Drozdowska Bogna A, McGill Kris, McKay Michael, Bartlam Roisin, Langhorne Peter, Quinn Terence J
Institute of Cardiovascular and Medical Sciences, University of Glasgow, UK.
School of Medicine, Dentistry & Nursing, University of Glasgow, UK.
Eur Stroke J. 2021 Mar;6(1):18-27. doi: 10.1177/2396987321997045. Epub 2021 Feb 23.
Stroke survivors are at high risk of developing cognitive syndromes, such as delirium and dementia. Accurate prediction of future cognitive outcomes may aid timely diagnosis, intervention planning, and stratification in clinical trials. We aimed to identify, describe and appraise existing multivariable prognostic rules for prediction of post-stroke cognitive status.
We systematically searched four electronic databases from inception to November 2019 for publications describing a method to estimate individual probability of developing a cognitive syndrome following stroke. We extracted data from selected studies using a pre-specified proforma and applied the Prediction model Risk Of Bias Assessment Tool (PROBAST) for critical appraisal.
Of 17,390 titles, we included 10 studies (3143 participants), presenting the development of 11 prognostic rules - 7 for post-stroke cognitive impairment and 4 for delirium. Most commonly incorporated predictors were: demographics, imaging findings, stroke type and symptom severity. Among studies assessing predictive discrimination, the area under the receiver operating characteristic (AUROC) in apparent validation ranged from 0.80 to 0.91. The overall risk of bias for each study was high. Only one prognostic rule had been externally validated. Research into the prognosis of cognitive outcomes following stroke is an expanding field, still at its early stages. Recommending use of specific prognostic rules is limited by the high risk of bias in all identified studies, and lack of supporting evidence from external validation. To ensure the quality of future research, investigators should adhere to current, endorsed best practice guidelines for conduct of prediction model studies.
中风幸存者发生认知综合征（如谵妄和痴呆）的风险很高。准确预测未来的认知结果可能有助于及时诊断、干预计划制定以及临床试验中的分层。我们旨在识别、描述和评估现有的用于预测中风后认知状态的多变量预后规则。
我们系统地检索了四个电子数据库，从建库至2019年11月，以查找描述估计中风后发生认知综合征个体概率方法的出版物。我们使用预先指定的表格从选定的研究中提取数据，并应用预测模型偏倚风险评估工具（PROBAST）进行批判性评价。
在17390个标题中，我们纳入了10项研究（3143名参与者），呈现了11条预后规则的制定情况——7条用于中风后认知障碍，4条用于谵妄。最常纳入的预测因素有：人口统计学特征、影像学检查结果、中风类型和症状严重程度。在评估预测辨别力的研究中，表观验证中受试者操作特征曲线下面积（AUROC）范围为0.80至0.91。每项研究的总体偏倚风险都很高。只有一条预后规则得到了外部验证。中风后认知结果的预后研究是一个不断扩展的领域，仍处于早期阶段。推荐使用特定的预后规则受到所有已识别研究中高偏倚风险以及缺乏外部验证支持证据的限制。为确保未来研究的质量，研究者应遵循当前认可的预测模型研究实施最佳实践指南。
