Sensitivity and Specificity of Potential Diagnostic Features Detected Using Fundus Photography, Optical Coherence Tomography, and Fluorescein Angiography for Polypoidal Choroidal Vasculopathy

Key Points

Question  What are the potential diagnostic features detected using fundus photography, optical coherence tomography, and fluorescein angiography for polypoidal choroidal vasculopathy without the use of indocyanine green angiography?

Findings  In this cohort study, when at least 2 of the following 4 potential diagnostic signs were graded on deidentified images without the use of indocyanine green angiography, high sensitivity and specificity were noted: notched or hemorrhagic pigment epithelial detachment detected using fundus photography or optical coherence tomography and a sharply peaked pigment epithelial detachment or a hyperreflective ring detected using optical coherence tomography.

Meaning  These data suggest that fundus photography and optical coherence tomography provide high sensitivity and specificity for polypoidal choroidal vasculopathy without the use of indocyanine green angiography, especially when at least 2 of 4 highly suggestive signs are present.

Importance  The use of indocyanine green angiography (ICGA) is a criterion standard for diagnosing polypoidal choroidal vasculopathy (PCV), an endemic and common cause of vision loss in Asian and African individuals that also presents in white individuals. However, the use of ICGA is expensive, invasive, and not always available at clinical centers. Therefore, knowing the value of certain features detected using fundus photography (FP), optical coherence tomography (OCT), and fluorescein angiography (FA) to diagnose PCV without ICGA could assist ophthalmologists to identify PCV when ICGA is not readily available.

Objective  To explore the sensitivity, specificity, and predictive accuracy of potential diagnostic features detected using FP, OCT, and FA in diagnosing PCV without ICGA.

Design, Setting, and Participants  Deidentified images of FP alone, OCT alone, and FA alone were graded by 3 retina specialists masked to ICGA findings for potentially diagnostic features of PCV prespecified before grading compared with the criterion standard grading of 2 other retina specialists with access simultaneously to FP, OCT, FA and ICGA. Specialists graded images of 124 eyes of 120 patients presenting between January 1, 2013, and December 31, 2016, with newly identified serous or serosanguinous maculopathy who had undergone FP, OCT, FA, and ICGA before treatment at a large referral eye center in Thailand.

Main Outcomes and Measures  Sensitivity, specificity, positive predictive value, negative predictive value, and predictive accuracy from the area under the receiver operating characteristic curve (AUC).

Results  The mean (SD) age of the patients was 57.7 (12.6) years, 52 were women, 68 were men, and the diagnosis (from ICGA) was PCV for 65 eyes (52.4%), central serous chorioretinopathy for 45 eyes (36.3%), and typical neovascular age-related macular degeneration for 12 eyes (9.7%). With the use of FP, a potential diagnostic feature for PCV was notched or hemorrhagic pigment epithelial detachment (AUC, 0.77; 95% CI, 0.70-0.85). With the use of OCT, potential diagnostic features for PCV were pigment epithelial detachment notch (AUC, 0.90; 95% CI, 0.85-0.96), sharply peaked pigment epithelial detachment (AUC, 0.86; 95% CI, 0.80-0.92), and a hyperreflective ring (AUC, 0.86; 95% CI, 0.80-0.92). When at least 2 of these 4 signs were present, the AUC was 0.93 (95% CI, 0.89-0.98), with a sensitivity of 0.95 (95% CI, 0.87-0.99), a specificity of 0.95 (95% CI, 0.82-0.97), a positive predictive value of 0.92 (95% CI, 0.83-0.97), and a negative predictive value of 0.95 (95% CI, 0.86-0.99).

Conclusions and Relevance  These data suggest that the potential diagnostic features detected using FP and OCT provide high sensitivity and specificity for a diagnosis of PCV, especially when at least 2 of 4 highly suggestive signs are present.

Polypoidal choroidal vasculopathy (PCV) is a pattern of choroidal neovascularization (CNV) that may or may not have other features of age-related macular degeneration (AMD).¹ Polypoidal choroidal vasculopathy can be an important cause of vision loss, especially in populations of Asian or African ancestry^2-5 and in at least 10% of cases of CNV in white populations older than 50 years. The use of indocyanine green angiography (ICGA) is a criterion standard for diagnosing PCV in clinical trials,^6,7 but ICGA is invasive and not always available or feasible to perform, including in many areas in the developing world where PCV is endemic. Nevertheless, differentiating PCV from other conditions, such as typical CNV due to AMD or central serous chorioretinopathy, is desirable because the diagnosis of PCV can determine the recommendations for patients regarding management and prognosis.

