To identify a clinically meaningful threshold for change in retinal thickness measured by optical coherence tomography for patients with uveitic macular edema using correlation with change in visual acuity.
Cross-sectional and longitudinal study.
One hundred twenty-eight eyes (101 individuals) with macular edema enrolled in the Multicenter Uveitis Steroid Treatment (MUST) trial. At enrollment and after 6 months of follow-up, retinal thickness was measured at the central subfield with time-domain optical coherence tomography and visual acuity was measured with logarithmic (Early Treatment Diabetic Retinopathy Study) visual acuity charts. Participants were classified as having macular edema if the retinal thickness was 260 μm or more.
A threshold for change in retinal center subfield thickness of 20% balanced the percentage of false positives and false negatives for predicting more than a 10-letter change in visual acuity with a sensitivity of 77% and a specificity of 75%. The results were similar for more than 5-letter changes and for 15-letter or more changes. Those with a 20% or more reduction in retinal thickness had a mean 11.0-letter improvement (95% confidence interval, 7.7 to 14.3) as compared with a −0.4-letter change (95% confidence interval, −4.1 to 3.3) in visual acuity for those without a 20% reduction ( P < .01).
In addition to being above the level of measurement uncertainty, a 20% change in retinal thickness in patients with macular edema seems to be optimal for clinically important changes in visual acuity and may be considered as an outcome for clinical trials of treatments for uveitic macular edema.
Macular edema (ME) is among the most frequent structural complications of uveitis and is a common cause of vision loss. Initially, ME was evaluated by measuring the area of macular leakage with fluorescein angiography. Optical coherence tomography (OCT) measures retinal thickness in a quantifiable and reproducible fashion and largely has replaced fluorescein angiography for managing ME in part because visual acuity is associated more strongly with retinal thickness than with macular leakage.
For measurements graded as abnormal, there are 2 possible response targets in clinical research: normalization (or resolution) and improvement. Visual acuity changes often are reported in both ways: recovery to a good acuity (ie, 20/40 or better) or a 1-, 2-, or 3-line improvement on a standard acuity chart. Similarly, changes in ocular inflammation are reported as resolution (grade 0 inflammation) or improvement (a 2-step change in the standardized semiquantitative grading scales).
Similarly, retinal thickness changes in patients with uveitic ME can be quantified as achieving resolution (return to a normal retinal thickness) or achieving improvement (change by a clinically meaningful amount). Previous data have suggested that the inter-measurement variability of OCT, regardless of machine, is less than 10%. Nevertheless, it is unclear whether a change of 10% is clinically meaningful or is the optimal threshold for change in retinal thickness as an outcome measurement for clinical research.
Visual acuity loss in patients with uveitis often is multifactorial, and the causes of visual acuity loss include ME, cataract, disc damage from uveitic glaucoma, and other media opacities from the inflammation. Nevertheless, ME is the most common cause of vision loss in patients with uveitis. Therefore, one potential way to determine a threshold for considering a change in ME to be clinically significant is to correlate change in macular thickness with change in visual acuity. However, currently there are no data to determine what level of change in retinal thickness best correlates with clinically meaningful changes in visual acuity in patients with uveitic ME. Although visual acuity is affected by multiple factors and OCT changes would not be expected to correlate perfectly with changes in VA, if it were necessary to decide whether ME had improved, then the change in thickness that best correlated with improvement in visual acuity would seem to be the most reasonable choice. Therefore, we decided to define a clinically meaningful change in retinal thickness for uveitic ME as the change in thickness that has optimal sensitivity and specificity for predicting clinically meaningful changes in VA and is higher than the reproducibility threshold. Furthermore, to be machine independent, we based the definition on a percentage change, rather than a fixed change, in the number of micrometers of thickness.
The Multicenter Uveitis Steroid Treatment (MUST) trial provides an opportunity to identify this threshold because masked visual acuity and objective OCT measurements were obtained at regularly specified time points throughout the trial. In this article, we report the association between changes in retinal thickness and visual acuity to determine a threshold of retinal thickness change that best predicts clinically meaningful changes in visual acuity.
This analysis includes data collected as of June 1, 2010, in the MUST trial, a prospective, multicenter, international clinical trial ( ClinicalTrials.gov registration no.: NCT00132691 ) comparing the safety and effectiveness of local therapy with a fluocinolone acetonide implant with systemic therapy for patients with severe noninfectious intermediate, posterior, or panuveitis.
