Predicting Acute Kidney Injury After Lung Transplantation: Development and Internal Validation of Perioperative Prediction Models
Shogo Trevena, Gregory I. Snell, Glen P. Westall, Mohammad Asghari‐Jafarabadi, Mark A. Shulman, Tim G. CoulsonABSTRACT
Background
Acute kidney injury (AKI) is common following lung transplantation (LTx), and is associated with increased morbidity and mortality. Existing renal injury risk models have not systematically evaluated perioperative variables. We aimed to develop comprehensive preoperative and early postoperative prediction models for AKI and renal replacement therapy (RRT) following LTx.
Methods
Adult LTx recipients at a single Australian tertiary center (January 2016–April 2025) were included in a retrospective cohort study. The KDIGO 2012 serum creatinine criteria were used to define AKI. Stable predictors identified via penalized regression with bootstrap resampling were incorporated into multivariable logistic regression models. Bootstrap internal validation provided optimism‐adjusted performance estimates, and decision curve analysis evaluated net clinical benefit.
Results
Of 534 patients, 216 (40.4%) developed AKI, and 47 (8.8%) required RRT. The preoperative AKI model, incorporating baseline demographics, comorbidities, and laboratory variables, achieved a C‐statistic of 0.632 (95%CI: 0.589–0.679). The postoperative AKI model demonstrated improved discrimination (C‐statistic 0.741, 95%CI: 0.697–0.795), incorporating preoperative variables such as chronic lung allograft dysfunction as a transplant indication, alongside procedure duration and novel early postoperative predictors: P/F ratio and cumulative noradrenaline dosage. The RRT models achieved C‐statistics of 0.720 (95%CI: 0.643–0.796) and 0.847 (95%CI: 0.781–0.917). Calibration slopes ranged from 0.889‐0.949. Decision curve analysis demonstrated net clinical benefit over treat‐all strategies across clinically relevant threshold probabilities.
Conclusions
Early postoperative models demonstrated superior discrimination and clinical utility, with timepoint‐specific risk stratification capturing distinct perioperative determinants of renal injury. These models provide a foundation for external validation and targeted nephroprotective research in LTx.