When ICGA is not available, information from other retinal imaging may be helpful in the diagnosis of PCV. Features of optical coherence tomography (OCT) have been described and demonstrated high sensitivity and specificity in differentiating PCV from neovascular AMD.^8,9 In an earlier investigation, assessment of a combination of color fundus photographs and OCT images provided high accuracy in differentiating PCV from central serous chorioretinopathy and differentiating PCV from AMD.¹⁰ In that report, examples of clinical clues for PCV detected using color fundus photography (FP) included a subretinal orange nodule, hemorrhagic or fibrovascular pigment epithelial detachment (PED), massive subretinal hemorrhage, peripapillary location of lesions, multifocal lesions, or no large drusen in the fellow eye; examples of clinical clues for PCV detected using OCT included a sharply peaked retinal pigment epithelium protrusion, notched or multilobulated PED, a hyperreflective ring surrounding a hyporeflective halo underneath the PED, or a double-layer sign; and examples of clinical clues for PCV detected using fluorescein angiography (FA) include an occult CNV leakage pattern. Although graders used this information in combination with their own background knowledge or experience to determine whether each case was PCV or not, no official grading standardization was performed. The results of this study suggested that, without ICGA, FP combined with OCT provides high sensitivity and high specificity to diagnose PCV. However, some differences were found among results from graders working inside vs outside areas where PCV is endemic.

Therefore, this investigation expands on that previous report to determine the accuracy of diagnosing PCV from prespecified potential diagnostic features detected using FP, OCT, or FA in the absence of ICGA. Knowing the value of each PCV feature detected using such imaging might be able to assist ophthalmologists with different clinical backgrounds to identify PCV in a standardized way in clinical care or clinical research when ICGA is not readily available, feasible, or desired.

Ophthalmologists also perform FA in some cases of suspected or confirmed PCV. However, the value or redundancy of specific features detected using any of these image modalities to detect PCV might be worth knowing to aid the physician as to what features to identify to achieve high accuracy of PCV diagnosis without the invasiveness of ICGA. Therefore, this study determined the sensitivity, specificity and predictive accuracy of prespecified potential diagnostic features detected using FP, OCT, and FA, alone and in combination, to diagnose PCV without the use of ICGA.

Individuals who presented to Chiang Mai University Hospital between January 1, 2013, and December 31, 2016, with newly diagnosed serous or serosanguinous maculopathy in 1 or both eyes and who underwent all 4 imaging types (FP, OCT, FA, and ICGA) before undergoing any treatments were identified and included. Eligible patients were identified from a list of patients who underwent ICGA at the ocular photography department at Chiang Mai University Hospital during the study period. Those with previous treatment (89 patients), those who did not receive all 4 imaging procedures (eg, missing FP, OCT, FA, or ICGA at their first visit; 17 patients), or those whose imaging results were of poor quality (20 patients) were excluded. This study was approved by the institutional review boards of Chiang Mai University and Johns Hopkins University. Patient consent was not required in this study as it is a retrospective review of deidentified data.

All images of each case were deidentified and categorized into 3 sets (FP, OCT, and FA). Subsequently, they were sequenced randomly and reviewed independently by 3 graders (J.K., M.S., and P.C.) to confirm the presence or absence of each of the following prespecified potential diagnostic features: (1) for FP, subretinal orange nodule, massive subretinal hemorrhage (4 disc area or larger), and notched or hemorrhagic PED (eFigure 1 in the Supplement), peripapillary lesion location (within 1 disc diameter surrounding the optic disc), multifocal lesions, and no large drusen (≥125 μm in diameter) in the fellow eye (eFigure 2 in the Supplement); (2) for OCT, multiple PEDs, sharply peaked PED (PED with an angle between 70° and 90°) (eFigure 3 in the Supplement), notched or multilobulated PED, hyperreflective ring surrounding hyporeflective halo beneath PED, and double-layer sign (eFigure 4 in the Supplement); and (3) for FA, classic CNV, occult CNV, peripapillary location, multifocal lesions, and expansile dot or smoke-stack leakage typical of central serous chorioretinopathy (eFigure 5 and eFigure 6 in the Supplement).

