Agreement in reading center grading of Isotretinoin(Accutane)-induced meibography changes

Abstract

Purpose : Meibography imaging has becomes increasingly common in clinical ophthalmic practice. Several different grading techniques exist for evaluating the images generated by meibographers. In addition, many practitioners have noticed pathological changes in the meibomian gland structure and function in patients who have a history of isotretinoin (Accutane) medical therapy. This study examines the interobserver reliability between different masked observers for two of the most common grading methods on patients with isotretinoin-related meibomian gland atrophy.

Methods : Nine patients (six male, three female) had a history of isotretinoin use for less than 12 month duration. Eighteen lower eyelid images were obtained primarily from Oculus Keratograph 5M and ReSeeVit instruments. Four experienced masked observers graded the 18 images. The Pult and Arita scales were evaluated using Cohen kappa, weighted kappa, and intraclass correlation to determine agreement.

Results : In all 18 images, at least one observer graded some degree of meibomian gland atrophy. For the Pult system, unweighted kappa indicated slight agreement between graders (0.14 [0.00-0.46]) and an ICC of 0.733 [0.426-0.890], p < 0.001. The Arita scale kappa indicated fair agreement between graders (0.23 [0.00-0.55]) and an ICC of 0.524 [0.101-0.79], p = 0.009. Weighted kappa with the Pult scale indicated moderate agreement (0.48) between graders and fair agreement (0.36) using the Arita scale.

Conclusions : Isotretinoin has been documented to cause structural changes in meibomian gland architecture visible via meibography. As meibography becomes more widespread in clinical practice, incorporation of a pictoral scale which is reliable and repeatable is necessary. The Pult scale seemed to fare better than the Arita scale with respect to interobserver agreement, and thus may be preferred for diagnosing and monitoring meibomian gland atrophy and interlaboratory collaborations.

This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.

 

Image A- Average masked grading: Heiko Pult: 4 Arita: 3