A detailed version of the enrollment criteria for the MUST trial is published elsewhere. Eyes were included in this analysis if the following criteria were met at enrollment: (1) a high-quality OCT scan was available and (2) retinal thickness, measured at the central subfield, was 260 μm or more.
Best-corrected visual acuity was measured in a masked fashion using logarithmic (Early Treatment Diabetic Retinopathy Study) visual acuity charts according to a standardized protocol. Optical coherence tomography was performed by certified photographers using Stratus OCT 3 machines (Carl Zeiss Meditec, Jena, Germany) to obtain fast macular retinal thickness maps and high-resolution cross-hair scans. The images were graded by a centralized reading center at the University of Wisconsin, Madison. The current analysis was restricted to those eyes for which high-quality scans (ie, sufficient quality to derive reliable central subfield data from the automated retinal thickness map) could be attained, providing the best possible data for center point and center subfield measurements. The parameters required for high-quality scans included proper centration of the scan, proper determination of retinal pigment epithelium and internal limiting membrane boundaries by automated algorithm, signal strength of 5 or more, and standard deviation of the center point and center subfield measurements of less than 10%. An eye was defined as having ME if the retinal thickness at the central subfield was more than 260 μm (normative values, 212 ± 20 μm).
Main Outcome Measures
The primary outcomes of interest were the change in visual acuity from enrollment to 6 months and the percentage change in retinal thickness over the same period. The 6-month time point was chosen to allow sufficient follow-up in which to observe meaningful changes in both retinal thickness and visual acuity, since the latter may lag behind the former. Furthermore, it was necessary to allow sufficient recovery time after surgery for patients treated with an implant to reduce the likelihood of observing any transient vision loss associated with implant surgery.
Three cutoffs for change in visual acuity were evaluated: 5, 10, and 15 letters. A change of more than 5 letters is considered reproducible. A change of more than 10 letters is considered clinically meaningful. A change of 15 or more letters represents a doubling of the visual angle (eg, a change from 20/20 to 20/40) and has been a standard outcome measurement in clinical trials.
The percentage change in retinal thickness was calculated rather than the absolute change. An absolute change assumes that a specific value has the same influence regardless of the thickness at enrollment. In contrast, the percentage change takes into account the initial value. This quantification parallels the logOCT scale proposed by Ferris and associates in which a 1-step change is approximately equivalent to a 20% change. Furthermore, this method has the advantage of being generalizable to all OCT machines, each of which has a different normal range.
Summary statistics were computed for continuous (medians, interquartile ranges, and ranges) and categorical (counts and percentages) variables. The Spearman rank correlation was used to estimate the association between visual acuity and retinal thickness for specific time points as well as change over time. The linear, logistic, and ordinal relationships between visual acuity and retinal thickness outcomes were modeled using generalized estimating equations to adjust for within-person, between-eye correlations.
Sensitivity and specificity were used to assess the ability of changes in retinal thickness to predict clinically meaningful changes in visual acuity accurately. Receiver operating characteristic curves and plots of the sensitivity and specificity versus percentage change in retinal thickness were generated to display graphically the ability of the change in retinal thickness to predict changes in visual acuity for a range of thresholds. Bootstrap confidence intervals were computed for the correlations and prediction characteristics to adjust for within-person, between-eye correlations. Multiple imputation techniques were used to assess the effect of missing data. Statistical analyses were performed using SAS software version 9.1 (SAS Institute, Cary, North Carolina, USA) and the R version 2.11.1 (The R project for Statistical Computing, Vienna, Austria).
A total of 255 patients (479 eyes) with uveitis were enrolled in the MUST trial. Of the 322 eyes with high-quality OCT scans at enrollment, 128 (40%) eyes (101 individuals) were diagnosed with ME and 194 (60%) did not have ME. The remaining eyes were excluded from the analysis because the OCT was missing (n = 26; 6%), the OCT was unreadable (n = 18; 4%), or the OCT did not have sufficient quality to derive reliable central subfield data from the automated retinal thickness map (n = 113; 24%). The large number eyes without a high-quality OCT at enrollment was to be expected since there was no requirement for high-quality images at enrollment in order to enroll patients representative along the spectrum of the disease and to encourage recruitment. Indeed, patients with media opacity (primarily cataract) and inability to dilate the pupil (primarily due to posterior synechiae) were eligible for the trial. The percentage of eyes that had a cataract or were aphakic or pseudophakic was significantly higher ( P = 0.04) in those eyes for which the OCT was missing or ungradable (92% and 94%, respectively) as compared with those eyes with an OCT that was gradable manually (81%) or using the automated retinal thickness map (74%).
