Improving precision for detecting change in the shape of the cornea in patients with keratoconus

M. Brunner, G. Czanner, R. Vinciguerra, V. Romano, S. Ahmad, M. Batterbury, C. Britten, C.E. Willoughby, S.B. Kaye

Research output: Contribution to journalArticlepeer-review

44 Citations (Scopus)
43 Downloads (Pure)


To investigate a method for precision analysis to discriminate true corneal change from measurement imprecision in keratoconus (KC). Thirty patients with KC and 30 healthy controls were included. Coefficients of repeatability and limits of agreement (LOA) were compared using multiple measurements for inter-observer and inter-device agreement with the Pentacam HR, Orbscan IIz, and Tomey Casia SS-1000. Correlation of repeated measurements was evaluated using a linear mixed effect model (also called random effect model). A formula was derived for the theoretical expected change in precision and compared with measured change. Correlation between measurements from the same eye was small (R = 0.13). The 99.73% LOA (3 SD) of the mean of three measurements, provided better precision than 95% LOA (2 SD) of single cut-off values as expected from statistical theory for uncorrelated measurements for evidence of a significant change in corneal shape in patients with keratoconus. This enabled the determination of cut-off values for the detection of true change in corneal shape. The mean of three repeated measurements will provide better precision when there is minimal correlation. Three (rather than two) standard deviations provides a precise estimate of the LOA within or between observers and can be used as a reliable measure for identifying stage-independent corneal shape changes (progression) in keratoconus.
Original languageEnglish
JournalScientific Reports
Issue number1
Early online date17 Aug 2018
Publication statusPublished (in print/issue) - 1 Dec 2018


Dive into the research topics of 'Improving precision for detecting change in the shape of the cornea in patients with keratoconus'. Together they form a unique fingerprint.

Cite this