The Reliability and Reproducibility of Corneal Confocal Microscopy in Children.

PURPOSE To assess the image and patient level interrater agreement and repeatability within 1 month for corneal nerve fiber length (CNFL) measured using in vivo corneal confocal microscopy (IVCCM) in children. METHODS Seventy-one subjects (mean [SD] age 14.3 [2.6] years, range 8-18 years; 44 with type 1 diabetes and 27 controls; 36 males and 35 females) were included. 547 images (∼6 images per subject) were analyzed manually by two independent and masked observers. One-month repeat visit images were analyzed by a single masked observer in 21 patients. Automated image analysis was then performed using a specialized computerized software (ACCMetrics). RESULTS For CNFL, the ICC (95% CI) were 0.94 (0.93-0.95) for image-level, 0.86 (0.78-0.91) for patient-level, and 0.88 (0.72-0.95) for the 1-month repeat assessment, and the Bland-Altman plots showed minimal bias between observers. Although there was excellent agreement between manual and automated analysis according to an ICC 0.89 (0.82-0.93), the Bland-Altman plot showed a consistent bias with manual measurements providing higher readings. CONCLUSIONS In vivo corneal confocal microscopy image analysis shows good reproducibility with excellent intraindividual and interindividual variability in pediatric subjects. Since the image-level reproducibility is stronger than the patient-level reproducibility, refinement of the method for image selection will likely further increase the robustness of this novel, rapid, and noninvasive approach to detect early neuropathy in children with diabetes. Further studies on the use of IVCCM to identify early subclinical neuropathy in children are indicated.