The Causal Artificial Intelligence Clinician for early haemodynamic management of septic shock in ICU

Rationale: Standardizing fluid and vasopressor resuscitation in septic shock is challeng ing due to patient heterogeneity. Objectives: To identify optimal intravenous fluid and vasopressor doses in first six hours of ICU admission in septic shock. Methods: Graphical causal inference models grounded in expert clinical knowledge were applied to estimate heterogeneous treatment effects, limiting bias from spurious correlations. The model was trained on 1,706 MIMIC admissions and externally validated on 1,450 eICU admissions. The primary outcome was in-hospital survival; clinical improvement was the secondary outcome. Measurements and Main Results: The cohort comprised 3,156 participants (age 65 years [IQR 53-75]; 42.8% female). Deviation from vasopressor recommendations was associ ated with failed clinical improvement (median OR 1.26, 95% CI 1.21-1.32), the most stable association across twelve sensitivity configurations, and with in-hospital mortality (median OR 1.16, 95% CI 1.10-1.24), which was null in five of twelve. Fluid deviations yielded 1.12 (1.08-1.18) and 1.02 (.96-1.07) respectively. These associations are confounded by indica tion by construction. External validation AUROCs were 0.72 (95% CI 0.67-0.76) for survival and 0.70 (95% CI 0.66-0.73) for improvement, comparable to predictive baselines. Treat ment response varied with baseline physiology, fluids showing neutral or positive effects and high-dose norepinephrine worse ones. Conclusions: An expert-grounded causal framework produced recommendations that be haved as an informative policy in external validation, using one-third the variables of a con ventional predictive baseline. Whether alignment with those recommendations improves out comes requires prospective evaluation. Code and instructions to reproduce this analysis, in cluding a dashboard to explore our results, are available at https://github.com/IDSIA/causal ai-clinician.