Optimal Optical Coherence Tomography–Based Measures in the Diagnosis of Clinically Significant Macular Edema: Retinal Volume vs Foveal Thickness

Objective  To compare optical coherence tomography–based measures of retinal thickness and volume as quantitative tests for clinically significant macular edema (CSME).

Design  Diagnostic validation study.

Methods  Sixty-five eyes with diabetic retinopathy underwent stereo photographic and optical coherence tomographic examination. Stereo photographs were examined in a masked fashion to determine the presence or absence of CSME according to criteria from the Early Treatment Diabetic Retinopathy Study. Optical coherence tomography–based measurements of central foveal thickness as well as retinal volumes within a series of radii of fixation were generated. The main outcome measures were areas under receiver operating characteristic curves. Likelihood ratios, sensitivities, and specificities for the diagnosis of CSME were also evaluated.

Results  Retinal volumes within radii of 0.50 mm and 1.11 mm of fixation and central foveal thickness were the best variables for discriminating between those with and without CSME as evidenced by analysis of receiver operating characteristic curves. There were no significant differences among these 3 variables in their performance as diagnostic tests for CSME.

Conclusions  Optical coherence tomography–based retinal volume and central foveal thickness variables display comparable abilities to discriminate between those with and without CSME. Both measures may have clinical applications as quantitative diagnostic tests for CSME.

Diabetic macular edema is a major cause of decreased visual acuity in western nations.^1,2 Laser photocoagulation improves outcomes in patients who have reached the stage of clinically significant macular edema (CSME).³ As the diagnosis of CSME requires expert clinical evaluation, the provision of diagnostic services is suboptimal in many jurisdictions. Thus, there exists a need for a simple, noninvasive test capable of quantitatively discriminating between those with CSME and those without. The introduction of retinal optical coherence tomography (OCT) has been a major step toward this goal. While OCT elegantly demonstrates the anatomic effects of retinal edema in a qualitative fashion, the great power of this technology resides in its potential for quantitative measures.^4-7

Numerous quantitative variables may be generated by retinal OCT. Among these, single-point central foveal thickness and mean sectoral retinal thickness measures have been shown to be good predictors of CSME.^8,9 Browning and Fraser¹⁰ have also shown that the intraindividual comparison of retinal sectors may provide important additional information. While these varied retinal thickness measurements have been suggested as diagnostic tests for CSME, the optimal OCT-based variable for use in quantitatively detecting CSME is currently unknown. In this role, retinal volume measurement may have a number of practical and theoretical advantages over retinal thickness determinations. Thus, we have compared the use of OCT-based foveal thickness measurements with a series of retinal volume measurements and a novel retinal volume sector analysis in the detection of CSME as defined by the Early Treatment Diabetic Retinopathy Study (ETDRS).³

Sixty-five eyes from 34 patients with type 1 or type 2 diabetes mellitus referred to the retina and comprehensive ophthalmology services of our department were prospectively enrolled in this study. Patients were excluded if they exhibited clinical evidence of any retinal disease other than diabetic retinopathy. The degree of diabetic retinopathy in the sample was representative of the spectrum of this disease process. Sample size was chosen to provide an estimated power of 0.85 to detect a 0.10-unit difference in area under receiver operating characteristic (ROC) curves at an α level of .05.

All subjects underwent OCT and stereo fundus photography on the same day. Technical details of OCT have been described previously.^4-7 All patients underwent a “fast macular thickness” scanning protocol using a commercially available OCT system (Stratus OCT 3.0 model 3000; Carl Zeiss Meditec, Dublin, Calif). Briefly, this protocol comprised 768 individual A-scans carried out at equally spaced locations along six 6-mm–long radial lines intersecting at their midpoints. This intersection is located at fixation. All individual line scans were manually reviewed to verify that the software algorithm had accurately identified the retinal internal limiting membrane and retinal pigment epithelium.

Estimates of retinal volumes within 9 ETDRS retinal sectors were generated (Figure 1). Commercial software was used to generate total volume estimates within radii of 0.50 mm, 1.73 mm, and 3.00 mm of fixation. As total retinal volume measurements within radii of 1.11 mm and 1.50 mm from fixation are not automatically generated by commercial software, the calculation of these values required data export. Central foveal thickness, designated foveal thickness by commercial software, was calculated as the average thickness at the intersection of the 6 scan lines.

Additionally, because the definition of CSME includes subtypes in which macular edema does not involve the center of the fovea, an effort was made to develop an OCT-based variable capable of detecting such CSME subtypes. Thus, a novel retinal volume sector analysis was designed specifically for this study. This analysis examined the 5 central sectors in the most magnified OCT output mode (Figure 1: sectors 1, T1, S1, N1, and I1 in output mode 1) and was defined as the number of sectors with a volume greater than the 95th percentile among diabetic eyes without CSME. Hence, individual scores for this variable ranged from 0 to 5.

