Blood pressure response index trajectories identify distinct hemodynamic phenotypes and predict mortality in septic shock: a two-database retrospective cohort study

Background: Vasopressor responsiveness in septic shock is typically assessed using static metrics that cannot capture temporal hemodynamic evolution. The Blood Pressure Response Index (BPRI = mean arterial pressure / Vasoactive-Inotropic Score) integrates hemodynamic response and treatment intensity into a single metric, but its longitudinal trajectory patterns remain unexplored. Methods: We applied latent class mixed models to BPRI trajectories during the first 48 h of vasopressor therapy in 4,389 septic shock patients from MIMIC-IV (development cohort). External validation was performed via parameter transport to 1,240 eICU-CRD patients. The prognostic significance of trajectory phenotypes was assessed using multivariable logistic and Cox regression with a three-level adjustment framework, restricted mean survival time analysis, and incremental predictive value assessment beyond conventional severity scores. Results: Six distinct hemodynamic phenotypes were identified with ICU mortality ranging from 21.9% (C3 Responders) to 54.5% (C2 Non-Responders). Parameter transport validation showed preserved class separation and prognostic gradient (average posterior probability 0.960) in eICU-CRD. After full multivariable adjustment, C2 (OR 3.67, 95% CI 2.76-4.86) and C1 (OR 2.68, 95% CI 2.08-3.46) remained independently associated with ICU mortality. Restricted mean survival time analysis showed the largest adjusted losses for C2 (- 2.56 days at τ = 14 days) with minimal attenuation from unadjusted estimates, suggesting an association that persisted after comprehensive adjustment. Adding trajectory classification to severity scores yielded statistically significant incremental discrimination (ΔAUC + 0.020, P < 0.001), while static BPRI added no further information. Conclusions: BPRI trajectory analysis identifies six hemodynamic phenotypes in septic shock that are validated in an independent external database, are independently associated with mortality, and capture temporal hemodynamic response patterns missed by static assessments. These phenotypes may facilitate risk stratification and enrichment strategies for clinical trials targeting vasopressor-dependent patients.