Effect of intraocular pressure on the Bayesian estimation of rates of visual field progression in glaucoma.
Effect of intraocular pressure on the Bayesian estimation of rates of visual field progression in glaucoma.
复制标题
眼压对青光眼视野进展率贝叶斯估计的影响。
DOI：
10.1167/iovs.13-12348
复制
发表时间：
2013
影响因子：
4.4
通讯作者：
Medeiros,FelipeA
中科院分区：
文献类型：
--
作者：
Medeiros,FelipeA
我饶有兴趣地阅读了安德森和约翰逊的研究，该研究利用人口信息通过贝叶斯分析来修改对青光眼视野进展速率的估计。 1 然而，我很惊讶地发现作者忽略了之前关于这个主题的工作，这实际上是第一个调查这个问题的工作。 2, 3 使用来自 250 名青光眼患者 352 只眼睛的大型队列的真实数据，我和我的同事表明，使用贝叶斯分析将风险因素纳入视野变化率的估计中，比普通最小二乘法有显着改善。 2 我们表明，在随访期间结合有关 IOP 的信息以及角膜厚度和是否存在进行性椎间盘损伤，可以得到更准确和精确的斜率，并更好地预测未来标准自动视野检查平均偏差 (MD) 值。与我们的结果相反，安德森和约翰逊得出的结论是，不考虑眼压信息并不会改变视野进展贝叶斯估计器的性能。 1 他们还认为，由于眼压缺乏显着影响，考虑其他风险因素的意义甚至更小。 1 然而，对于安德森和约翰逊提出的结果缺乏意义有一些简单的解释，它们主要与他们的研究中使用的先验的定义和表征不佳有关。作者评估了代表治疗与未治疗青光眼人群变化斜率分布的两个不同先验是否会不同地影响视野进展速率的估计。 1 然而，他们的研究中使用的先验数据源自先前发表的来自两个截然不同人群的数据，这些数据是加拿大青光眼研究（治疗组）和早期明显青光眼试验（EMGT，未治疗组）的一部分。 4, 5 然而，这种方法有很大的局限性。从随机接受治疗与不接受治疗的同质人群中获得两个分布会更合适。通过使用来自不同地理区域和临床环境的截然不同的人群，作者忽略了潜在的混杂因素，这些因素对于确定这些人群的变化率分布可能很重要。更重要的是，简单地使用两种先前的变化率分布（分类为“已治疗”与“未治疗”）来评估 IOP 的效果在很大程度上是不合适的。这本质上忽略了眼压对青光眼进展风险的持续影响，如几项主要临床试验所示。 4-6 例如，虽然 IOP 为 12 mm Hg 的接受治疗的青光眼人群的变化率分布与 IOP 为 30 mm Hg 的未治疗人群的变化率分布有很大不同，但如果简单地将这些眼睛视为接受治疗与未接受治疗的广泛群体的一部分，这种差异将在很大程度上被忽略。在加拿大青光眼研究的治疗组中，许多眼睛的眼压与 EMGT 的未治疗组的眼压重叠。此外，在这两组中的每一组中，所包括的眼睛的眼压都存在重大差异。作为另一个例子，人们不应期望平均 IOP 为 10 mm Hg 的眼睛的变化率分布与平均 IOP 为 20 mm Hg 的眼睛的变化率分布相同，即使这些眼睛都经过一段时间的治疗。 6 本质上，Anderson 和 Johnson 使用的方法无法捕捉 IOP 对变化率估计的影响。 1 相比之下……
I read, with great interest, the study by Anderson and Johnson on the use of population information to modify estimates of rates of visual field progression in glaucoma through Bayesian analysis. 1 I was surprised, however, to verify that the authors omitted previous work on this subject, which was actually the first to investigate this issue. 2, 3 Using real data from a large cohort of 352 eyes of 250 glaucoma patients, my colleagues and myself showed that incorporating risk factors into the estimation of rates of visual field change using Bayesian analysis resulted in significant improvement over the ordinary least squares approach. 2 We showed that incorporating information about IOP during follow up, along with corneal thickness and presence/absence of progressive disc damage, resulted in more accurate and precise slopes, with better prediction of future standard automated perimetry mean deviation (MD) values. In contrast to our results, Anderson and Johnson concluded that failure to consider information on IOP did not alter the performance of a Bayesian estimator of visual field progression. 1 They also suggest that because of a lack of significant influence of IOP, efforts of considering other risk factors would have even smaller significance. 1 However, there are simple explanations for the lack of significance of the results presented by Anderson and Johnson and they are mostly related to the poor definition and characterization of the priors used in their study. The authors evaluated whether two different priors, representing the distributions of slopes of change in treated versus untreated glaucomatous populations, would differently influence the estimates of rates of visual field progression. 1 The priors used in their study, however, were derived from previously published data from two widely different populations followed as part of the Canadian Glaucoma Study (the treated group) and Early Manifest Glaucoma Trial (EMGT, the untreated group). 4, 5 This approach, however, has major limitations. It would have been more appropriate to obtain the two distributions from a homogenous population randomized to treatment versus no treatment. By using widely different populations from different geographic areas and clinical settings, the authors are ignoring potentially confounding factors that could be important in determining the distribution of rates of change in these populations. Even more importantly, to evaluate the effect of IOP by simply using two prior distributions of rates of change categorized as ‘‘treated’’versus ‘‘untreated’’is largely inappropriate. This essentially ignores the continuous effect of IOP on the risk of glaucoma progression, as shown by several major clinical trials. 4–6 For example, while the distribution of rates of change for a population of treated glaucomatous eyes with IOP of 12 mm Hg will be largely different than that of an untreated population with IOP of 30 mm Hg, such differences will be largely missed by simply considering these eyes as part of broad groups of treated versus untreated eyes. In the treated group from the Canadian Glaucoma Study there were many eyes with IOPs that overlapped with those of the untreated group from the EMGT. Additionally, within each one of these two groups there are major differences in the IOPs of the included eyes. As another example, one should not expect that the distribution of rates of change for eyes with a mean IOP of 10 mm Hg would be the same as that for eyes with a mean IOP of 20 mm Hg, even if these eyes were both treated over time. 6 In essence, the methodology used by Anderson and Johnson was not able to capture the effect of IOP on estimation of rates of change. 1 In contrast …