AI 中文总结
本研究以HIV临床试验数据对比两阶段方法与贝叶斯联合模型，发现二者对ddI与死亡率关联估计相近，但CD4保护关联估计存在差异，表明纵向生物标志物信息的处理方式会影响统计推断。
AI 中文摘要
两阶段方法和贝叶斯联合模型是常用于联合分析纵向生物标志物测量值与时间-事件结局的方法。本研究利用一项包含467例患者的HIV临床试验数据，该数据包含重复CD4测量值和全因死亡率作为生存结局，将基线两阶段方法与贝叶斯联合模型进行比较。两阶段分析拟合线性混合效应模型，并将每位患者的预测基线CD4值作为固定协变量纳入Cox比例风险模型；联合模型则同时对纵向CD4过程和生存进行建模，将死亡风险与当前潜在CD4值关联。两种方法下，ddI与死亡率的关联估计在方向和大小上相似：两阶段方法的风险比HR=1.342（95%置信区间CI：1.006-1.789），联合模型的风险比HR=1.383（95%可信区间CrI：0.952-2.010）；联合模型估计的CD4的保护关联更强（HR=0.776），高于两阶段方法的HR=0.826。由于两种方法采用了纵向CD4过程的不同汇总方式，观察到的差异不能仅归因于测量误差或信息缺失。研究结果表明，纵向生物标志物信息的处理方式会对统计推断产生实质性影响。
英文摘要
The two-stage approach and Bayesian joint modeling are commonly used to analyze longitudinal biomarker measurements together with time-to-event outcomes. Using data from an HIV clinical trial of 467 patients with repeated CD4 measurements and all-cause mortality as the survival outcome, we compared a baseline two-stage approach with a Bayesian joint model. The two-stage analysis fitted a linear mixed-effects model and included each patient's predicted baseline CD4 value as a fixed covariate in a Cox proportional hazards model. The joint model simultaneously modeled the longitudinal CD4 process and survival while linking mortality risk to the current underlying CD4 value. The estimated association between ddI and mortality was similar in direction and magnitude across the two approaches. The two-stage estimate was HR = 1.342 (95% CI: 1.006-1.789), whereas the joint-model estimate was HR = 1.383 (95% CrI: 0.952-2.010). The estimated protective association of CD4 was stronger under the joint model (HR = 0.776) than under the two-stage approach (HR = 0.826). Because the approaches used different summaries of the longitudinal CD4 process, the observed differences cannot be attributed solely to measurement error or informative dropout. The findings demonstrate that the treatment of longitudinal biomarker information can materially affect statistical inference.
Comments15 pages, 5 figures, 2 tables
