Purchase this article with an account.
AH Kim, JY Chen, B Lengauer, WL Ottar-Pfeiffer, SI Holgado, WE Scott, PY Tong; Screening for Amblyopia in Preverbal Children With Photoscreening Photographs: Identifying the Source of Interobserver Variability and Improving Agreement . Invest. Ophthalmol. Vis. Sci. 2002;43(13):2666.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose: One limitation of photograph-based photoscreening is inter-grader variability. To decrease this variability, we dissected the grading process for refractive error into three parts: (i) determining the adequacy of the photograph, (ii) measuring the dimensions of pupils and crescents, and (iii) deducing the presence of supra-threshold refractive error from the measurements obtained. Because step (ii) is the major source of variability, we determined its magnitude and devised a method to minimize it. Methods:Using digital calipers, two graders measured the bright crescent and pupil dimensions of 955 photographs from two prior studies. The 5% most discrepant were remeasured. The R squared correlation values of interclass plots measured the agreement between the graders. Results:Between the two graders, the more consistent 95% differed by 0.16mm±0.27 (mean±SD, range of 0.00-1.13mm) and 0.22mm±0.15 (range of 0.00-0.60mm)for bright crescents and pupils, respectively. Their more discrepant 5% differed by 1.94mm±1.01 (range of 1.14-6.76mm) and 0.86mm±0.39 (range of 0.61-3.25mm) for bright crescents and pupils, respectively. Remeasurements of this 5% differed by 1.00mm±0.69 (range of 0.00-7.98mm) and 0.36mm±0.24 (range of 0.00-3.55mm) for bright crescents and pupils, respectively. The mean difference more than halved with remeasurement. In the interclass plots for the 955 photographs, the correlation values were ∼ 0.62 and 0.87 for bright crescents and pupils, respectively. The more discrepant 5% had correlation values of ∼ 0.01 and 0.43 for bright crescents and pupils, respectively. The remeasured 5% had higher values of ∼ 0.11 and 0.86 for bright crescents and pupils, respectively. Agreement was consistently higher for pupils than it was for crescents. In addition, remeasurements improved agreement. Conclusion:We conclude; (i) multiple gradings reduce variability; (ii) measurements of crescents, as compared to those of pupils, result in greater variability; (iii) an objective method to identify the bright crescent-dark crescent junction will reduce the variability.
This PDF is available to Subscribers Only