Association of albumin trajectories and cumulative exposure with in-hospital mortality in acute pancreatitis: a retrospective cohort study
Yaoyu Zou, Maobin Kuang, Shixuan Xiong, Xin Xu, Xue-Yang Li, Ling Ding, Cong He, Nianshuang Li, Huajing Ke, Xin Huang, Yupeng Lei, Huifang Xiong, Wenhua He, Lingyu Luo, Xia Liang, Nonghua Lü, Jianhua Wan, Yin Zhu
Abstract
BACKGROUND: Hypoalbuminemia is common in acute pancreatitis (AP) and associated with poor outcomes, but its cumulative and dynamic impacts are not well characterized. METHODS: This study included 3,214 AP patients (2005-2023). Cumulative albumin exposure (CumALB) was calculated over the first 7 days using the trapezoidal method. Latent class mixed modeling (LCMM) identified distinct albumin trajectories. Multivariable logistic regression and restricted cubic splines (RCS) assessed associations between CumALB and in-hospital mortality. Kaplan-Meier curves compared survival across trajectories. A simplified nomogram was developed using predictors selected via Boruta and LASSO algorithms, with performance assessed by AUC, calibration, and decision curves. Validation was conducted in three cohorts: the MIMIC-IV database ( n = 514), the eICU-CRD database ( n = 211), and the local cohort from 2024 ( n = 880). RESULTS: Patients in the lowest CumALB tertile had significantly higher mortality (11.0%) than those in the highest tertile (1.3%, P < 0.001). Each 1-SD increase in CumALB reduced mortality risk by 42% (adjusted OR = 0.58, 95% CI: 0.44-0.76, P < 0.001), with RCS confirming a linear inverse relationship. CumALB outperformed single-day albumin in mortality prediction. LCMM identified four trajectories; low-stable (LS-T1) had highest mortality (17.7%), whereas low-increasing (LI-T4) had lower mortality (8.6%). Compared with LS-T1, high-stable (HS-T2) and LI-T4 groups had reduced risks (adjusted OR = 0.16 and 0.49, respectively; P < 0.05). The CumALB-based nomogram achieved an AUC of 0.836 in the training set (70%) and 0.864 in the internal test set (30%), outperforming APACHE II (AUC = 0.76), SIRS (AUC = 0.72), Ranson (AUC = 0.72), and BISAP (AUC = 0.80). Validation was conducted using three independent cohorts: the MIMIC-IV cohort (AUC = 0.631), the eICU-CRD cohort (AUC = 0.681), and the local cohort from 2024 (AUC = 0.844). Sensitivity analyses further confirmed the robustness of these results. CONCLUSION: CumALB may provide a useful measure of the cumulative burden of hypoalbuminemia in patients with acute pancreatitis, and albumin trajectories could help capture its dynamic changes. The CumALB-based predictive model demonstrated good performance in predicting mortality risk, potentially assisting clinicians in early risk stratification and clinical decision-making for high-risk patients within heterogeneous populations. However, prospective validation is warranted before clinical implementation.