Abstract / Summary
Severe acute pancreatitis (SAP) remains associated with substantial ICU mortality, while early fluid management remains uncertain. Whether dynamic abdominal pressure–perfusion trajectories identify reproducible prognostic subphenotypes and whether these classes exhibit distinct fluid–outcome associations remain unknown. In this multicentre observational cohort study, we jointly modelled mean arterial pressure (MAP) and intra-abdominal pressure (IAP) trajectories over ICU days 1–7 using group-based trajectory modelling in 1008 adults with SAP (development cohort, 2005–2023) and 315 patients in an independent external cohort (TRACE trial plus Nanjing Jinling Hospital, 2023–2024). Multivariable Cox models compared 28-day mortality across subphenotypes. Lagged linear mixed-effects models examined associations between fluid volume at prespecified time points and subsequent MAP/IAP changes at the next scheduled assessment; Bayesian models were implemented as sensitivity analyses to assess model robustness. Five-fold out-of-fold (OOF)-predicted Day-1 labels were used for hypothesis-generating analyses of fluid–mortality associations, with sensitivity analyses comparing OOF-predicted labels with GBTM-derived trajectory assignments. Four prognostic abdominal pressure–perfusion subphenotypes were identified (average posterior probability 0.913): Stable (26.3%), Hypoperfusion (26.7%), Hypertensive decompensated (18.1%), and Hypertensive compensated (29.0%). 28-day mortality was 5.7%, 17.8%, 29.7%, and 8.2% (log-rank P < 0.001); adjusted HRs (95% CIs) versus Stable were 4.35 (2.16–8.74), 5.49 (2.68–11.24), and 1.55 (0.71–3.40), respectively. The external cohort reproduced the mortality ranking (log-rank P = 0.005). Day-t fluid volume showed a global fluid × subphenotype interaction for ΔIAP at the next scheduled measurement ( P = 0.007); within the Hypertensive decompensated class, each additional litre was associated with a 0.40 mmHg greater increase in IAP (95% CI 0.23–0.57). In a secondary exploratory analysis, greater cumulative Days 1–2 fluid volume was associated with higher 28-day mortality within the Day-1 OOF-predicted Hypoperfusion class (adjusted HR 1.26 per litre, 95% CI 1.08–1.48; P = 0.004); the corresponding global fluid × class interaction was not statistically significant ( P = 0.085). Four prognostic abdominal perfusion trajectory subphenotypes with divergent 28-day mortality were identified, with similar patterns and mortality ranking in an independent cohort. Joint MAP–IAP trajectories may inform prognostic stratification, whereas class-stratified fluid–outcome associations varied by endpoint and label source and require prospective validation. Not applicable (observational study). External validation included the TRACE trial (ClinicalTrials.gov NCT02473406).