Kamerman Peter R, Vollert Jan
Brain Function Research Group, School of Physiology, Faculty of Health Sciences, University of the Witwatersrand, South Africa.
Pain Research, Department of Surgery and Cancer (MSK), Imperial College London, London, United Kingdom.
Pain. 2022 Jun 1;163(6):e748-e758. doi: 10.1097/j.pain.0000000000002468. Epub 2021 Aug 27.
We modelled the effects of pain intensity inclusion thresholds (3/10, 4/10, and 5/10 on a 0- to 10-point numerical pain rating scale) on the magnitude of the regression to the mean effect under conditions that were consistent with the sample mean and variance, and intermeasurement correlation observed in clinical trials for the management of chronic pain. All data were modelled on a hypothetical placebo control group. We found a progressive increase in the mean pain intensity as the pain inclusion threshold increased, but this increase was not uniform, having an increasing effect on baseline measurements compared with study endpoint measurements as the threshold was increased. That is, the regression to the mean effect was magnified by increasing inclusion thresholds. Furthermore, the effect increasing pain inclusion thresholds had on the regression to the mean effect was increased by decreasing sample mean values at baseline and intermeasurement correlations, and increasing sample variance. At its smallest, the regression to the mean effect was 0.13/10 (95% confidence interval: 0.03/10-0.24/10; threshold: 3/10, baseline mean pain: 6.5/10, SD: 1.6/10, and correlation: 0.44), and at its greatest, it was 0.78/10 (95% confidence interval: 0.63/10-0.94/10; threshold: 5/10, baseline mean pain: 6/10, SD: 1.8/10, and correlation: 0.19). We have shown that using pain inclusion thresholds in clinical trials drives progressively larger regression to the mean effects. We believe that a threshold of 3/10 offers the best compromise between maintaining assay sensitivity (the goal of thresholds) and the size of the regression to the mean effect.
我们模拟了疼痛强度纳入阈值（在0至10分的数字疼痛评分量表上为3/10、4/10和5/10）对均值回归效应大小的影响，模拟条件与慢性疼痛管理临床试验中观察到的样本均值、方差和测量间相关性一致。所有数据均基于一个假设的安慰剂对照组进行建模。我们发现，随着疼痛纳入阈值的增加，平均疼痛强度逐渐增加，但这种增加并不均匀，随着阈值的增加，与研究终点测量相比，对基线测量的影响越来越大。也就是说，通过提高纳入阈值，均值回归效应会被放大。此外，通过降低基线时的样本均值和测量间相关性以及增加样本方差，提高疼痛纳入阈值对均值回归效应的影响会增大。均值回归效应最小为0.13/10（95%置信区间：0.03/10 - 0.24/10；阈值：3/10，基线平均疼痛：6.5/10，标准差：1.6/10，相关性：0.44），最大为0.78/10（95%置信区间：0.63/10 - 0.94/10；阈值：5/10，基线平均疼痛：6/10，标准差：1.8/10，相关性：0.19）。我们已经表明，在临床试验中使用疼痛纳入阈值会导致均值回归效应逐渐增大。我们认为，3/10的阈值在保持检测灵敏度（阈值的目标）和均值回归效应大小之间提供了最佳折衷。
