Severity-stratified vasopressor-escalation strategies in septic shock: a cross-fitted doubly robust analysis with external replication.

Background: The optimal vasopressor-escalation strategy after norepinephrine in septic shock remains uncertain, with randomized trials showing no clear winner and observational data suggesting severity-dependent effects. We characterized severity-stratified heterogeneity for adding vasopressin or epinephrine to norepinephrine and tested external replicability. Materials and Methods: We emulated a target trial in Medical Information Mart for Intensive Care IV (MIMIC-IV) (derivation) and eICU Collaborative Research Database (eICU-CRD) (external validation), enrolling adults with Sepsis-3 septic shock who started norepinephrine. Patients were classified as S1 (norepinephrine alone), S2 (plus vasopressin), or S3 (plus epinephrine) by the agent added in a baseline window. The primary outcome was 14-day administratively truncated in-hospital mortality. Average treatment effects were estimated with a cross-fitted doubly robust learner, anchored on a prespecified norepinephrine-equivalent dose below 0.25 μg/kg/min, with subgroups, sensitivity analyses, and E-values. A depth-2 policy tree summarized heterogeneity, not a decision rule. Results: After trimming, 2,415 MIMIC-IV and 568 eICU-CRD patients were analyzed. Cohort-level effects were heterogeneous and did not replicate: in MIMIC-IV, S3 Conclusion: Vasopressor-escalation effects were heterogeneous and concentrated at low baseline severity, with a directionally consistent but unconfirmed signal favoring epinephrine below the Surviving Sepsis Campaign dose threshold. These findings are hypothesis-generating and do not support changing first-line practice; future trials should stratify on baseline norepinephrine-equivalent dose rather than enroll uniform cohorts.