’s kappa of 0.79 (95% bootstrap CI 0.74–0.84), indicating substantial agreement (Fig. 2). The paired cross-tabulation of prehospital and in-hospital diagnostic categories is provided in Supplementary Table S1. Fig. 2Agreement between prehospital and in-hospital diagnoses for dyspnea-related etiologies. Caption: Points represent Cohen’s kappa coefficients with horizontal bars indicating bootstrap 95% confidence intervals. The vertical solid line indicates no agreement (κ=0), and the dashed line indicates the threshold commonly interpreted as substantial agreement (κ=0.60). Point size is proportional to the number of cases in each diagnostic category. Confidence intervals are wide for rare etiologies because of small sample sizes. The overall kappa coefficient is shown as a diamond. Legend: COPD= Chronic Obstructive Pulmonary Disease Agreement between prehospital and in-hospital diagnoses for dyspnea-related etiologies. Caption: Points represent Cohen’s kappa coefficients with horizontal bars indicating bootstrap 95% confidence intervals. The vertical solid line indicates no agreement (κ=0), and the dashed line indicates the threshold commonly interpreted as substantial agreement (κ=0.60). Point size is proportional to the number of cases in each diagnostic category. Confidence intervals are wide for rare etiologies because of small sample sizes. The overall kappa coefficient is shown as a diamond. Legend: COPD= Chronic Obstructive Pulmonary Disease The four most frequent prehospital MICU diagnostic categories were AHF (33.3%), acute COPD exacerbation (23.6%), pneumonia (19.2%), and asthma exacerbation, together accounting for 87.7% (n = 279) of all cases. These common cardiorespiratory presentations showed the highest category-specific agreement: AHF (κ = 0.76, 95%CI 0.70–0.82; n = 106), acute COPD (κ = 0.88, 95% CI 0.83–0.93; n = 75), pneumonia (κ = 0.67, 95%CI 0.59–0.75; n = 61), and asthma exacerbation (κ = 0.80, 95%CI 0.71–0.89; n = 37). Because these etiologies represented most cases, the overall kappa largely reflects agreement in these common cardiorespiratory presentations. Lower and imprecise agreement estimates were observed for less frequent etiologies, including upper airway dyspnea (κ = 0.56, 95%CI 0.30–0.82, n = 8), digestive causes (κ = 0.18, n = 5), PE (κ = 0.50, 95%CI 0.22–0.78, n = 6), and neurological causes (κ = 0.33, n = 2). When grouped into broader categories, prehospital MICU diagnoses were classified as pulmonary in 56.3% (n = 173), cardiac in 34.0% (n = 108), and other or mixed causes in 10.0% (n = 32). Patterns of diagnostic discordance Among the 52 discordant cases, 46 (88.5%) involved at least one of the four common cardiorespiratory diagnoses (AHF, acute COPD exacerbation, pneumonia, or asthma exacerbation). Twenty-two discordant cases (42.3%) occurred within this group. The most frequent bidirectional mismatch was between AHF and pneumonia (9/52, 17.3% of discordant cases), followed by AHF versus acute COPD exacerbation (5/52, 9.6%) and asthma versus acute COPD exacerbation (3/52, 5.8%). Beyond these recurrent overlaps, most other discordant transitions were individually rare and occurred only once. Exploratory 28-day mortality analysis Among the 313 patients with both diagnoses available, 45 died within 28 days (14.4%). Twenty-eight-day mortality was 23.1% (12/52) in discordant cases and 12.6% (33/261) in concordant cases, corresponding to a crude RR 1.83 (95% CI 1.01–3.29). In adjusted complete cases analysis (n = 269; 41 deaths), the aRR for 28-day mortality associated with diagnostic discordance was 1.59 (95% CI 0.83–3.05). Among patients who died within 28 days, the median time to death was 5 days [IQR 2–11]. Prehospital management consistency according to concordance status Overall, prehospital management was judged consistent with key principles of acute dyspnea management in 90.9% (289/318). Among patients with paired diagnoses, consistent management was less frequent in discordant cases than in concordant cases (75.0% [39/52] vs. 93.9% [245/261]). Among the 29 cases judged inconsistent, 44.8% (n = 13) occurred in the context of diagnostic discordance, while 55.2% (n = 16) occurred despite diagnostic concordance. Discussion This study adds data from a French physician-staffed MICU system on diagnostic agreement between prehospital and in-hospital diagnoses in patients managed for dyspnea. Overall agreement was substantial, with 83.4% agreement and a Cohen’s kappa of 0.79, a level of agreement consistent with previous international studies [14–16]. Notably, the prospective multicenter study by Veldhuis et al. reported a comparable diagnostic concordance, supporting the external coherence of our findings despite differences in healthcare systems and EMS organization [19]. These results confirm that diagnostic concordance is highest for frequent and prototypical cardiorespiratory causes of dyspnea, including AHF, acute COPD, pneumonia, and asthma, whereas agreement was lower for less common or atypical, mixed or overlapping presentations. This pattern has been reported consistently in previous studies [1, 9, 14–17] and likely reflects intrinsic constraints of prehospital emergency medicine, including restricted time, limited access to complementary investigations, and the need for rapid decision-making based on incomplete or evolving clinical information [10, 12]. Notably, most discordances involved overlap between common cardiorespiratory syndromes, particularly AHF, pneumonia, and acute COPD