摘要
目的 对慢性病共病患者预后预测模型进行范围综述,了解该类模型的建模方法、预测因子和预测效能,为慢性病共病患者预后评估提供参考.方法 检索中国生物医学文献数据库、中国知网、万方数据知识服务平台、维普中文科技期刊数据库、PubMed、Embase、Cochrane Library和Web of Science,收集建库至2023年11月1日发表的关于慢性病共病患者预后预测模型的文献,采用预测模型的偏倚风险评估工具进行文献质量评价,对建模方法、预测因子和预测效能等进行综述.结果 初期检索到2 130篇文献,最终纳入9篇文献,总体偏倚风险均为高风险.涉及13种模型,3种采用机器学习法建模,10种采用logistic回归法建模.4种模型的预测结局为死亡,预测因子主要为年龄、性别、体质指数(BMI)、Barthel指数和压疮;9种模型的预测结局为再入院,预测因子主要为年龄、BMI、住院次数、住院时间和住院费用.11种模型报告了受试者操作特征曲线下面积,范围为0.663～0.991 6;2种报告了一致性指数,范围为0.64～0.70.8种模型进行了内部验证;1种进行了外部验证;4种未报告验证方法.结论 本文分析的慢性病共病患者预后预测模型主要采用logistic回归和机器学习法建模,预测因子以日常护理评估指标为主,模型总体预测效能较好.
Abstract
Objective To conduct a scoping review on prognostic prediction models for patients with comorbidity of chronic diseases,and understand modeling methods,predictive factors and predictive effect of the models,so as to pro-vide the reference for prognostic evaluation on patients with comorbidity of chronic diseases.Methods Literature on prognostic prediction models for patients with comorbidity of chronic diseases was collected through SinoMed,CNKI,Wanfang Data,VIP,PubMed,Embase,Cochrane Library and Web of Science published from the time of their estab-lishment to November 1,2023.The quality of literature was assessed using prediction model risk of bias assessment tool(PROBAST),then modeling methods,predictive factors and predictive effects were reviewed.Results Totally 2 130 publications were retrieved,and nine publications were finally enrolled,with an overall high risk of bias.Thir-teen models were involved,with three established using machine learning methods and ten established using logistic re-gression.The prediction results of four models were death,with main predictive factors being age,gender,body mass index(BMI),Barthel index and pressure ulcers;the prediction results of nine models were rehospitalization,with main predictive factors being age,BMI,hospitalization frequency,duration of hospital stay and hospitalization costs.Eleven models reported the area under the receiver operating characteristic curve(AUC),ranging from 0.663 to 0.991 6;two models reported the C-index,ranging from 0.64 to 0.70.Eight models performed internal validation,one model performed external validation,and four models did not reported verification methods.Conclusions The prognostic prediction models for patients with comorbidity of chronic diseases are established by logistic regression and machine learning methods with common nursing evaluation indicators,and perform well.Laboratory indicators should be considered to add in the models to further improve the predictive effects.
基金项目
山西省社科联至重点课题研究项目(2023-2024)(SSKLZDKT2023117)
山西省教育厅研究生创新创新计划(2023)(2023SJ272)
山西省教育厅山西省高等学校一般性教学改革创新立项项目(2023)(J20230894)
山西中医药大学科技创新能力培育计划软科学研究专项(2023)(2023PY-RKX-03)
山西中医药大学研究生教育改革及创新创业项目(2023CX050)
维普
万方数据