The images included 88 cases from a previous study¹⁰ but sent to graders who were not used in that previous study. Fundus photography included images of both the study eye and the fellow eye, OCT included at least 25 cross-sectional B-scan images of the study eye, and FA included multiple images from early, middle, and late phases of the study eye and the fellow eye. No en face OCT images or OCT angiography images were available at the time of this study. Majority rule was applied to determine the group decision of graders; opinions of at least 2 of 3 graders were determined as final results. In parallel, 2 other graders (V.C. and D.P.) reviewed all the images (including ICGA) to confirm the presence or absence of PCV in each case based on the EVEREST criteria,¹ including the presence of subretinal focal ICGA hyperfluorescence within 6 minutes after injection of indocyanine green and at least 1 of the following additional criteria: (1) association with abnormal vascular channel supplying the polypoidal lesion, (2) presence of pulsatile polypoidal lesions, (3) nodular appearance of polypoidal lesion on stereoscopic viewing, (4) presence of hypofluorescent halo around the nodules (in the first 6 minutes), (5) orange subretinal nodules on FP corresponding to the hyperfluorescent area identified by ICGA, or (6) association with massive submacular hemorrhage (at least 4 disc areas). Any disagreement between these 2 graders was reviewed and discussed using open adjudication.

Diagnostic parameters, including sensitivity, specificity, positive predictive value, negative predictive value, and predictive accuracy (area under the receiver operating characteristic curve [AUC]) of FP, OCT, and FA to diagnose PCV were determined. Given that the perfect score of predictive accuracy or AUC is 1, this study preferred features with an AUC of 0.8 or more. Subsequently, to select the final potential diagnostic features for the diagnosis of PCV, the stepwise method¹¹ based on the Akaike information criterion (AIC) was used to analyze features with an AUC of 0.8 or more. The variables were removed from the model by the decreasing order of AIC, until the reduced model with the lowest AIC (final model) was established. The 4 major criteria were the variables that were presented in the final model. The remaining features were considered minor criteria. The diagnostic parameters mentioned previously were calculated to determine the optimal number of features to diagnose PCV.

This study included 124 eyes of 120 patients. The mean (SD) age of the patients was 57.7 (12.6) years, 100% were of Thai ethnicity, 52 were women, and 68 were men. Of the 124 study eyes, the definitive diagnosis by the expert retina specialists who had all 4 imaging modalities (FP, OCT, FA, and ICGA) was PCV in 65 eyes (52.4%), central serous chorioretinopathy in 45 eyes (36.3%), typical neovascular AMD in 12 eyes (9.7%), idiopathic CNV in 1 eye (0.8%), and retinal angiomatous proliferation in 1 eye (0.8%).

The sensitivity, specificity, and AUC for the potential diagnostic features detected using FP are shown in Table 1; the sensitivity, specificity, and AUC for the potential diagnostic features detected using OCT are shown in Table 2; and the sensitivity, specificity, and AUC for the potentially diagnostic features detected using FA are shown in Table 3. Note that an AUC of 0.8 or higher refers to high predictive accuracy and an AUC less than 0.8 refers to relatively lower accuracy. On FP, the diagnostic feature of a subretinal orange nodule had high specificity (0.92; 95% CI, 0.82-0.97) but rather low sensitivity (0.39; 95% CI, 0.27-0.52) to identify PCV based on ICGA. Many features detected using OCT (Table 2) showed an AUC of 0.8 or higher (multiple PED, sharply peaked PED, notched or multilobulated PED, hyperreflective ring underneath PED, and double-layer sign). No features detected using FA showed an AUC of 0.8 or higher (Table 3).

