Best Way to Grade Meibography: Best Way to Grade Meibomian Gland Loss

Currently we do not have an automatic computer generated scoring of meibomian gland loss. That should be coming soon. The key thing to see is if a patient has lost a full gland from scar tissue in the gland. If this is the case, we know that the gland will not return and the other glands are at risk. Every effort should be made to prevent that patient from going from that Stage 2 to Stage 3 and 4.
We await eagerly the day when a computer will be able to objectively tell us how significant is the meibomian gland dysfunction.

Sandra Lora Cremers, MD, FACS

ARVO Annual Meeting Abstract  |   March 2012

An Assement of Subjective and Objective Grading of Meibography Images
Heiko Pult; Britta H. Riede-Pult
 Author Affiliations & Notes
  • Heiko Pult
    Contact Lens & Anterior Eye Research Unit (CLAER), School of Optometry and Vision Sciences, 
    Cardiff University, Cardiff, United Kingdom
    Optometry & Vision Research, Weinheim, Germany
  • Britta H. Riede-Pult
    School of Optometry and Vision Sciences, 
    Cardiff University, Cardiff, United Kingdom
    Optometry & Vision Research, Weinheim, Germany
  • Footnotes
    Commercial Relationships  Heiko Pult, None; Britta H. Riede-Pult, None
  • Support  None
Investigative Ophthalmology & Visual Science March 2012, Vol.53, 588. doi:

Abstract
Purpose: : To analyze repeatability of subjective and objective grading of meibography images.
Methods: : Meibography images of 24 subjects (female= 14; mean age= 46; range= 19 – 69 years) were classified in two different sessions (day 1 vs. 2) by two experienced observers. Loss of meibomian glands (MGL) was classified three times applying three different grading scales in randomized order: A four grade pictorial grading scale (4S) (degree 0= No MGL; 1=<25% MGL; 3= 25% to 75% MGL; 3= >75% MGL) a five grade pictorial grading scale (5S) (degree 0= No MGL; 1=<25% MGL; 3= 25% to 50% MGL; 3= 50% to 75%; 4= >75% MGL) and objectively by digital analyses (DA) obtaining Image J Software (degree: 0 – 100 MGL). Observers were masked to results. Differences between observers and session were analyzed by repeated measurements ANOVA (or non-parametric equivalent) and Bland Altman plots and relations by Pearson correlation (or non-parametric equivalent). 95% confidence intervals (CI) were calculated from the distributions of differences between observer and observations
Conclusions: : Intra-observer and inter-observer agreement was better in digital grading followed by the 5 grade scale and the 4 grade scale being the least repeatable one.
Shopping Cart