Department of Orthopaedic Surgery, University of Michigan, Ann Arbor, MI, USA; Department of Epidemiology, School of Public Health, University of Michigan, Ann Arbor, MI, USA.
Department of Community Health & Epidemiology, Dalhousie University, Halifax, Nova Scotia, Canada.
J Clin Epidemiol. 2019 Dec;116:36-38. doi: 10.1016/j.jclinepi.2019.07.012. Epub 2019 Jul 30.
The objective was to assess the potential for biased treatment effects associated with patient-reported outcome measures (PROMs) of varying psychometric quality in randomized clinical trials (RCTs) for rotator cuff disease (RCD).
We searched for RCTs published in the past 5 years (January 2011 to December 2016) in the top five 2015 impact factor orthopedic journals. We accepted RCTs including human participants with RCD, published in English, and using PROMs specific to RCD. We extracted data on study design, sample size, risk of bias for RCTs, quality of PROM used, estimates of effect, and associated measures of variance. PROMs were given numerical ratings of psychometric quality from a prior publication. Continuous measures of effect were transformed by dividing the effect estimate by the standard deviation. Multilevel linear regression analyses were performed to determine whether PROM quality was associated with the magnitude of effect.
Overall, we included 72 RCTs reporting 174 separate outcomes. Mean sample size was 66.8 (95% CI 62.30 to 71.27), mean risk of bias score across all studies was 7.00/10 (95% CI 6.72 to 7.29), psychometric quality summary scores ranged from -2 to 10, and the standardized mean effect estimate was 0.47 (95% CI -0.17 to 1.11). Regression revealed that higher-quality PROMs had smaller estimates of effect (β = -0.32; 95% CI -0.51 to -0.13; P = 0.001). We also found that a longer follow-up period predicted slightly increased effect estimates (β = 0.08; 95% CI 0.02 to 0.13; P = 0.007).
PROMs with poor or unknown psychometric properties overestimate treatment effects in clinical research of RCD by 68.4% (β -0.32/standardized mean effect 0.47). To our knowledge, this is the first empirical evidence that variations in the quality of PROMs bias treatment effect estimates. Researchers and clinicians using data from PROMs must be cautious to explore the quality of that measure so as to not mislead decision-making resulting from biased outcomes.
评估与肩袖疾病（RCD）相关的患者报告结局测量（PROM）的心理测量质量不同的随机临床试验（RCT）中偏倚治疗效果的可能性。
我们在过去 5 年（2011 年 1 月至 2016 年 12 月）的五大顶级影响因子骨科杂志中搜索了发表的 RCT。我们接受了包含 RCD 人类参与者的 RCT，以英文发表，并使用了特定于 RCD 的 PROM。我们提取了研究设计、样本量、RCT 的偏倚风险、使用的 PROM 质量、效应估计值以及相关方差测量数据。先前发表的出版物对 PROM 进行了心理测量质量的数值评分。通过将效应估计值除以标准差来转换连续效应测量值。进行了多水平线性回归分析，以确定 PROM 质量是否与效应大小相关。
总体而言，我们纳入了 72 项 RCT，报告了 174 项独立结果。平均样本量为 66.8（95%置信区间 62.30 至 71.27），所有研究的平均偏倚风险评分均为 7.00/10（95%置信区间 6.72 至 7.29），心理测量质量综合评分范围为-2 至 10，标准化平均效应估计值为 0.47（95%置信区间-0.17 至 1.11）。回归显示，更高质量的 PROM 具有更小的效应估计值（β=-0.32；95%置信区间-0.51 至-0.13；P=0.001）。我们还发现，较长的随访期预测略微增加的效应估计值（β=0.08；95%置信区间 0.02 至 0.13；P=0.007）。
在 RCD 的临床研究中，具有较差或未知心理测量特性的 PROM 会高估治疗效果，高出 68.4%（β-0.32/标准化平均效应 0.47）。据我们所知，这是第一个实证证据，表明 PROM 质量的变化会影响治疗效果估计。使用 PROM 数据的研究人员和临床医生必须谨慎探索该措施的质量，以免因结果偏差而误导决策。