Of the features with an AUC of 0.8 or higher, a logistic regression model based on AIC using a backward and forward stepwise method provided the final 4 potential diagnostic features (Figure), which included a notched or hemorrhagic PED detected using FP (AUC, 0.77; 95% CI, 0.70-0.85), a sharply peaked PED at an angle of 70° to 90° detected using OCT (AUC, 0.86; 95% CI, 0.80-0.92), a notched or multilobulated PED detected using OCT (AUC, 0.90; 95% CI, 0.85-0.96), or a hyperreflective ring underneath PED detected using OCT (AUC, 0.86; 95% CI, 0.80-0.92).

When these 4 potential diagnostic features were considered as major criteria, and all other features as minor criteria, the sensitivity and specificity of a diagnosis of PCV using multiple features can be seen in Table 4. If at least 2 of 4 major criteria were present, the predictive accuracy (AUC) of PCV was 93% (95% CI, 89%-98%), with 95% sensitivity (95% CI, 87%-99%), 95% specificity (95% CI, 82%-97%), 92% positive predictive value (95% CI, 83%-97%), and 95% negative predictive value (95% CI, 86%-99%). In other words, for clinical application of positive predictive value and negative predictive value, these results imply that if physicians find at least 2 of 4 major criteria in an eye with serous or serosanguinous maculopathy, 92 of 100 cases will have a diagnosis of PCV based on our criterion standard of 2 retina specialists evaluating FP, OCT, FA, and ICGA imaging results. If physicians look for all 4 major criteria and at least 2 features were not found (none or only 1 feature was found), 95 of 100 cases will not have a PCV diagnosis. When only 1 of 4 major criteria were present, sensitivity was high (0.98; 95% CI, 0.92-1.00), but specificity was low (0.67; 95% CI, 0.53-0.78). When only 1 major criterion was present, with 1 or more minor criteria, the AUC decreased to 43%, and when only 1 major criterion was present, with 2 or more minor criteria, the AUC decreased to 52%; these AUCs would be considered relatively low accuracy for diagnosing PCV in the absence of ICGA.

Although, to our knowledge, there is no universal recognition of PCV currently, a well-accepted definition of PCV from 2 independent randomized clinical trials is based on the presence of subretinal nodular hyperfluorescence detected using ICGA.^1,7 However, ICGA is neither routinely performed in diagnosing exudative AMD nor readily available. This limitation could make it difficult to diagnose PCV when lacking the criterion standard of ICG and could impede the possibility of applying results of clinical trials on PCV that used ICGA or performing a clinical research on PCV when access to ICGA is limited. Furthermore, it is hypothesized that PCV may be underdiagnosed in people who are not Asian or of African ancestry, in areas where ICGA may be performed even less frequently than in countries with a majority of individuals who are of Asian or African ancestry in which PCV is more common than elsewhere. Noninvasive and feasible diagnostic features with both high sensitivity and specificity would be valuable to improve the ability of diagnosing PCV without performing ICGA.

To address this need, this study evaluated the sensitivity, specificity, and accuracy of prespecified potential diagnostic features detected using FP, OCT, and FA, alone and in combination, to diagnose PCV without performing ICGA. The results of this study suggest that when at least 2 of 4 highly suggestive features detected using FP and OCT are present, there is 95% sensitivity (95% CI, 87%-99%) and 95% specificity (95% CI, 82%-97%) of a diagnosis of PCV. The 4 highly suggestive features for PCV include a notched or hemorrhagic PED detected using FP, a sharply peaked PED at an angle of 70° to 90° detected using OCT, a notched or multilobulated PED detected using OCT, and a hyperreflective ring underneath PED detected using OCT. Physicians should consider looking for these features when PCV is one of the differential diagnoses.

Several other studies have evaluated the diagnostic value of spectral-domain OCT for patients with PCV. A retrospective study showed that sharply peaked PED, notched or multilobulated PED, a hyporeflective lumen, or multiple PEDs are 4 signs that have both high sensitivity (94.6%) and specificity (92.9%) to differentiate PCV from occult CNV; however, the sample size was relatively small, with only 51 eyes of 44 patients included.⁸ Furthermore, the inclusion criteria of the study were set to 1 or more PEDs in at least 1 eye, which excluded patients with PCV who did not have a PED. Liu et al⁹ performed a prospective study with a broader spectrum of patient inclusion criteria and a relatively larger sample size to distinguish PCV from typical CNV in AMD by using OCT and found a lower sensitivity of 89.4% and a lower specificity of 85.3% in diagnosing PCV. However, in contrast to our present investigation, these studies did not consider the information potentially added by FP as a proxy for, or in place of, features noted during fundus examination when added to features detected using OCT, with or without potential information added from FA, to differentiate PCV from CNV in AMD.