The gold standard for the detection of CSME adhered to the strict definitions set out in the ETDRS.³ Standard stereo photographs encompassing the macula and optic disc were evaluated by an experienced, fellowship-trained retina specialist (P.J.K.) in a masked fashion. Eyes were then classified as either CSME present or CSME absent according to ETDRS definitions.³

Foveal thickness and retinal volumes within the series of radii of fixation were compared between the CSME present and CSME absent groups using nonpaired t tests. Variance estimates were adjusted to compensate for correlation between eyes from individual subjects using the design effect correction factor according to the method of Kish (Stata, College Station, Tex).¹¹ To evaluate which variables best discriminate between those with and without CSME, areas under ROC curves were compared via a z test according to the method of Hanley and McNeil (MedCalc, Mariakerke, Belgium).¹² Finally, sensitivities, specificities, and likelihood ratios were generated for all predictor variables.

Overall, 29 eyes were classified as CSME present and 36 as CSME absent according to ETDRS criteria. As expected, all retinal volumes as well as foveal thickness were significantly greater in patients with CSME than in those without (Table 1).

Receiver operating characteristic curves provide a useful overall measure of the performance of a continuous variable as the basis for classifying patients as diseased or not diseased. Receiver operating characteristic curves plot the true-positive rate vs the false-positive rate (ie, sensitivity vs 1-specificity) at a series of predictor variable cutoff values. A useless test has an equal chance of designating those with and without disease as positive regardless of the cutoff employed. Consequently, such a test has equal true-positive and false-positive rates at all cutoffs and the ROC curve is a straight line at 45° to the horizontal (Figure 2; line A). The area under this ROC curve is 0.5. On the other hand, a perfect test—that is, a test that at one particular cutoff perfectly separates patients into a group with disease and one without—would have an ROC curve that proceeds straight up from the origin and then across the top of the graph (Figure 2; line B). The area under this ROC curve equals 1. In the real world, all continuous variables used as diagnostic tests fall somewhere in between these 2 extremes (Figure 2; line C). As the cutoff is adjusted, improvement in the true-positive rate (sensitivity) comes at the expense of some increase in the false-positive rate (1-specificity). The better the diagnostic test, the faster the true-positive rate will rise relative to the false-positive rate. As a consequence, the ROC curve will rise above the 45° line and the area under the curve will increase toward the perfect value of 1.

In the present study, foveal thickness and retinal volumes within radii of 0.5 mm and 1.11 mm of the scan center displayed areas under the ROC curves that were not significantly different (Figure 3 and Figure 4; P>.05). However, all 3 of these variables were significantly better at discriminating between those with and without CSME than retinal volumes within the larger radii (Figure 3 and Figure 4; P<.05 for all comparisons). The sector analysis variable developed for this study did not provide additional predictive ability over retinal volume and thickness measures as evidenced by the area under the ROC curve for this variable (Figure 3 and Figure 4).

While ROC curves help determine which variables provide the greatest clinical utility, other statistics are needed to provide a clinical feel for test results in an individual patient. Likelihood ratios compare the odds of a given test result occurring in a patient with the target disorder to the odds of the same result occurring in a patient without that disorder. This ratio encompasses both sensitivity and specificity information in 1 variable and lends itself to clinical usage as one simply needs to multiply the pretest odds of the disorder by the likelihood ratio to arrive at a posttest estimate of disorder likelihood.

In the present study, retinal volume measures within smaller radii as well as foveal thickness displayed the most powerful likelihood ratios (Table 2). For comparison, we have also included sensitivities and specificities that echo this message with foveal thickness and retinal volumes within smaller radii performing best (Table 2). Table 3 demonstrates the use of likelihood ratios to refine estimates of disease likelihood from the pretest to posttest situation.

Quantitative OCT-based measures have become an integral part of macular disease assessment.¹³ However, as this technology can generate many different measures, efforts continue to determine which variables provide the greatest clinical utility. Previous studies have evaluated single-point central foveal thickness and mean sectoral retinal thickness measures and found these variables to be good predictors of CSME.^8,9,14 Furthermore, Browning and Fraser¹⁰ have demonstrated that intraindividual retinal sector comparisons also provide important information. The analysis of ROC curves in the present study suggests that retinal volumes within smaller radii of fixation are more clinically useful variables than volumes within larger radii. Specifically, retinal volumes within 0.5 mm and 1.11 mm of fixation appear to be the best retinal volume measures for use in detecting CSME.

