Identifying clinical subtypes in acute dyspnoea patients admitted to the emergency department: a secondary model-based clustering of prospective cohorts.

INTRODUCTION

Acute dyspnoea is a frequent reason for emergency department (ED) presentation, with heterogeneous causes and prognosis often insufficiently captured by traditional diagnostic categories. We applied model-based clustering (i.e. latent class analysis [LCA]) to identify distinct subtypes of acute dyspnoea and assess their prognostic relevance.

METHODS

We analysed two prospective ED acute dyspnoea adult cohorts (LEDA and BASEL V). LCA was performed in the LEDA derivation cohort using nine admission clinical/biological variables. A simplified decision tree was derived in the LEDA cohort and externally validated to classify BASEL V patients. Subtypes were compared regarding clinical characteristics, biomarker profiles, and 90-day mortality.

RESULTS

Among 1392 LEDA and 1886 BASEL V patients, four reproducible subtypes were identified and labelled as 'non-inflammatory' (A), 'tachycardic' (B), 'anaemic' (C) and 'hypoxemic' (D). Subtypes transcended conventional diagnostic categories. Inflammatory and cardiovascular biomarker levels increased significantly from A to D (all P < .001). After adjusting for traditional risk factors, assignment remained independently associated with the 3-month mortality risk, with the worst outcome for patients assigned to the 'hypoxemic' subtype [adjusted HR (95% CI) for subtype D: LEDA 3.87 (2.17-6.87), P < .0001; BASEL V 5.15 (2.71-9.76), P < .001]. Adding subtype membership improved the Harrell C-index for 3-month mortality beyond a linear combination of the three main class-defining variables (LEDA 0.73 to 0.77, P = .04; BASEL V 0.76 to 0.79, P = .03).

CONCLUSIONS

A model-based clustering approach identified four reproducible acute dyspnoea subtypes with distinct clinical, biomarker, and prognostic profiles. This framework may improve early risk stratification and personalized ED management beyond nosologically classified diagnoses.