Early Postoperative Predictors of 30-Day Mortality After Pediatric Liver Transplantation: A Trajectory-Based Analysis

Background/Objectives: Early mortality after pediatric liver transplantation remains a clinical challenge, yet few studies have specifically addressed 30-day outcomes. Conventional pretransplant scores such as the age-appropriate MELD/PELD score were not designed for post-transplant risk prediction. We aimed to evaluate whether dynamic postoperative biomarker trajectories and novel composite ratios can identify high-risk patients. Methods: This single-center retrospective cohort study included 140 consecutive pediatric patients (<18 years) who underwent primary liver transplantation between 2015 and 2023. Patients were classified as deceased (≤30 days, n = 11) or survivors (>30 days, n = 129). PRISM-III, PELOD-2, and age-appropriate MELD/PELD scores were evaluated. Serial laboratory parameters were collected at pretransplant and at 0, 24, and 72 h. Delta (Δ) values and composite ratios-including lactate clearance, lactate-to-albumin ratio (LAR), INR×lactate product, platelet ratio, and fibrinogen/INR-were calculated. Penalized logistic regression (Firth method) was used for multivariate analysis. Internal validation was performed using bootstrap resampling (1000 iterations) and leave-one-out cross-validation (LOO-CV). Because two of the three components of the multivariable model (ΔINR, ΔALT) were derived from 72-h values, the model is best understood as a 72-h landmark risk model rather than as an immediate post-transplant early-warning tool. Results: The 30-day mortality rate was 7.9% (11/140), with central nervous system complications as the leading cause (36.4%). PRISM-III demonstrated excellent discrimination (AUROC 0.957; cut-off ≥ 14); the age-appropriate MELD/PELD score, a pretransplant tool not designed for post-transplant prediction, showed near-chance performance (AUROC 0.513; p = 0.576). A distinctive biomarker crossover pattern was observed: non-survivors had paradoxically lower pretransplant INR, ALT, and LAR values, but trajectories diverged sharply by 24 h. The INR×lactate product achieved an AUROC of 0.981 at 72 h. LAR at 24 h achieved 0.909, and lactate clearance at 0 → 72 h achieved 0.783. Postoperative hypernatremia emerged as a strong predictor (AUROC 0.884). In multivariate analysis, PRISM-III (OR 4.00), ΔINR (OR 3.28), and ΔALT (OR 3.46) were independent predictors (apparent AUROC 0.989). Internal validation confirmed model stability: bootstrap-corrected AUROC was 0.978; LOO-CV AUROC was 0.957 (sensitivity 90.9%, specificity 96.9%). Conclusions: Dynamic postoperative factors-rather than pretransplant disease severity-appeared more strongly associated with 30-day mortality after pediatric liver transplantation in this single-center exploratory analysis. The INR×lactate product, a novel two-variable composite, showed very high apparent discrimination (AUROC 0.981) and is proposed as a hypothesis-generating candidate marker requiring prospective external validation before any clinical use. The combined PRISM-III + ΔINR + ΔALT model (best understood as a 72-h landmark risk model, since two of its three components are defined at 72 h postoperatively) demonstrated robust internal validation performance (LOO-CV AUROC 0.957); however, given the small number of events (n = 11) and the absence of external validation, the model should be regarded as exploratory.