exacerbation, rather than clearly unrelated diagnostic categories. In crude analyses, diagnostic discordance was associated with higher 28-day mortality. However, this association was attenuated after adjustment for baseline severity, with wide confidence intervals. These data therefore do not rule out a clinically relevant association, but they do not support a causal interpretation. Overall, this pattern suggests that part of the excess crude mortality observed in discordant cases may reflect greater initial severity or clinical complexity rather than discordance itself [18, 19]. Importantly, approximately three-quarters of patients with diagnostic discordance still received prehospital care judged consistent with key principles of acute dyspnea management [22]. This finding likely reflects the French physician-staffed MICU model, in which on-scene assessment and early stabilization are central [26]. This apparent discrepancy is clinically plausible. In acute dyspnea, early prehospital care is often syndrome-oriented and physiology-driven, so initial treatment may remain appropriate even when the final coded hospital diagnosis differs from the prehospital working diagnosis. From a pragmatic point of view, these findings suggest that additional diagnostic tools may be useful when the initial clinical presentation is ambiguous, overlapping, or atypical, rather than as systematic tools for all dyspneic patients. Previous studies have demonstrated that lung and cardiac ultrasound can improve diagnostic accuracy in acute dyspnea, particularly for differentiating cardiac and pulmonary causes [27–31]. However, comprehensive protocols such as “BLUE” or “FATE” were primarily developed for in-hospital use and may be difficult to apply systematically in the prehospital setting. Simplified, goal-directed ultrasound strategies tailored to specific clinical questions may therefore represent a more feasible and impactful approach in prehospital care. Conversely, point-of-care blood gas analysis may offer added value in certain patients by enabling the early identification of severe respiratory or metabolic derangements and facilitating dynamic reassessment of treatment response [32, 33]. However, its impact on diagnostic concordance and outcomes remains to be formally evaluated. Limitations Several limitations should be acknowledged. First, this was a single-centre study conducted in a physician-staffed MICU system and restricted to patients who were subsequently hospitalized at Orléans University Hospital. Patients not hospitalized at the study hospital, those who died before MICU arrival or during transport, and those without a documented prehospital diagnosis were excluded. These exclusions may have selected patients with more complete documentation and diagnostic workup and may have excluded more unstable, complex, or chaotic presentations, thereby potentially overestimating diagnostic agreement. Second, the in-hospital reference was the first coded diagnosis recorded during initial hospital management rather than an adjudicated final diagnosis. Findings should therefore be interpreted as inter-setting diagnostic agreement rather than confirmed etiologic accuracy. In addition, prehospital and in-hospital clinicians belonged to the same institutional ecosystem and communicated directly at handover. Shared training, local protocols, documentation habits, and continuity of care may have favoured diagnostic alignment, particularly for common conditions such as acute heart failure, COPD exacerbation, asthma, and pneumonia. Third, although the exploratory mortality analysis was adjusted a priori for several markers of baseline severity, residual confounding is likely because important determinants such as comorbidity burden, laboratory data, imaging results, and some aspects of treatment response were not available or not modelled. This analysis should therefore not be interpreted causally. Fourth, this study relied on routinely collected clinical records that were not created for research purposes. Misclassification, incomplete documentation, missing data, and changes in documentation practices may have affected both diagnostic classification and the retrospective assessment of management consistency. Because the guidance used for the post hoc assessment was published after part of the study period, it was used as a standardizing interpretive framework reflecting long-standing principles of acute dyspnea care rather than as a strict audit of contemporaneous compliance. Finally, although dates of death were available for deceased patients, complete follow-up time was not available for survivors. Mortality was therefore analysed as a binary 28-day outcome rather than with a formal time-to-event approach. Conclusion In this selected population of patients managed by a physician-staffed MICU for dyspnea and with paired diagnostic information available, diagnostic agreement between the prehospital MICU diagnosis and the first in-hospital coded diagnosis was substantial. Most discordant cases still received prehospital management judged consistent with key principles of acute dyspnea management. The association between discordance and 28-day mortality was attenuated after adjustment, with imprecise estimates. Larger multicenter studies using adjudicated diagnoses are needed to determine whether diagnostic discordance primarily reflects clinical complexity or independently influences outcomes. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1