模拟研究显示线性回归与Mann-Whitney检验可用于分析随机对照试验中30天内居家存活天数结局
该研究从统计学角度审视30天内居家存活天数(DAH30)的属性,并基于经验数据开展1:1分配的随机对照试验模拟,评估简单(未校正)线性回归与Mann-Whitney检验的分析表现。结果显示,线性回归可提供未校正平均治疗效应的有效估计,Mann-Whitney检验在某些情形下检验效能更高。
要点
- 线性回归可有效估计DAH30的未校正平均治疗效应,零值膨胀相对较低时适用。
- 在不同加性治疗效应、样本量和分布情形下,线性回归在偏倚、方差估计、无效应时拒绝率和95%置信区间覆盖方面表现有效。
- 小效应和小样本情形下,Mann-Whitney检验检测组间差异的能力优于线性回归。
核心结论
- 简单线性回归是DAH30的合理分析选择,用于估计随机对照试验队列中的平均治疗效应(即意向性治疗或「治疗策略」估计目标),适用于零值膨胀相对较低的情形。
- 在某些情形下(小效应和较小样本量),Mann-Whitney检验具有更好的检测组间差异能力(同等条件下比较)。
研究设计与样本
- 研究审视DAH30的统计属性,并基于经验数据开展模拟,模拟采用随机对照试验中1:1分配,涵盖不同的加性与实际治疗效应、样本量、分布及不同中心位置和零值水平。
主要发现
- 线性回归在加性治疗效应下提供未校正平均治疗效应的有效估计,这一点在偏倚、方差估计、无效应时的拒绝率以及真实实际效应95%置信区间的覆盖方面得到确认。
- 在某些情形下,Mann-Whitney检验的检验效能大于线性回归。
临床怎么用
- 原文未据此给出推荐。
局限性
- 原文未报告局限性。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
BACKGROUND AND OBJECTIVES: The aims of the work were to consider the properties of the Days Alive and at Home by Day 30 (DAH30) from a statistical perspective, and to conduct a simulation study exploring the use of simple (unadjusted) linear regression and Mann-Whitney test as the method of analysis reflect realized analysis options. STUDY DESIGN AND SETTING: The Days Alive and at Home by Day 30 (DAH30) has been proposed a patient-centric outcome, and clinically relevant outcome suitable for clinical trials. It has unusual statistical properties, and suitability of standard statistical analysis methods is unclear. The properties of DAH30 were reviewed. Simulations based upon 1:1 allocation in an randomized controlled trial (RCT) based upon empirical data were conducted reflecting different additive and realized (reflecting the DAH30) treatment effects, sample sizes and distributions with varying and central locations and zero value level. A variety of metrics were used to assess performance (including bias, coverage, and rejection rate). RESULTS: Linear regression provided a valid estimate of the unadjusted average treatment effect with an additive treatment. This was confirmed in terms of bias, estimation of variance, rejection rate in the absence of an effect, and coverage of the 95% confidence interval for the true realized effect. Mann-Whitney provided greater (power) than linear regression in some situations. CONCLUSION: Simple linear regression is a reasonable analytic option for the DAH30 for estimating the average treatment effect in the RCT cohort (ie, an intention to treat, or "treatment policy" estimand) where zero-inflation is relatively low. Mann-Whitney test in some circumstances (small effects and smaller samples sizes) provides better ability (like for like) to detect a difference between the groups. PLAIN LANGUAGE SUMMARY: Simple linear regression can be used to analyze DAH30 outcome in a randomized trial for a range of scenarios which were considered in this study (including relatively proportions of zero values). The DAH30's properties affect the treatment effect than can be estimated. Mann-Whitney test offered better ability to detect a difference of a smaller magnitude for smaller samples sizes.
引用
- Cook JA. A simulation study showed that linear regression and Mann-Whitney test can be used to analyze the Days Alive and at Home by Day 30 outcome in a randomized controlled trial. J Clin Epidemiol. 2025. PMID: 39826625. doi:10.1016/j.jclinepi.2025.111674
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。