UNASSIGNED: A comprehensive literature search was conducted across multiple databases, including PubMed, EBSCO, Web of Science Core Collection, Ovid Full-Text Medical Journal Database, ProQuest, Elsevier ClinicalKey, China National Knowledge Infrastructure, China Biology Medicine, and WanFang Database, spanning from January 1990 to July 2023. To assess the quality of the included models, the Predictive Model Risk of Bias Assessment Tool (PROBAST) was employed.
UNASSIGNED: Fourteen relevant studies were identified and included in the final review, all focusing on model development. The discrimination ability of the included models ranged from 0.725 to 0.940, indicating satisfactory prediction accuracy. However, a notable limitation was that nine of these models (64.3%) did not provide clear guidelines on the selection of potential predictors. Furthermore, only six models (42.86%) underwent internal validation, with none undergoing external validation. A high risk of bias was observed across the included models. Logistic regression, Cox regression, and machine learning were the primary methods employed in the construction of these models.
UNASSIGNED: The risk prediction models included in this review demonstrated favorable prediction accuracy. However, due to variations in construction methodologies, direct comparison of their performance is challenging. These models exhibited certain shortcomings, such as inadequate handling of missing data and a lack of internal and external validation, resulting in a high risk of bias. Therefore, it is recommended that these models be updated and externally validated. The development of prospective, multi-center studies is encouraged to construct predictive models with low risk of bias and high clinical applicability, ultimately guiding evidence-based clinical practice.
UNASSIGNED: The online version contains supplementary material available at 10.1007/s40200-023-01330-1.
■在多个数据库中进行了全面的文献检索,包括PubMed,EBSCO,WebofScience核心合集,Ovid全文医学期刊数据库,ProQuest,ElsevierClinicalKey,中国国家知识基础设施,中国生物医学,和万方数据库,从1990年1月到2023年7月。为了评估包含的模型的质量,采用偏倚风险预测模型评估工具(PROBAST)。
■确定了14项相关研究,并将其纳入最终审查,都专注于模型开发。所包含模型的辨别能力范围为0.725至0.940,表明令人满意的预测准确性。然而,一个显著的局限性是,这些模型中有9个(64.3%)没有为潜在预测因子的选择提供明确的指南.此外,只有六个模型(42.86%)进行了内部验证,没有正在进行外部验证。在所包括的模型中观察到偏差的高风险。Logistic回归,Cox回归,机器学习是构建这些模型的主要方法。
■本综述中包含的风险预测模型显示出良好的预测准确性。然而,由于施工方法的变化,直接比较他们的表现是具有挑战性的。这些模型表现出某些缺点,例如对缺失数据的处理不当以及缺乏内部和外部验证,导致偏见的高风险。因此,建议更新这些模型并进行外部验证。前瞻性的发展,鼓励多中心研究构建低偏倚风险和高临床适用性的预测模型,最终指导循证临床实践。
■在线版本包含补充材料,可在10.1007/s40200-023-01330-1获得。