BPD预测模型的演进、比较与临床转化挑战

Evolution, Comparison, and Clinical Translation Challenges of Bronchopulmonary Dysplasia (BPD) Prediction Models

摘要: 支气管肺发育不良(Bronchopulmonary Dysplasia, BPD)是早产儿，尤其是极早产儿与极低出生体重儿中常见的慢性肺疾病之一，早期预测对改善预后及实现个体化干预至关重要。近四十年来已开发大量BPD预测模型，但总体上看其可操作性与实用性较差，临床转化率低。本文回顾了BPD预测模型从早期临床评分、标准化多变量模型到现代机器学习预测的演进历程，重点比较了Henderson-Smart 2006、Laughon 2011、Valenzuela-Stutman 2019、Shim 2021等关键模型的区分度、校准度与临床实用性；深入剖析了阻碍模型临床转化的核心障碍，包括方法学质量不足、外部验证缺失、数据时效性滞后、预测因子依赖临床实践、BPD严重程度分级不统一以及缺乏改善患儿结局的前瞻性证据。本文还总结了新型生物标志物，如基因组学、蛋白组学，在预测模型中的应用潜力。未来研究亟需遵循TRIPOD等国际规范，基于当代多中心数据构建并持续验证模型，同时推动模型与临床工作流程的整合，开展前瞻性效用研究，以真正实现预测模型在BPD个体化防治中的临床价值。

Abstract: Bronchopulmonary Dysplasia (BPD) is a common chronic lung disease in preterm infants, particularly in very preterm and very low birth weight infants. Early prediction of BPD is crucial for improving prognosis and enabling individualized intervention. Although numerous BPD prediction models have been developed over the past four decades, their overall operability and practicality remain limited, resulting in low rates of clinical translation. This article reviews the evolution of BPD prediction models from early clinical scores and standardized multivariable models to modern machine learning-based approaches. Key models such as Henderson-Smart 2006, Laughon 2011, Valenzuela-Stutman 2019, and Shim 2021 are critically compared in terms of discrimination, calibration, and clinical utility. The core barriers hindering clinical translation are thoroughly analyzed, including methodological shortcomings, lack of external validation, outdated data, dependence of predictors on clinical practices, inconsistency in BPD severity grading and absence of prospective evidence on patient-important outcomes. Additionally, the review summarizes the potential of novel biomarkers, such as those from genomics and proteomics in predictive modeling. Future research should adhere to international reporting guidelines such as TRIPOD, develop and continuously validate models using contemporary multicenter data, promote integration into clinical workflows, and conduct prospective utility studies to realize the clinical value of prediction models in the individualized prevention and management of BPD.