压力性损伤风险预测机器学习模型的系统评价

Machine learning models for risk prediction of Pressure Injury:a systematic review

目的 :系统评价压力性损伤风险预测的机器学习模型。方法 :检索数据库中有关压力性损伤风险预测机器学习模型的研究,检索时限为建库至2022年3月1日。2名研究者独立筛选文献、提取数据,并应用PROBAST分析文献质量。结果 :共纳入17篇文献,包括4项开发模型,13项开发和验证模型。受试者工作特征曲线下面积为0.790~0.897。研究总体适用性较好,但存在一定偏倚,主要是因为未采取或未报告盲法、未报告缺失数据处理方法或处理不当、样本量不足、自变量处理不恰当、未考虑模型性能及拟合。结论 :压力性损伤风险预测机器学习模型的开发尚处于发展阶段,外推性有待进一步探讨,未来应关注研究设计和临床数据的处理,开发适用于中国人群的模型。

Objective:To systematically review the machine learning models for risk prediction of Pressure Injury(PI).Methods:A search was performed on the machine learning models for PI risk prediction in PubMed,Embase,Cochrane Library,CBM,CNKI and Wanfang databases from the establishment of the database to March 1,2022.Two researchers independently screened literature,extracted data and evaluated the quality of the included studies based on PROBAST.Results:Totally 17 studies were included,including 4 development studies and 13 development and validation studies.The area under the subject operating characteristic curve ranged from 0.790 to 0.897.The overall applicability of the study was good,but there was a certain bias,mainly because blind method was not adopted or reported,the missing data processing method was not reported or improper,sample size was insufficient,independent variable processing was inappropriate,and model performance and fitting situation were not considered.Conclusion:The development of PI risk prediction machine learning models is still in the developing stage,and the extrapolation needs to be further discussed.In the future,attention should be paid to study design and clinical data processing,and the development of models suitable for the Chinese population.