Florida State University, Tallahassee, USA.
Merck Research Laboratory, Rahway, NJ, USA.
Lifetime Data Anal. 2022 Oct;28(4):723-743. doi: 10.1007/s10985-022-09570-8. Epub 2022 Aug 7.
Genitourinary surgeons and oncologists are particularly interested in whether a robotic surgery improves times to Prostate Specific Antigen (PSA) recurrence compared to a non-robotic surgery for removing the cancerous prostate. Time to PSA recurrence is an example of a survival time that is typically interval-censored between two consecutive clinical inspections with opposite test results. In addition, success of medical devices and technologies often depends on factors such as experience and skill level of the medical service providers, thus leading to clustering of these survival times. For analyzing the effects of surgery types and other covariates on median of clustered interval-censored time to post-surgery PSA recurrence, we present three competing novel models and associated frequentist and Bayesian analyses. The first model is based on a transform-both-sides of survival time with Gaussian random effects to account for the within-cluster association. Our second model assumes an approximate marginal Laplace distribution for the transformed log-survival times with a Gaussian copula to accommodate clustering. Our third model is a special case of the second model with Laplace distribution for the marginal log-survival times and Gaussian copula for the within-cluster association. Simulation studies establish the second model to be highly robust against extreme observations while estimating median regression coefficients. We provide a comprehensive comparison among these three competing models based on the model properties and the computational ease of their Frequentist and Bayesian analysis. We also illustrate the practical implementations and uses of these methods via analysis of a simulated clustered interval-censored data-set similar in design to a post-surgery PSA recurrence study.
泌尿科医生和肿瘤学家特别关注机器人手术与非机器人手术相比是否能提高前列腺特异性抗原 (PSA) 复发的时间，以去除癌变的前列腺。PSA 复发时间是生存时间的一个例子，通常在两次连续的临床检查之间以相反的检测结果进行区间截断。此外，医疗器械和技术的成功通常取决于医疗服务提供者的经验和技能水平等因素，从而导致这些生存时间的聚类。为了分析手术类型和其他协变量对手术后 PSA 复发时间的集中区间截断中位数的影响，我们提出了三种具有竞争力的新型模型以及相关的频率分析和贝叶斯分析。第一种模型基于生存时间的双侧变换，并带有高斯随机效应，以解释簇内关联。我们的第二种模型假设转换后的对数生存时间具有近似的边缘拉普拉斯分布，并带有高斯 Copula 以适应聚类。我们的第三种模型是第二种模型的特例，其中边缘对数生存时间具有拉普拉斯分布，簇内关联具有高斯 Copula。模拟研究表明，第二种模型在估计中位数回归系数时对极端观测值具有高度稳健性。我们基于模型特性和频率分析与贝叶斯分析的计算简便性，对这三种竞争模型进行了全面比较。我们还通过分析类似于手术后 PSA 复发研究的模拟集中区间截断数据集，说明了这些方法的实际实施和用途。
