Causal Effect and Personalization of Intraoperative Hypotension Burden on Postoperative Acute Kidney Injury: A Doubly Robust Analysis of the VitalDB Cohort.

Background

Intraoperative hypotension (IOH) is the leading modifiable contributor to postoperative acute kidney injury (AKI), yet most evidence is associational and the heterogeneity of its effect is unknown. We estimated the causal effect of IOH burden on AKI and tested whether the most susceptible patients can be identified preoperatively.

Methods

In a retrospective cohort of 2726 general-anesthesia cases from VitalDB, the exposure was the time-integrated mean arterial pressure (MAP) <65 mmHg burden (≥30, ≥60 and ≥120 mmHg·min) and the outcome was KDIGO-defined AKI within 7 days. The primary estimator was pre-treatment-adjusted augmented inverse-probability weighting (AIPW; doubly robust) with bootstrap 95% confidence intervals (CIs). Sensitivity analyses comprised a controlled-direct-effect model, negative control outcomes, E-values and vasopressor-stratified estimates. Effect heterogeneity was estimated with a causal forest; preoperative gradient-boosted models and decision-curve analysis assessed personalization and clinical utility.

Results

AKI occurred in 205 (7.52%) cases. At 60 mmHg·min the AIPW risk difference was +3.00 percentage points (pp; 95% CI +0.84 to +5.26), with a monotonic dose-response (+2.78 to +7.62 pp across thresholds) and E-values rising from 2.08 to 3.44. The effect was concentrated in patients with elevated preoperative creatinine (conditional effect +9.86 pp, more than twice the cohort average). This susceptibility was recoverable from routine preoperative variables alone, with intraoperative waveform features conferring no measurable improvement (ΔAUROC -0.001). For predicting AKI itself, a parsimonious 4-feature preoperative score matched a 27-feature model (AUROC 0.775 vs. 0.768) and provided positive net benefit.

Conclusions

Intraoperative hypotension burden shows a dose-dependent association with postoperative AKI that is consistent with a causal effect, concentrated in patients with reduced baseline renal reserve who are identifiable from routine preoperative data without intraoperative waveform infrastructure.