目的:研究针灸随机对照试验(RCTs)是否采用了合适的统计学方法。
方法:我们搜索了PubMed,以确定2010年1月至2019年12月(10年)在核心临床期刊和补充和替代医学(CAM)期刊上发表的具有连续结局的针灸RCT作为主要结局。我们比较了在核心临床期刊和CAM期刊上发表的纳入试验的统计特征。
结果:我们包括262个RCT,其中46篇发表在核心临床期刊上,216篇发表在CAM期刊上。包括RCT,只有132(50.4%)明确预定义了主要结果,72(27.5%)指定使用意向治疗或修改意向治疗人群进行初步分析。在167项试验中报告了缺失的参与者数据(MPD),118(70.7%)使用次优方法(例如,完整病例分析)处理MPD;11(6.6%)进行了关于MPD的敏感性分析。在161个重复措施设计的试验中,只有21个(13.0%)使用先进的统计模型(例如,混合效应模型),用于处理初级分析中的重复测量数据。在涉及多名针灸师的72项试验中,只有4人(5.6%)调整了针灸师变量或在分析中考虑了针灸师的聚类。核心临床期刊上的试验更有可能预先确定主要结局(78.3%vs.44.4%,P<0.001),使用多个估算来处理MPD(40%与1.5%,P<0.001),并使用统计学上先进的方法在单个时间点评估治疗效果(26.1%vs.2%,P=0.001)。
结论:统计学方法在针灸随机对照试验中的应用远不能令人满意。我们的发现强调了研究人员需要仔细使用最佳统计方法,期刊编辑需要加强统计方法的使用。
OBJECTIVE: To examine whether appropriate statistical methods were used in acupuncture randomized controlled trials (RCTs).
METHODS: We searched PubMed to identify acupuncture RCTs with continuous outcome as primary outcome published in the core clinical journals and complementary and alternative medicine (CAM) journals between January 2010 and December 2019 (10 years). We compared statistical characteristics of included trials published in core clinical journals and CAM journals.
RESULTS: We included 262 RCTs, including 46 published in core clinical journals and 216 in CAM journals. Of included RCTs, only 132 (50.4%) clearly predefined the primary outcome, 72 (27.5%) specified the use of intention to treat or modified intention to treat population for primary analysis. In the 167 trials reported missing participant data (MPD), 118 (70.7%) used suboptimal methods (e.g., complete case analysis) for dealing with MPD; 11 (6.6%) conducted sensitivity analysis regarding MPD. Among the 161 trials with repeated measures design, only 21 (13.0%) used advanced statistical models (e.g., mixed-effects models) for handling repeated-measure data in the primary analysis. In the 72 trials involving multiple acupuncturists, only 4 (5.6%) adjusted acupuncturist variable or considered the clustering by acupuncturist in analysis. Trials in core clinical journals were more likely to predefine primary outcome (78.3% vs. 44.4%, P < 0.001), use multiple imputations for handling MPD (40% vs. 1.5%, P < 0.001), and use statistically advanced methods for assessing treatment effect at a single time point (26.1% vs. 2%, P = 0.001).
CONCLUSIONS: The use of statistical methods among acupuncture RCTs is far from satisfactory. Our findings highlighted the need for researchers to carefully use the optimal statistical methods and for journal editors to strengthen the use of statistical methods.