Shen Yanfei, Cheng Xuping, Zhang Weimin
Department of Intensive Care Unit, Dongyang People's Hospital, No. 60, Wuning West Road, Dongyang, Jinhua, 322100, Zhejiang, People's Republic of China.
Neurosurg Rev. 2019 Mar;42(1):9-14. doi: 10.1007/s10143-017-0870-8. Epub 2017 Jun 20.
Fragility of randomized controlled trials (RCTs) has been evaluated using a novel metric called fragility index (FI), which measures how many events the statistical significance of a dichotomous outcome depends on. This study aimed to evaluate the fragility of RCTs in intracranial hemorrhage. Literature search (PubMed/Embase) identified all RCTs of intracranial hemorrhage since 2006. The overall distribution of FI was evaluated. Subgroup and spearman correlation analyses were made to explore potential factors that may affect FI value. All the included RCTs were divided into two groups (positive and negative trials) according to the statistical significance of selected outcomes. Finally, 47 positive and 51 negative trials were included. Both the median FI ([2; IQR, 1-4] vs. [6; IQR, 4-9], p < 0.001) and the proportion of trials with FI ≤1 (2 vs. 18, p < 0.001) in positive trials were smaller than negative trials. In subgroup comparison within positive trials, sample size ([165; IQR, 87-200] vs. [83; IQR, 60-120], p = 0.015) and number of events ([35; IQR, 20-72] vs. [24; IQR, 11-32], p = 0.015) were higher in subgroup with FI >1 than the subgroup with FI ≤1. Weak positive correlations were found between FI and sample size and number of events. In the field of intracranial hemorrhage, trials reporting significant conclusions often depend on a small number of events. Compared to sample size, this phenomenon is more likely to be affected by statistical approach and trial methodology.
随机对照试验（RCT）的脆弱性已通过一种名为脆弱性指数（FI）的新指标进行评估，该指标衡量二分结果的统计学显著性取决于多少事件。本研究旨在评估颅内出血RCT的脆弱性。文献检索（PubMed/Embase）确定了自2006年以来所有颅内出血的RCT。评估了FI的总体分布。进行亚组和Spearman相关性分析以探索可能影响FI值的潜在因素。根据所选结果的统计学显著性，将所有纳入的RCT分为两组（阳性和阴性试验）。最后，纳入了47项阳性试验和51项阴性试验。阳性试验的FI中位数（[2；四分位间距，1 - 4] 对比 [6；四分位间距，4 - 9]，p < 0.001）和FI≤1的试验比例（2对比18，p < 0.001）均低于阴性试验。在阳性试验的亚组比较中，FI > 1亚组的样本量（[165；四分位间距，87 - 200] 对比 [83；四分位间距，60 - 120]，p = 0.015）和事件数（[35；四分位间距，20 - 72] 对比 [24；四分位间距，11 - 32]，p = 0.015）高于FI≤1亚组。FI与样本量和事件数之间存在弱正相关。在颅内出血领域，报告显著结论的试验通常依赖于少量事件。与样本量相比，这种现象更可能受统计方法和试验方法学的影响。
