HLAP早期重症预测：微观指标与联合模型

Early Prediction of Severe Hyperlipidemic Acute Pancreatitis: Microscopic Indicators and Combined Models

摘要: 高脂血症性急性胰腺炎(HLAP)的发病率逐年上升，其特有的“脂毒性”与游离钙耗竭机制导致早期风险分层面临挑战。本文系统梳理HLAP重症化的病理生理基础，重点评估甘油三酯–葡萄糖(TyG)指数与白蛋白校正血钙(Ca_corr)等微观代谢指标及其联合模型的早期预测价值。现有证据表明：TyG指数及其衍生指标(如TyG-BMI)预测HLAP重症化的受试者工作特征曲线下面积(AUC)可达0.830~0.891，但易受早期应激性高血糖与液体复苏的混杂干扰；发病24小时内Ca_corr锐减预测持续性器官衰竭的AUC可达0.888，但在并发酸碱失衡时存在方法学失效风险。基于多维指标构建的列线图及机器学习模型预测效能优异，C-index最高可达0.966，但多数现有模型局限于单中心回顾性设计，缺乏独立的外部验证与Brier分数等校准度评估，存在过拟合风险且临床干预阈值界定模糊。后续研究应重点关注动态指标(如ΔTyG、ΔCa_corr)监测、融合多维特征(炎症、凝血、影像)构建复合模型，并依托多中心前瞻性队列进一步探索启动干预的量化阈值。

Abstract: The incidence of hyperlipidemic acute pancreatitis (HLAP) is steadily rising, and its unique mechanisms of “lipotoxicity” and free calcium depletion pose significant challenges for early risk stratification. This review systematically outlines the pathophysiological basis of HLAP progression to severe disease, evaluating the early predictive value of microscopic metabolic biomarkers, such as the triglyceride-glucose (TyG) index and albumin-corrected calcium (Ca_corr), alongside their combined models. Current evidence demonstrates that the TyG index and its derivative indicators (e.g., TyG-BMI) achieve an area under the receiver operating characteristic curve (AUC) of 0.830~0.891 for predicting severe HLAP; however, they are susceptible to confounding by early stress-induced hyperglycemia and fluid resuscitation. Within 24 hours of onset, a sharp decline in Ca_corr predicts persistent organ failure with an AUC reaching up to 0.888, but its methodological validity is compromised in the presence of concurrent acid-base imbalance. Nomograms and machine learning models constructed based on multidimensional indicators exhibit excellent predictive performance, with C-indices reaching up to 0.966. Nevertheless, most existing models are constrained by single-center retrospective designs, lacking independent external validation and calibration assessments such as Brier scores, which introduces a substantial risk of overfitting and leaves clinical intervention thresholds poorly defined. Future research should prioritize the monitoring of dynamic indicators (e.g., ΔTyG, ΔCa_corr), the integration of multidimensional features (inflammation, coagulation, imaging) to construct composite models, and further exploration of quantitative thresholds for initiating interventions based on multicenter prospective cohorts.