The 128 eyes identified as having ME with a high-quality OCT constitute the basis of the report. Of these, 75 eyes (53 individuals) had complete enrollment and follow-up data for visual acuity as well as a high-quality OCT scan at 6 months. Incomplete data were the result of missing the 6-month visit (n = 9), missing an OCT scan (n = 4), being ungradable OCT (n = 3), missing visual acuity data (n = 1), or having an OCT scan that was not of sufficient quality (n = 36) at the 6-month visit.
Characteristics of the Study Population
The characteristics of the entire MUST cohort are described elsewhere. Table 1 compares the demographic and clinical characteristics for eyes with ME at enrollment who had complete data for visual acuity and retinal thickness at enrollment and 6 months with those who did not. In brief, for those eyes with complete follow-up, the median age was 50 years (twenty-fifth to seventy-fifth percentile, 40 to 57), the median time from diagnosis with uveitis was 3.8 years (twenty-fifth to seventy-fifth percentile, 1.1 to 7.9), 33% were male, and 69% were white. These eyes were likely to have vitreous haze (n = 53; 73%), vitreous cells (n = 65; 89%), and cataracts (n = 33; 44%) or be aphakic or pseudophakic (n = 30; 40%). The median visual acuity was 68 letters (twenty-fifth to seventy-fifth percentile, 57 to 79), which is approximately 20/40 Snellen equivalent (twenty-fifth to seventy-fifth percentile, 20/70 to 20/25). Most of the characteristics of uveitis and its complications were similar for those eyes that did not have a complete follow-up. However, those eyes that did not have a complete follow-up were more likely to be aphakic or pseudophakic (66% vs 40%). There was no significant difference between the change in visual acuity over 6 months for those who had an OCT of sufficient quality and those who did not (P = .2). The remainder of this article focuses on the characteristics of those 75 eyes for which high-quality OCTs and visual acuity measurements were available at enrollment and 6 months.
|Characteristic||Eyes with Complete Follow-up (n = 75)||Eyes with Incomplete Follow-up (n = 53)|
|Age (years) a||50 (40 to 57)||58 (48 to 69)|
|Female, n (%) b||50 (67%)||42 (79%)|
|White, non-Hispanic, n (%) b||52 (69%)||31 (58%)|
|Any systemic disease, n (%) b||25 (33%)||11 (21%)|
|Bilateral uveitis, n (%) b||68 (91%)||47 (89%)|
|Time since diagnosis (years) a||3.8 (1.1 to 7.9)||4.2 (1.2 to 9.9)|
|Missing, n (%) b||3 (4%)||3 (5%)|
|Vitreous haze c , n (%) b||54 (73%)||41 (77%)|
|Missing, n (%) b||1 (1%)|
|Vitreous cells d , n (%) b||65 (89%)||51 (96%)|
|Missing, n (%) b||2 (3%)|
|Lens status, n (%) b|
|Phakic||12 (16%)||3 (6%)|
|Cataract||33 (44%)||15 (28%)|
|Aphakic/pseudophakic||30 (40%)||35 (66%)|
Change in Retinal Thickness and Visual Acuity
Visual acuity and retinal thickness measurements at enrollment and after 6 months of follow-up and change during the initial 6 months of follow-up are summarized in Table 2 . Median visual acuity at enrollment was 68 letters (range, 7 to 94 letters), and the median change in visual acuity was 4 letters (range, −28 to 35 letters). The median change for the 47 eyes with an improvement in vision was 10 letters (range, 1 to 35 letters) and that for the 28 eyes without improvement in vision was −6 letters (range, −28 to 0 letters). Thirty-five (47%) eyes had an improvement of more than 5 letters, 22 (29%) had an improvement of more than 10 letters, and 13 (17%) had a halving of the visual angle (ie, an improvement of 15 letters or more). The median retinal thickness at the central subfield at enrollment was 327 μm (range, 260 to 1195 μm), and the median percentage change from enrollment to 6 months was a 15% reduction (range, −86% to 152%). The median change for those with improvement was a reduction in retinal thickness of 25% (range, 1% to 86%), and that for those without improvement was an increase of 13% (range, 0% to 52%).
