甲状腺相关性眼病临床试验受试者脱落原因分析及预测模型构建

Analysis of dropping out causes of subjects in thyroid-associated ophthalmopathy clinical research and construction of prediction model

目的 回顾性分析甲状腺相关性眼病(thyroid-associated ophthalmopathy,TAO)临床研究中受试者脱落发生的原因,并建立受试者脱落预测模型,为TAO临床试验受试者管理提供依据。方法 收集2017年11月至2021年4月于上海交通大学医学院附属第九人民医院眼科参加TAO临床试验的384例受试者资料,通过Lasso回归进行变量筛选,构建Logistic回归预测模型,绘制受试者操作特征(receiver operator characteristic, ROC)曲线和校准曲线验证模型的区分度和校准度。结果 384例受试者的平均年龄为(44.55±13.25)岁,其中男性173例(45.1%)、女性211例(54.9%),有53例受试者脱落,脱落率为13.8%。受试者脱落的原因主要有入组后未治疗、未追踪到原因、拒绝随访、新冠疫情影响及电话无人接听。训练集多因素Logistic回归分析结果显示,治疗方式[OR=0.16,95%CI(0.06,0.40),P <0.001]、吸烟情况[OR=0.19,95%CI(0.03,0.78),P=0.04]、复视评分[OR=0.36,95%CI(0.19,0.61),P <0.001]、来源[OR=12.09,95%CI(3.41,48.76),P <0.001]为受试者脱落的独立预测因子。验证集中ROC曲线下面积(area under curve, AUC)为0.786,表明训练集所建模型具备较好的预测能力,同时校准曲线在验证集中表现出良好的一致性。结论 应用建立的预测模型对即将开展的TAO临床研究受试者脱落情况进行预测,重点关注脱落发生概率高的受试者,对可能导致脱落发生的问题进行预警,同时加强研究者临床研究管理培训,有助于降低受试者脱落率,提高临床研究质量。

Objective To retrospectively analyze the causes of the dropping out in subjects in thyroid-associated ophthalmopathy(TAO)clinical research,and to establish a predictive model of subject dropping out,to provide the basis for the subject management in TAO clinical trials.Methods Data of 384 subjects participating in the TAO clinical trial in the department of ophthalmology,Shanghai Ninth People’s Hospital,Shanghai Jiao Tong University School of Medicine,from November 2017 to April 2021 were collected.Lasso regression was used for variable screening and construting a logistic regression prediction model,and the differentiation and calibration of the receiver operator characteristic(ROC)curve and calibration curve validation model were drawn.Results A total of 384 subjects,of mean age(44.55±13.25)years,were 173 males(45.1%),211 females(54.9%),and 53 subjects dropped out,with a dropping out rate of 13.8%.The main reasons for subject dropping out were untreated after enrollment,untraced cause,refusal to follow-up,the impact of COVID-19,and unanswered phone calls.The results of multivariate Logistic regression analysis in the training set showed that treatment modality(OR=0.16,95%CI 0.06 to 0.40,P<0.001),smoking(OR=0.19,95%CI 0.03 to 0.78,P=0.04),diplopia score(OR=0.36,95%CI 0.19 to 0.61,P<0.001),and source(OR=12.09,95%CI 3.41 to 48.76,P<0.001)were independent predictors of subject dropping out.The area under curve(AUC)under the ROC curve in the validation set is 0.786,which indicates that the model built in the training set has good prediction ability,while the calibration curve shows good consistency in the validation set.Conclusion Applying the model established in this study to predict the dropping out of subjects in the upcoming TAO clinical reasearch,focusing on subjects with a probability of dropping out,providing warnings for problems that may lead to dropping out,and strengthening clinical research management training for researchers,can help reduce the subject dropping out rate and improve the quality of clinical research.