Early risk assessment model for sepsis-associated acute kidney injury combining quantitative contrast-enhanced ultrasound and biomarkers: a prospective pilot study

Background Sepsis-associated acute kidney injury (SA-AKI) is associated with high mortality in the intensive care unit (ICU), yet early diagnosis remains challenging due to the diagnostic lag of conventional functional biomarkers. We aimed to develop and validate a Ridge-penalized multimodal model integrating quantitative renal contrast-enhanced ultrasound (CEUS) with clinical biomarkers for early SA-AKI risk stratification. Methods Adult septic patients were prospectively enrolled in a 35-bed general ICU. Quantitative renal CEUS parameters—including peak intensity (PI), time to peak (TTP), mean transit time (MTT), and area under the time-intensity curve (TIC-AUC)—and routine clinical variables were collected within 24 h of admission. SA-AKI was defined according to the Kidney Disease: Improving Global Outcomes criteria within 7 days. A multimodal Ridge logistic regression model (Model C) was developed using stratified 5-fold cross-validation ( λ 1SE rule) and compared with CEUS-only (Model A) and clinical-only (Model B) candidate models. Internal validation was performed using 1,000 bootstrap resamples. Multivariable Firth-penalized regression was used to evaluate independent predictors. Results Among 63 analyzed patients, 25 (39.7%) developed SA-AKI at a median lead time of ICU day 2 (IQR: day 1–day 3) post-admission; none met AKI criteria at baseline. Quantitative CEUS revealed significant cortical hypoperfusion (PI and TIC-AUC) and prolonged microvascular transit (MTT) in the SA-AKI group (all P < 0.05), while other temporal or slope parameters showed non-significant trends. Model C achieved superior discrimination with a bias-corrected area under the curve (AUC) of 0.852 (95% CI: 0.789–0.932), outperforming Model A (0.818) and Model B (0.740). At the optimal threshold, Model C yielded 80.0% sensitivity and 84.2% specificity. Decision curve analysis confirmed higher standardized net benefit for Model C across threshold probabilities of 0.15–0.60. Adjusted age and baseline serum cystatin C confirmed lower PI (aOR = 0.31), lower TIC-AUC (aOR = 0.31), and prolonged TTP (aOR = 2.1) as independent predictors (all P < 0.05). Conclusions Quantitative renal CEUS parameters are independent early indicators of SA-AKI. Integrating renal CEUS with clinical biomarkers offers a promising, noninvasive strategy for early SA-AKI risk stratification up to 48 h before clinical diagnosis. Clinical Trial Registration https://www.chictr.org.cn/ , identifier ChiCTR2300067267.