Intraoperative Arterial Waveform Morphology Trajectories and Prediction of Postoperative Acute Kidney Injury: A Latent-Class Analysis of the VitalDB Registry

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Background: Intraoperative hypotension is an established risk factor for postoperative acute kidney injury (AKI), but whether arterial waveform morphology adds prognostic information beyond the hypotension burden is unknown. Methods: In this study, 2587 adult surgical cases from the Vital Signs Database (VitalDB) with continuous invasive arterial waveforms and creatinine-based (KDIGO) AKI ascertainment were analyzed retrospectively. Maximal systolic upstroke (dP/dtmax), a late-diastolic pressure-decay descriptor (τ), and a waveform-derived dynamic arterial elastance (Eadyn), modeled as latent-class trajectories and as case-level features, were added to a base logistic model of log-transformed hypotension burden (time-weighted MAP < 65 mmHg), age, sex, and baseline eGFR. Results: AKI occurred in 143 cases (5.5%); the base model had an AUC of 0.725 (95% CI 0.680–0.770). A five-class dP/dtmax solution yielded distinct phenotypes whose membership was nominally associated with AKI (likelihood-ratio p = 0.02) but did not improve discrimination (change in AUC (ΔAUC), 0.008, 95% CI −0.005 to 0.021; optimism-corrected 0.002), and the association was attenuated after clinical adjustment (p = 0.13); τ showed no stable trajectory structure. No feature improved discrimination: the optimism-corrected ΔAUC was ≤0.007, upper confidence limits did not exceed 0.03 on any adequately fitting reference model, and the signal disappeared after adjustment for procedure type and comorbidity. The likelihood-ratio test had 80% power for a ΔAUC of 0.009. Conclusions: Arterial waveform morphology did not improve prediction of postoperative AKI beyond the hypotension burden; overall, these data exclude an incremental gain larger than approximately 0.02–0.03 AUC units.