School of Physiotherapy, Royal College of Surgeons in Ireland, Dublin, Ireland.
Department of Mathematics and Statistics, College of Science and Engineering, University of Limerick, Limerick, Ireland.
J Epidemiol Community Health. 2016 May;70(5):513-9. doi: 10.1136/jech-2015-206475. Epub 2016 Jan 14.
Falls are a significant cause of morbidity after stroke. The aim of this review was to identify, critically appraise and summarise risk prediction models for the occurrence of falling after stroke.
A systematic literature search was conducted in December 2014 and repeated in June 2015. Studies that used multivariable analysis to build risk prediction models for falls early after stroke were included. 2 reviewers independently assessed methodological quality. Data relating to model calibration, discrimination (C-statistic) and clinical utility (sensitivity and specificity) were extracted. A narrative review of models was conducted. PROSPERO reference: CRD42014015612.
The 12 included articles presented 18 risk prediction models. 7 studies predicted falls among inpatients only and 5 recorded falls in the community. Methodological quality was variable. A C-statistic was reported for 7 models and values ranged from 0.62 to 0.87. Models for use in the inpatient setting most frequently included measures of hemi-inattention, while those predicting community events included falls (or near-falls) history and balance measures most commonly. Only 2 studies reported any form of validation, and none presented a validated model with acceptable performance.
A number of falls-risk prediction models have been developed for use in the acute and subacute stages of stroke. Future research should focus on validating and improving existing models, with reference to the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) guidelines to ensure quality reporting and expedite clinical implementation.
跌倒在卒中后发病率中占很大比例。本综述的目的是识别、批判性评估和总结卒中后发生跌倒的风险预测模型。
2014 年 12 月进行了系统的文献检索，并于 2015 年 6 月重复检索。纳入了使用多变量分析构建卒中后早期跌倒风险预测模型的研究。2 位评审员独立评估方法学质量。提取与模型校准、区分度（C 统计量）和临床实用性（敏感性和特异性）有关的数据。对模型进行叙述性综述。PROSPERO 参考：CRD42014015612。
12 篇纳入文献提出了 18 个风险预测模型。7 项研究预测住院患者跌倒，5 项记录社区跌倒。方法学质量参差不齐。有 7 个模型报告了 C 统计量，范围从 0.62 到 0.87。用于住院患者的模型最常包括偏侧忽视的指标，而预测社区事件的模型最常包括跌倒（或接近跌倒）史和平衡指标。只有 2 项研究报告了任何形式的验证，没有一项提出了具有可接受性能的验证模型。
已经开发了许多用于卒中急性期和亚急性期的跌倒风险预测模型。未来的研究应集中于验证和改进现有模型，参考用于个体预后或诊断的多变量预测模型的透明报告（TRIPOD）指南，以确保高质量报告并加快临床实施。
