HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

Electronic Letters to:

Glaucoma: Julio E. DeLeón Ortega, Lisandro M. Sakata, Bobby Kakati, Gerald McGwin, Jr, Blythe E. Monheit, Stella N. Arthur, and Christopher A. Girkin Effect of Glaucomatous Damage on Repeatability of Confocal Scanning Laser Ophthalmoscope, Scanning Laser Polarimetry, and Optical Coherence Tomography Invest. Ophthalmol. Vis. Sci. 2007; 48: 1156-1163 [Abstract] [Full text] [PDF]

eLetters: Submit a response to this article

Electronic letters published:

Intraclass Correlation Analysis May Alter Conclusions

William J Feuer   (25 June 2007)

Author Response: Intraclass Correlation Analysis May Alter Conclusions

Christopher A. Girkin   (25 June 2007)

Intraclass Correlation Analysis May Alter Conclusions 25 June 2007
William J Feuer, Reseach Associate Bascom Palmer Eye Institute Send letter to journal: Re: Intraclass Correlation Analysis May Alter Conclusions wfeuer{at}miami.edu William J Feuer	In their article comparing reproducibility of measurements of retinal nerve fiber layer (RNFL), disc parameters, and optic nerve parameters made with StratusOCT, GDx-VCC, and the HRT II, DeLeón Ortega et al.1 use the Bland-Altman repeatability coefficient (RC) to support their conclusion that the OCT performs worse than the GDx-VCC in glaucoma patients with severe visual field loss. This coefficient quantifies test-retest measurement variability and is useful, but is not itself sufficient for comparing reproducibility between instruments. Careful examination of the Bland-Altman plots (their Fig. 1), especially for RNFL, reveals that in addition to differences in test-retest variability between instruments there are also differences in the range of measurements in the severe loss group made by GDx-VCC (approximately 20-60 mm) and StratusOCT (approximately 35-95 mm). That is, the OCT measurements have, in addition to more within-patient test-retest variability, more variability between patients, suggesting better discrimination among patients by the OCT. The intraclass correlation coefficient (ICC)2 is a statistic which scales the test-retest measurement variability (undesirable variability) for the variability between subjects (those true differences between RNFL thicknesses of different patients appropriate for the instrument to detect in a cross-sectional study). I encourage DeLeón Ortega et al. to provide a table of ICCs and modify their conclusions if appropriate. While the published article does not contain data sufficient for estimating ICCs, dividing the RC by the range gives a very crude estimate of the effect of accounting for between-subject variability. For the RNFL measurements in the severe patient group, RC/range is about 3/40 = 0.08 for GDx-VCC, which is not appreciably different from 4/60 = 0.07 for StratusOCT. It is difficult to say if the correct analysis with intraclass correlations will be similar. The ICC is not useful for all reproducibility analyses. The effect of scaling for between-subject variability, which is advantageous in this situation, becomes a liability when comparing groups with substantially different biologic ranges of measurement, say, normal controls versus patients with progressive disease, but ICCs would be a useful addition to the data presented in the article by DeLeón Ortega et al. William J. Feuer Bascom Palmer Eye Institute, Miller School of Medicine, Miami, Florida References 1. DeLeón Ortega JE, Sakata LM, Kakati B, et al. Effect of glaucomatous damage on repeatability of confocal scanning laser ophthalmoscope, scanning laser polarimetry, and optical coherence tomography. Invest Ophthalmol Vis Sci. 2007;48:1156-1163. 2. Fleiss JL. Design and Analysis of Clinical Experiments. New York: John Wiley & Sons; 1986:1-32.
Author Response: Intraclass Correlation Analysis May Alter Conclusions 25 June 2007
Christopher A. Girkin Send letter to journal: Re: Author Response: Intraclass Correlation Analysis May Alter Conclusions cgirkin{at}uab.edu Christopher A. Girkin	We appreciate Mr. Feuer's comments about our article1 addressing repeatability and imaging techniques in glaucoma patients. In addition to within-subject variance, repeatability coefficient (RC), and Bland-Altman plots, we did consider other analyses of test-retest repeatability including intraclass correlation coefficients (ICC) and coefficient of variation (CV). We are aware of advantages and disadvantages of each of these analyses and that there is no standard for repeatability analysis. We agree with Mr. Feuer in that the range of measures of nerve fiber layer (RNFL) thickness given by GDx and OCT differ and that a relative measure of measurement error such as ICC may account for this.2 However, ICC is dependent on study population since it uses between-subject variation. With this limitation in mind we report ICC as suggested by Mr. Feuer in the Table below. Results show similarities between the ICC and RC data. For GDx, ICC was greater than for OCT at all stages of disease severity, except for the severe field loss group. Furthermore, 95% confidence interval (CI) were narrower for GDx, and some slight overlapping for cases with early and moderate field loss. We choose to report RC for its simplicity, easy comparison between instruments, and, unlike ICC, Bland-Altman plots are able to demonstrate the agreement across the range of observed values making this approach useful. We stand by our published conclusions, with ICC analysis providing further support for our conclusions. Table 1. Reliability analysis between repeated measurements by scanning laser polarimetry and optical coherence tomography GDx-VCC StratusOCT ICC 95% CI ICC 95% CI Controls 0.979 0.961 to 0.989 0.937 0.886 to 0.966 Early 0.985 0.971 to 0.992 0.962 0.927 to 0.980 Moderate 0.991 0.981 to 0.996 0.962 0.922 to 0.982 Severe 0.986 0.972 to 0.993 0.984 0.967 to 0.992 ICC, intraclass correlation coefficient CI, confidence interval Julio De León Ortega, Gerald McGwin, Jr., Christopher A. Girkin University of Alabama at Birmingham, Birmingham, Alabama References 1. De León Ortega JE, Sakata LM, Kakati B, et al. Effect of glaucomatous damage on repeatability of confocal scanning laser ophthalmoscope, scanning laser polarimetry, and optical coherence tomography. Invest Ophthalmol Vis Sci. 2007;48:1156-1163. 2. Shrout PE, Fleiss JL. Intraclass correlations: uses in assessing rater reliability. Psychological Bulletin. 1979;86:420-428.