|Characteristic||Eyes with Macular Edema a at Enrollment (n = 75)|
|Visual acuity (EDTRS letters) b|
|Enrollment c||68 (7 to 94)|
|Six months c||73 (19 to 93)|
|Change from enrollment to 6 months c||4 (–28 to 35)|
|>5-letter improvement, n (%)||35 (47%)|
|>10-letter improvement, n (%)||22 (29%)|
|≥15-letter improvement, n (%)||13 (17%)|
|Retinal thickness (μm) d|
|Enrollment c||327 (260 to 1195)|
|Six months c||261 (133 to 840)|
|Change from enrollment to 6 months c , e||–43 (–870 to 428)|
|Percentage change c , d , e , f||–15% (–86% to 152%)|
Association between Retinal Thickness and Visual Acuity
For each 100-μm lower retinal thickness, the visual acuity was 6.5 letters higher (95% confidence interval, 5.2 to 7.6; P < .01) at enrollment and 5.3 letters higher (95% CI, 1.9 to 8.8; P < .01) after 6 months of follow-up, respectively. The cross-sectional Spearman rank correlation between visual acuity and retinal thickness at enrollment was −0.56 (95% CI, −0.74 to −0.35) and that at 6 months was −0.24 (95% CI, −0.46 to 0.02). As retinal thickness decreased over time, vision improved ( Figure 1 ) ; Spearman rank correlation, −0.46; 95% CI, −0.62 to −0.27; P < .01). Over the 6-month follow-up period, the average change in visual acuity was 1.5 letters (95% CI, 0.7 to 2.2; P < .01) for each 10% decrease in retinal thickness, the minimum reproducible difference.
Identifying a Threshold for Retinal Thickness
The performance of a range of potential retinal thickness cutoffs for predicting changes in visual acuity was similar for all 3 of the vision thresholds, as demonstrated by the overlap in the receiver operating characteristic curves, which plot the sensitivity versus 1 minus specificity ( Supplemental Figure ). The relationship between the percentage change in retinal thickness and the sensitivity (percentage of true positives) and specificity (percentage of true negatives) for identifying a more than 10-letter change in visual acuity is presented in Figure 2 . The point at which the 2 curves cross represents the percentage change in retinal thickness for which the sensitivity and specificity are equal, that is, the level of false positives and false negatives are equal. Overall, a 20% reduction in retinal thickness seems to balance the 2 error types for a change in visual acuity of more than 10 letters with a sensitivity of 77% (95% CI, 60% to 95%) and a specificity of 75% (95% CI, 62% to 86%). Those with a 20% or more reduction in retinal thickness had on average a 11.0-letter improvement (95% CI, 7.7 to 14.3) as compared with a −0.4-letter change (95% CI, −4.1 to 3.3) in visual acuity for those without a 20% reduction (difference, 11.4; 95% CI, 6.9 to 15.9; P < .01).
The association between a 20% change in retinal thickness and clinically meaningful changes in visual acuity (>5 letters, >10 letters, and ≥15 letters) is explored in Table 3 . Comparisons were made for changes in either direction (eg, ≥20% reduction, < 20% change, or ≥20% increase) as well as for improvement only (eg, ≥20% reduction vs < 20% reduction). In both cases, the 20% change was significantly associated with all 3 visual acuity thresholds.
|Change in Visual Acuity||≥20% Reduction||<20% Change||≥20% Increase||P Value|
|Overall trend a|
|>5-letter improvement||25 (83%)||9 (24%)||1 (14%)||<.001 b|
|≤5-letter change||3 (10%)||19 (50%)||3 (43%)|
|>5-letter decline||2 (7%)||10 (26%)||3 (43%)|
|>10-letter improvement||17 (57%)||5 (13%)||0 (0%)||<.001 b|
|≤10-letter change||11 (36%)||26 (69%)||4 (57%)|
|>10-letter decline||2 (7%)||7 (18%)||3 (43%)|
|>15-letter improvement||10 (33%)||3 (8%)||0 (0%)||<.001 b|
|≤15-letter change||19 (63%)||31 (82%)||6 (86%)|
|>15-letter decline||1 (3%)||4 (10%)||1 (14%)|
|>5-letter improvement||25 (83%)||10 (22%)||<.001 c|
|≤5-letter improvement||5 (17%)||35 (78%)|
|>10-letter improvement||17 (57%)||5 (11%)||<.001 c|
|≤10-letter improvement||13 (43%)||40 (89%)|
|>15-letter improvement||10 (33%)||3 (7%)||.006 c|
|≤15-letter improvement||20 (67%)||42 (93%)|