Channing Division of Network Medicine, Brigham and Women's Hospital/Harvard Medical School, 181 Longwood Avenue, Boston, MA 02115, USA.
Berman-Gund Laboratory for the Study of Retinal Degenerations, Harvard Medical School, Massachusetts Eye and Ear Infirmary, 243 Charles Street, Boston, MA 02114, USA.
Stat Med. 2018 Jul 30;37(17):2586-2598. doi: 10.1002/sim.7662. Epub 2018 May 31.
Retinitis pigmentosa is one of the most common forms of inherited retinal degeneration. The electroretinogram (ERG) can be used to determine the severity of retinitis pigmentosa-the lower the ERG amplitude, the more severe the disease is. In practice for career, lifestyle, and treatment counseling, it is of interest to predict the ERG amplitude of a patient at a future time. One approach is prediction based on the average rate of decline for individual patients. However, there is considerable variation both in initial amplitude and in rate of decline. In this article, we propose an empirical Bayes (EB) approach to incorporate the variations in initial amplitude and rate of decline for the prediction of ERG amplitude at the individual level. We applied the EB method to a collection of ERGs from 898 patients with 3 or more visits over 5 or more years of follow-up tested in the Berman-Gund Laboratory and observed that the predicted values at the last (kth) visit obtained by using the proposed method based on data for the first k-1 visits are highly correlated with the observed values at the kth visit (Spearman correlation =0.93) and have a higher correlation with the observed values than those obtained based on either the population average decline rate or those obtained based on the individual decline rate. The mean square errors for predicted values obtained by the EB method are also smaller than those predicted by the other methods.
色素性视网膜炎是最常见的遗传性视网膜变性之一。视网膜电图（ERG）可用于确定色素性视网膜炎的严重程度——ERG 幅度越低，疾病越严重。在实际的职业、生活方式和治疗咨询中，预测患者未来的 ERG 幅度很重要。一种方法是基于个体患者的平均下降率进行预测。然而，在初始幅度和下降率方面都存在相当大的差异。在本文中，我们提出了一种经验贝叶斯（EB）方法，用于在个体水平上预测 ERG 幅度，该方法考虑了初始幅度和下降率的变化。我们将 EB 方法应用于 Berman-Gund 实验室的 898 名患者的一系列 ERG 中，这些患者在 5 年以上的随访中有 3 次或更多次就诊。结果观察到，使用基于前 k-1 次就诊数据的建议方法获得的最后（k 次）就诊的预测值与 k 次就诊的观察值高度相关（Spearman 相关系数=0.93），并且与观察值的相关性高于基于人群平均下降率或基于个体下降率获得的预测值。EB 方法获得的预测值的均方误差也小于其他方法预测的值。
