重症患者ICU获得性吞咽障碍风险预测模型的系统评价

Systematic evaluation of a predictive model for the risk of developing ICU-acquired dysphagia in critically ill patients

目的系统评价重症患者ICU获得性吞咽障碍风险预测模型,为医疗工作者选择或开发适合的风险评估工具提供参考依据。方法检索万方数据库、中国生物医学文献数据库、维普、中国知网、PubMed、Embase、CINAHL、Cochrane Library 8个数据库中有关ICU获得性吞咽障碍风险预测模型的横断面研究、病例对照研究、队列研究,语种限定为中文和英文,检索时限为建库至2023年7月5日。由2位研究人员独立筛选文献和提取数据,并应用预测模型研究的偏倚风险评估工具分析纳入文献的偏倚风险和适用性。结果共纳入8篇高质量文献,其中包括8项ICU获得性吞咽障碍风险预测模型。模型的AUC为0.750~0.980,其中6项模型报告了校准,2项模型的Hosmer-Lemeshow检验P>0.05,表明一致性较好。2项模型进行了外部验证。多变量模型重复报告的独立预测因子有:年龄≥65岁、慢性肺疾病,脑血管疾病、合并心律失常、镇静药物的使用及使用时间、APACHEⅡ评分≥15分、是否存在鼻胃管、留置胃管时间≥72 h、气管插管时间≥72 h。样本量(结局事件)较少、对缺失数据关注不足、变量筛选过程不合理以及模型过度拟合增加了模型的偏倚风险。结论纳入的模型整体呈现良好的区分、校准性能及适用性,但其偏倚风险较高,仅有极少数研究开展了外部验证。未来研究应参照多变量预测模型的透明报告来开发、更新和验证模型,以得到高质量的ICU获得性吞咽障碍风险预测模型,为制订相关预防策略提供参考依据。

Objective To systematically evaluate the risk prediction model of ICU-acquired dysphagia in critically ill patients,and to provide a reference basis for medical practitioners to select or develop suitable risk assessment tools.Methods Cross-sectional studies,case-control studies,and cohort studies of ICU-acquired dysphagia risk prediction models were searched in eight databases,including Wanfang database,China Biomedical Literature Database,VIP database,China National Knowledge Infrastructure,PubMed,Embase,CINAHL,and the Cochrane Library in both English and Chinese,and the search timeframe was from the construction of the database to July 5th,2023.and the search was limited to Chinese and English.Literature was independently screened and data were extracted by 2 investigators,and the risk of bias assessment tool for predictive modeling studies was applied to analyze the risk of bias and applicability of the included literature.Results A total of 8 high-quality papers were included,including 8 predictive models of risk of ICU-acquired dysphagia.The area under the subject operating characteristic curve for the models ranged from 0.750 to 0.980,with 6 models reporting calibration and 2 models having a Hosmer-Lemeshow test P>0.05,indicating good agreement.2 models were externally validated.Independent predictors reported repeatedly for multivariate models were age≥65 years,chronic lung disease,cerebrovascular disease,comorbid arrhythmias,use of sedative medications and duration of use,Acute Physiology and Chronic Health EvaluationⅡscore≥15,presence of a nasogastric tube,duration of indwelling gastric tube≥72 h,and duration of endotracheal intubation≥72 h.Smaller sample sizes(outcome events),insufficient attention to missing data,variable screening processes irrationality,and model overfitting increased the risk of model bias.Conclusions The included models showed overall good discrimination and applicability,but their risk of bias was high,and external validation was carried out in only a very few studies.Future studies should refer to transparent reports of multivariate predictive models to develop,update,and validate the models to obtain high-quality predictive models of the risk of ICU-acquired dysphagia,which can provide a reference basis for the development of relevant preventive strategies.