基于XGBoost算法构建的ICU死亡风险预测模型的系统评价

Systematic Review of the Research Progress of ICU Mortality Risk Prediction Model Based on XGBoost Algorithm

目的系统评价基于极端梯度提升(eXtreme Gradient Boosting,XGBoost)算法构建的重症加强护理病房(Intensive Care Unit,ICU)死亡风险预测模型的研究现况。方法检索知网、万方、维普、PubMed、Embase、Web of Science、Scopus数据库,搜集有关基于XGBoost算法构建的ICU死亡风险预测模型的研究,检索时限均为建库至2023年2月18日。由2名研究者独立筛选文献,提取资料并评价纳入研究的偏倚风险后,进行定性系统评价。结果共纳入12篇文献,纳入模型的受试者工作特征曲线下面积为0.750~0.941。10篇文献适用性较好,其余2篇文献适用性不清楚。12篇文献均存在高偏倚风险,偏倚主要来自于不合适的研究数据来源、研究对象的纳排标准不清晰、预测因子定义与评估不一致、基于单因素分析法筛选预测因子、缺乏完善的模型性能评估等。结论现有基于XGBoost算法构建的ICU死亡风险预测模型具有较好的区分度,但其临床预测的准确性还尚不明确。未来还需进一步完善相关研究设计,避免研究中的各类偏倚风险,加强模型的外部验证,确保模型在临床实践中的可行性及有效性。

Objective To systematically evaluate the current research status of the intensive care unit(ICU)mortality risk prediction model based on eXtreme gradient boosting(XGBoost).Methods A search of the CNKI,Wanfang,Weipu,PubMed,Embase,Web of Science,and Scopus databases was conducted to collect studies on ICU mortality risk prediction models based on XGBoost,and the search time was from the establishment of the database to February 18,2023.Two researchers independently screened the literature,extracted the data and evaluated the risk of bias in the included studies,and carried out a qualitative systematic review.Results A total of 12 papers were included,and the area under the receiver operating characteristic curve of the included model was 0.750-0.941.The applicability of 10 papers was good,and the applicability of the other 2 papers was not clear.All the 12 papers had high risk of bias,which mainly came from inappropriate research data sources,unclear exclusion criteria of research objects,inconsistent definition and evaluation of predictors,screening of predictors based on single factor analysis,lack of perfect model performance evaluation,etc.Conclusion The existing ICU mortality risk prediction model constructed based on the XGBoost algorithm has a good degree of differentiation,but the accuracy of its clinical prediction is still unclear.In the future,it is necessary to further improve the design of the relevant studies,avoid the risk of various types of bias in the study,strengthen the external validation of the model,and ensure the feasibility and effectiveness of the model in clinical practice.