原发性血小板增多症血栓与出血事件的风险 预测研究进展

Research Progress in Risk Prediction of Thrombosis and Hemorrhage Events in Essential Thrombocythemia

摘要: 原发性血小板增多症(essential thrombocythemia, ET)是BCR-ABL1阴性骨髓增殖性肿瘤里最常见的亚型，血栓与出血事件作为病程中引发残疾和致命性并发症的主要原因，能够严重损害到患者的生活质量并缩短其生存期，随着研究工作的不断深入，目前针对原发性血小板增多症的血栓风险分层方式已经无法充分满足临床上的需求，而且至今还没有形成公认的出血风险分层标准，所以明确这两类事件的危险因素并实现精准预测，已经成为临床研究需要重点关注的核心问题，现有研究已经证实，年龄、血栓病史、JAK2基因突变、心血管危险因素等都与原发性血小板增多症患者的血栓事件存在相关性，而对于出血事件来说，诊断时的年龄、白细胞计数、低血红蛋白等指标则是相关的影响因素，机器学习算法能够高效地处理海量的临床数据集，识别出各参数之间复杂的非线性关系，通过把临床参数与分子生物标志物整合起来构建预测模型，已经展现出更高的预测准确性与可靠性，尽管目前基于机器学习构建的原发性血小板增多症血栓和出血事件预测模型已经取得了良好的研究成果，但仍然面临着数据质量、模型可解释性以及临床泛化性等方面的挑战，未来的研究工作应当聚焦于算法优化、多维度数据整合，并推动此类模型在临床上的转化与应用，从而为原发性血小板增多症患者的个体化风险分层及治疗策略制定提供更为精准的工具支撑。

Abstract: Essential thrombocythemia (ET) is the most common subtype among BCR-ABL1-negative myeloproliferative neoplasms, with thrombotic and hemorrhagic events serving as the primary causes of disability and fatal complications during disease progression, capable of severely compromising patients’ quality of life and shortening their survival. As research continues to advance, current thrombotic risk stratification approaches for ET have become insufficient to meet clinical demands, and no universally accepted bleeding risk stratification standard has been established to date; therefore, identifying risk factors for both event types and achieving precise prediction have emerged as core priorities requiring focused attention in clinical research. Existing studies have confirmed that age, prior thrombotic history, JAK2 gene mutations, and cardiovascular risk factors demonstrate correlations with thrombotic events in ET patients, whereas for bleeding events, indicators such as age at diagnosis, leukocyte count, and low hemoglobin levels constitute relevant influencing factors. Machine learning algorithms can efficiently process massive clinical datasets, recognizing complex nonlinear relationships among various parameters; prediction models constructed through integrating clinical parameters with molecular biomarkers have already demonstrated superior predictive accuracy and reliability. Although current machine learning-based predictive models for thrombotic and hemorrhagic events in ET have achieved favorable research outcomes, they still face challenges regarding data quality, model interpretability, and clinical generalizability. Future research efforts should concentrate on algorithm optimization, multidimensional data integration, and promoting the clinical translation and application of such models, thereby providing more precise instrumental support for individualized risk stratification and therapeutic strategy formulation in ET patients.