Interestingly and importantly, foveal thickness determinations were as effective as these retinal volume measures in discriminating CSME from non-CSME. However, a companion article has demonstrated that retinal volume measures are less susceptible to artifact induced by changes in fixation.¹⁵ Thus, retinal volume measures within small radii of fixation appear to provide a good balance between diagnostic utility and long-term stability.

The greater utility of volumes within smaller radii and central foveal thickness is likely related to the potential to dilute the impact of important focal edema when measuring over larger retinal areas. However, measurements of thickness or volume within small central regions will fail to detect subtypes of CSME in which retinal thickening does not involve the central fovea. Thus, in an attempt to better detect these forms of CSME, we developed a sector analysis variable that analyzed retinal sectors individually. However, the use of this variable did not improve overall diagnostic utility. It is possible that, through the examination of even smaller sectors, one might detect CSME subtypes with focal edema more readily. Future software refinements could allow for this.

To date, most studies reporting the distribution of retinal thickness in normal subjects have used older OCT scanning systems.^4-8,15 However, normative data generated from the OCT system employed in the present study have been recently published.¹⁶ Chan et al¹⁶ reported a mean central foveal thickness of 182 μm, which is very close to the value of 186.7 μm in our CSME-absent group (Table 1). Chan et al¹⁶ also suggested 2 standard deviations above their mean as an arbitrary but commonly employed upper limit of normal. This thickness cutoff of 228 μm corresponds in the present study to a sensitivity of approximately 75% and specificity of approximately 90% (Table 2). Chan et al¹⁶ also report a mean thickness within a radius of 0.50 mm of the scan center of 212 μm, which converts to a retinal volume within this sector of approximately 0.17 mm³. This volume is very similar to the volume of 0.16 mm³ found in the CSME-absent group of the present study (Table 1). Moreover, for this variable the 2 standard deviation upper limit is 252 μm, which converts to a retinal volume of approximately 0.20 mm³. Again this agrees very well with our results with this cutoff yielding a sensitivity of approximately 78% and specificity of approximately 95% (Table 2).

The lack of complete agreement between the qualitative criteria used in stereo photograph examination and the quantitative measures generated by OCT is not unexpected. Indeed, previous studies have shown that OCT often detects retinal thickening in the absence of thickening detectable via biomicroscopy.^9,17,18 As we gain further clinical insight into the prognosis for patients with retinal thickening detectable only via OCT, our treatment paradigms may evolve. However, we focused on the ability of OCT-based measures to detect CSME as defined in the ETDRS because, at present, the best evidence regarding treatment protocols for diabetic macular edema comes from that trial.³

To minimize overestimation of diagnostic utility, every effort was made to ensure that the study population spanned the spectrum of disease severity and included a significant number of intermediate cases. Additionally, to maximize generalizability, this study employed commonly used OCT scanning protocols. A weakness of this study is the coarse integration process used by the commercial software to generate volume estimates. This involves multiplying retinal surface area by average retinal thickness within rather large retinal sectors. Additionally, current software does not automatically generate total retinal volumes within radii of 1.11 mm and 1.50 mm of fixation. Thus, this study required data export to generate these values, making the routine use of these particular measures impractical in the clinical setting at present. Future software refinements could allow more accurate volume calculations via finer integration and may also directly generate volumes within a wider variety of central radii.

Although clearly not a replacement for the clinical examination, OCT has found many important applications in clinical care. We have found retinal volumes within small radii of fixation and foveal thickness equally effective in detecting CSME. These OCT-based measures demonstrate impressive sensitivities and specificities and hold great promise as objective and quantitative diagnostic tools for use in the care of patients with diabetic retinopathy.

Correspondence: Robert J. Campbell, MD, MSc, FRCSC, Department of Ophthalmology, Queen's University, Hotel Dieu Hospital, 166 Brock St, Kingston, Ontario, Canada K7L 5G2 (rob.campbell@queensu.ca).

Submitted for Publication: June 4, 2006; final revision received September 21, 2006; accepted September 25, 2006.

Author Contributions: Dr Robert Campbell had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Financial Disclosure: None reported.

Funding/Support: This study was supported in part by a grant from the University of Ottawa Medical Research Fund.

Previous Presentation: The study was presented in part at the Annual Meeting of the Canadian Ophthalmological Society; June 18, 2004; Vancouver, British Columbia.

Acknowledgment: We thank Drs Sherif El-Defrawy, Brian Leonard, and George Mintsioulis for providing some of the subjects sampled. We also thank Ms Jeanne Koroluk and Ms Rosario Bate for technical assistance and Dr Erica de L. P. Campbell for help in manuscript preparation.