Gaynor Bruce D, Amza Abdou, Gebresailassie Sintayehu, Kadri Boubacar, Nassirou Baido, Stoller Nicole E, Yu Sun N, Cuddapah Puja A, Keenan Jeremy D, Lietman Thomas M
F. I. Proctor Foundation, Department of Ophthalmology, Department of Epidemiology and Biostatistics, Institute for Global Health, University of California, San Francisco, California; Programme National de Lutte Contre la Cecité Niamey, Niger; The Carter Center, Addis Ababa, Ethiopia
F. I. Proctor Foundation, Department of Ophthalmology, Department of Epidemiology and Biostatistics, Institute for Global Health, University of California, San Francisco, California; Programme National de Lutte Contre la Cecité Niamey, Niger; The Carter Center, Addis Ababa, Ethiopia.
Am J Trop Med Hyg. 2014 Sep;91(3):577-9. doi: 10.4269/ajtmh.13-0658. Epub 2014 Jul 7.
We assessed trachoma grading agreement among field graders using photographs that included the complete spectrum of disease and compared it with cases where there was consensus among experienced graders. Trained photographers took photographs of children's conjunctiva during a clinical trial in Ethiopia. We calculated κ-agreement statistics using a complete set of 60 cases and then recalculated the κ using a consensus set where cases were limited to those cases with agreement among experienced graders. When the complete set of 60 cases was used, agreement was moderate (κ = 0.61, 95% confidence interval [95% CI] = 0.56-0.67). When the consensus set was used, agreement improved significantly (κ = 0.75, 95% CI = 0.68-0.80). The κ of the consensus set was higher than the complete set by 0.14 (95% CI = 0.12-0.16) (P < 0.001). If testing sets remove difficult-to-grade cases, agreement in trachoma grading may be higher than actually seen in population-based trachoma surveys.
我们使用涵盖疾病全谱的照片评估了现场分级人员之间沙眼分级的一致性，并将其与经验丰富的分级人员达成共识的病例进行了比较。在埃塞俄比亚的一项临床试验中，经过培训的摄影师拍摄了儿童结膜的照片。我们使用完整的60例病例计算κ一致性统计量，然后使用一个共识集重新计算κ，该共识集中的病例仅限于经验丰富的分级人员达成一致的病例。当使用完整的60例病例时，一致性为中等（κ = 0.61，95%置信区间[95%CI] = 0.56 - 0.67）。当使用共识集时，一致性显著提高（κ = 0.75，95%CI = 0.68 - 0.80）。共识集的κ比完整集高0.14（95%CI = 0.12 - 0.16）（P < 0.001）。如果测试集去除难以分级的病例，沙眼分级的一致性可能高于基于人群的沙眼调查中实际观察到的一致性。
