Krogh Jesper, Hjorthøj Carsten Rygaard, Jakobsen Janus Christian, Lindschou Jane, Kessing Lars Vedel, Nordentoft Merete, Gluud Christian
Mental Health Centre Copenhagen, University of Copenhagen, Copenhagen, Denmark.
Mental Health Centre Copenhagen, University of Copenhagen, Copenhagen, Denmark.
J Affect Disord. 2015 Jul 1;179:121-7. doi: 10.1016/j.jad.2015.03.042. Epub 2015 Apr 1.
Systematic errors in randomized clinical trials (RCTs) overestimate treatment effects. We systematically assessed the risks of bias in RCTs assessing the effects of drug and non-drug interventions for patients with unipolar depression.
We searched bibliographic databases for drug and non-drug RCTs including patients with depression. We assessed the following risk of bias domains: sequence generation, allocation concealment, baseline imbalance, blinding, intention-to-treat analysis, selective outcome reporting, and funding. Risks of bias were compared for drug and non-drug trials and according to year of publication (before 1990; from 1990 to 1999; and 2000 to 2010).
Comparing drug trials (N=775) to non-drug trials (N=73), the proportion of drug trials with low risk of bias seemed superior regarding blinding of participants (p<0.001), blinding of health-care providers (p<0.001), and blinded outcome assessment (p<0.001). Non-drug trials were superior regarding sequence generation (p<0.001), allocation concealment (p=0.002), intention-to-treat analysis (p<0.001), and baseline imbalance (p=0.006). Adequate blinding of data managers (p=0.45), blinding of statisticians (p=0.69), and selective outcome reporting (p=0.55) did not differ. 41.5% of drug trials were funded by for-profit organizations compared to 12.3% of non-drug trials (p<0.001). In drug trials, the risk of bias decreased significantly over time. This did not reach statistical significance in non-drug trials.
This study only included trials published before 2010.
Included trials were associated with high risks of bias which may distort effect estimates. The risks of bias decreased with time for drug trials.
随机临床试验（RCT）中的系统误差会高估治疗效果。我们系统评估了评估单相抑郁症患者药物和非药物干预效果的RCT中的偏倚风险。
我们在文献数据库中搜索了包括抑郁症患者的药物和非药物RCT。我们评估了以下偏倚风险领域：序列产生、分配隐藏、基线不平衡、盲法、意向性分析、选择性结果报告和资金来源。比较了药物试验和非药物试验以及根据发表年份（1990年之前；1990年至1999年；以及2000年至2010年）的偏倚风险。
将药物试验（N = 775）与非药物试验（N = 73）进行比较，在参与者盲法（p < 0.001）、医疗保健提供者盲法（p < 0.001）和盲法结果评估（p < 0.001）方面，偏倚风险低的药物试验比例似乎更高。在序列产生（p < 0.001）、分配隐藏（p = 0.002）、意向性分析（p < 0.001）和基线不平衡（p = 0.006）方面，非药物试验更优。数据管理人员的充分盲法（p = 0.45）、统计学家的盲法（p = 0.69）和选择性结果报告（p = 0.55）没有差异。41.5%的药物试验由营利性组织资助，而非药物试验为12.3%（p < 0.001）。在药物试验中，偏倚风险随时间显著降低。在非药物试验中这未达到统计学显著性。
本研究仅纳入了2010年之前发表的试验。
纳入的试验存在较高的偏倚风险，这可能会扭曲效应估计。药物试验的偏倚风险随时间降低。
