Supplementary Material for: A Novel Strategy to Identify Placebo Responders: Prediction Index of Clinical and Biological Markers in the EMBARC Trial

Background: One in three clinical trial patients with major depressive disorder report symptomatic improvement with placebo. Strategies to mitigate the effect of placebo responses have focused on modifying study design with variable success. Identifying and excluding or controlling for individuals with a high likelihood of responding to placebo may improve clinical trial efficiency and avoid unnecessary medication trials. Methods: Participants included those assigned to the placebo arm (n = 141) of the Establishing Moderators and Biosignatures for Antidepressant Response in Clinical Care (EMBARC) trial. The elastic net was used to evaluate 283 baseline clinical, behavioral, imaging, and electrophysiological variables to identify the most robust yet parsimonious features that predicted depression severity at the end of the double-blind 8-week trial. Variables retained in at least 50% of the 100 imputed data sets were used in a Bayesian multiple linear regression model to simultaneously predict the probabilities of response and remission. Results: Lower baseline depression severity, younger age, absence of melancholic features or history of physical abuse, less anxious arousal, less anhedonia, less neuroticism, and higher average theta current density in the rostral anterior cingulate predicted a higher likelihood of improvement with placebo. The Bayesian model predicted remission and response with an actionable degree of accuracy (both AUC > 0.73). An interactive calculator was developed predicting the likelihood of placebo response at the individual level. Conclusion: Easy-to-measure clinical, behavioral, and electrophysiological assessments can be used to identify placebo responders with a high degree of accuracy. Development of this calculator based on these findings can be used to identify potential placebo responders.背景：每三名重度抑郁症（major depressive disorder）临床试验受试者中，就有一人报告接受安慰剂应答（placebo response）治疗后症状得到改善。为缓解安慰剂应答带来的干扰效应，学界已探索多种调整试验设计的策略，但收效参差不齐。甄别并排除或管控高安慰剂应答风险的受试者，或可提升临床试验效率，避免不必要的药物试验。方法：本研究纳入「临床护理中抗抑郁应答调节因子与生物标志物确立（Establishing Moderators and Biosignatures for Antidepressant Response in Clinical Care, EMBARC）」试验安慰剂组（n=141）的受试者。采用弹性网（elastic net）算法，对283项基线临床、行为、影像学及电生理指标进行评估，以筛选出可预测双盲8周试验结束时抑郁症严重程度的最优且简约的特征。对在100份插补数据集（imputed data sets）的至少50%中保留的变量，采用贝叶斯多元线性回归模型（Bayesian multiple linear regression model），同时预测应答与缓解的发生概率。结果：基线抑郁症严重程度更低、年龄更轻、无忧郁特征（melancholic features）或躯体虐待（physical abuse）史、焦虑唤醒（anxious arousal）程度更低、快感缺失（anhedonia）程度更低、神经质（neuroticism）水平更低，以及喙前扣带回（rostral anterior cingulate）的平均θ电流密度（theta current density）更高的受试者，安慰剂治疗后症状改善的可能性更高。该贝叶斯多元线性回归模型对缓解与应答的预测精度具备临床实用性（曲线下面积（Area Under the Receiver Operating Characteristic Curve, AUC）均>0.73）。本研究据此开发了一款可在个体层面预测安慰剂应答概率的交互式计算器。结论：通过易于实施的临床、行为及电生理评估，可高精度甄别安慰剂应答者。基于本研究结果开发的这款计算器，可用于识别潜在的安慰剂应答者。