阶梯楔形整群随机试验的亚组分析:2016至2023年统计实践与报告的系统评价
该系统评价纳入2016至2023年数据库中的211篇已完成的健康相关阶梯楔形整群随机试验(SW-CRT)主要报告,评估亚组分析的流行程度、变量特征、预设与论证依从性、统计方法及报告细节。结果显示47%的试验报告至少1项亚组分析,但仅23%清晰预设所有亚组分析,交互检验、功效计算和多重性校正均少见,提示亚组分析常欠规范且报告不一致。
要点
- 47%的SW-CRT报告亚组分析,但仅23%清晰预设所有亚组分析。
- 仅35%报告与治疗指示变量的交互检验结果,功效计算仅4%。
- 多重性校正仅2%,提示亚组分析结果解释需谨慎。
核心结论
- 亚组分析在SW-CRT中常见,但常欠规范、分析不一致且报告不充分,交互检验有限、功效论证少且多重性校正罕见。
研究设计与样本
- 系统评价2016至2023年数据库中已完成的健康相关SW-CRT主要报告,纳入至少2个序列、3个周期且随机化至少5个整群的试验,由两名独立评价者提取信息并达成共识。
主要发现
- 211篇合格报告中99篇(47%)报告至少1项亚组分析,每项试验亚组分析中位数为4(Q1-Q3:2-8;范围:1-81)。
- 仅23篇(23%)清晰预设所有呈现的亚组分析,28篇(28%)为所有亚组分析提供理由,53篇(54%)未提供任何理由。
- 分层分析常见(n=79,80%),49篇(50%)报告使用交互检验,但仅35篇(35%)报告与治疗指示变量的交互检验结果。
- 亚组分析的功效计算罕见(n=4,4%),多重性校正不常见(n=2,2%)。
- 69篇(70%)试验观察到至少1个具有统计学意义的层特异性治疗效果。
临床怎么用
- 在SW-CRT中规划亚组分析时,应清晰预设并论证所有亚组分析,报告与治疗指示变量的交互检验结果,并考虑多重性校正和功效计算。
局限性
- 本系统评价仅纳入2016至2023年数据库中的已完成SW-CRT主要报告,可能未覆盖更早或未发表的研究;原文未报告其他局限性。
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
OBJECTIVES: Subgroup analyses can inform whether intervention effects differ systematically across participants, but they are prone to false-positive findings, low power, and selective reporting. Stepped-wedge cluster randomized trials (SW-CRTs) introduce additional inferential complexities (eg, time effects, clustering, and complex interactions) yet subgroup-analysis practice in SW-CRTs has not been systematically evaluated. Our objectives were to describe (1) the prevalence and extent of subgroup analyses, (2) characteristics of the subgroup variables of interest, (3) adherence to key recommendations such as prespecification and justification, (4) statistical approaches used, and (5) details of reporting results. STUDY DESIGN AND SETTING: We reviewed primary reports of completed health-related SW-CRTs from a 2016-2023 database to capture statistical and reporting practice. Trials were eligible if they had at least 2 sequences, 3 periods, and randomized at least 5 clusters. Two independent reviewers extracted information from each trial and reached consensus through discussion. Results were summarized using median (interquartile ranges) and frequencies (%). RESULTS: Of 211 eligible reports, 99 (47%) presented results of at least 1 subgroup analysis; the median number of subgroup analyses (unique outcome by covariate combinations) per trial was 4 (Q1-Q3:2-8; range:1-81). Only 23 (23%) clearly prespecified all presented subgroup analyses. A rationale for all subgroup analyses was provided in 28 (28%), while 53 (54%) provided none. Stratified analyses were common (n = 79, 80%); the use of interaction testing was reported in 49 (50%), but only 35 (35%) reported an interaction test result with the treatment indicator. Power calculations for subgroup analyses were rare (n = 4, 4%), and multiplicity corrections were uncommon (n = 2, 2%). At least 1 statistically significant stratum-specific treatment effect was observed in 69 (70%) trials. CONCLUSION: Subgroup analyses are common in SW-CRTs but are frequently underspecified and inconsistently analyzed and reported, with limited interaction testing, minimal power justification, and rare adjustment for multiple testing. These patterns are broadly consistent with concerns identified in prior subgroup-analysis reviews of randomized trials, suggesting that longstanding vulnerabilities persist in SW-CRTs and may be compounded by design complexity. More explicit, design-aware guidance for planning, analysis, and reporting of subgroup investigations in SW-CRTs may improve the credibility and interpretability of subgroup findings.
引用
- Tong G, Li C, Wang H, et al. Subgroup analyses in stepped-wedge cluster randomized trials: a systematic review of statistical practice and reporting from 2016 to 2023. J Clin Epidemiol. 2026. PMID: 42753990. doi:10.1016/j.jclinepi.2026.112512
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。