According to the proposed guidelines for the diagnosis of PCV by the Japanese Study Group of Polypoidal Choroidal Vasculopathy,¹² definite cases of PCV should meet at least 1 of the following criteria: protruding orange-red elevated lesions observed during fundus examination or characteristic polypoidal lesions detected using ICGA. However, in our study, judging by the use of FP, the diagnostic feature of a subretinal orange nodule has high specificity (92%) but rather low sensitivity (39%) to identify PCV based on ICGA. Compared with FP, our study found that clinically relevant features detected using OCT had relatively lower specificity but higher sensitivity.

These results suggest that OCT alone may not be powerful enough to identify PCV in the absence of ICGA. Using a logistic regression model, we propose 4 highly suggestive features (notched or hemorrhagic PED detected using FP, a sharply peaked PED detected using OCT, notched or multilobulated PED detected using OCT, or a hyperreflective ring surrounding hyporeflectivity detected using OCT) as major criteria to identify PCV in the absence of ICGA, while other signs detected using FP and OCT are proposed as minor criteria. Identifying PCV by 1 of the 4 major criteria proposed could provide a high sensitivity of 98% but would have a lower specificity of 67%. In contrast, identifying at least 2 of these 4 major criteria had high specificity (95%), sensitivity (95%), positive predictive value (92%), and negative predictive value (95%).

One benefit of considering these 4 major criteria to identify PCV in the absence of ICGA is that noninvasive FP and OCT are routinely used and could be accessed easily in most clinics and most clinical trials. Results from this study also confirmed that FA may be unnecessary for the diagnosis of PCV in most cases when FP and OCT are available, as previously reported,¹⁰ because all FA features provided a relatively low AUC (<0.7) for PCV diagnosis.

Identifying PCV in the absence of ICGA may preclude identifying polypoidal lesions for which photodynamic therapy (PDT) with verteporfin might be considered in combination with ranibizumab, as performed in EVEREST2.⁶ However, 1-year results of the PLANET (Aflibercept in Polypoidal Choroidal Vasculopathy) study showed that improvement in visual and functional outcomes could be achieved for most participants with intravitreous aflibercept monotherapy, in the absence of rescue PDT for more than 85% of study participants.⁷ Although EVEREST2 showed that a combination of PDT and ranibizumab provides superior visual acuity outcomes with slightly fewer injections compared with ranibizumab monotherapy in the absence of rescue PDT,⁶ it remains unknown whether aflibercept combined with PDT provides superior visual acuity results with fewer injections compared with aflibercept monotherapy with rescue PDT. At this time, ICGA still may be needed to identify polypoidal lesions to which PDT should be applied if PDT is planned in combination with anti–vascular endothelial growth factor therapy at the initiation of treatment, as in EVEREST2,⁶ or as rescue therapy as in PLANET.⁷

The limitations of this study include its retrospective design; it remains to be seen whether clinicians can identify PCV in the absence of ICGA with similar sensitivity and specificity. Furthermore, this study was performed only among individuals of Thai ethnicity; therefore, it remains to be seen whether these results can be generalized to other Asian populations or those of African ancestry, for whom PCV is endemic, or even among white individuals with PCV. Also, this study evaluated only treatment-naive eyes. It is uncertain whether the findings from this study are applicable to eyes that have undergone previous treatment. Imaging technologies (eg, OCT angiography¹³ or en face OCT¹⁴) that were recently reported to be beneficial in detecting PCV were not included in this study because those images were not available at our center during the study period. It remains to be seen how the potential diagnostic features from this study might be affected when more recently developed imaging techniques are applied in these situations.

This study suggests that prespecified potential diagnostic features detected using FP and OCT of Thai individuals with serous or serosanguinous maculopathy provide high sensitivity and specificity to diagnose PCV without ICGA, especially when at least 2 of 4 highly suggestive signs are present, including a notched or hemorrhagic PED detected using FP or OCT, a sharply peaked PED detected using OCT, or a hyperreflective ring detected using OCT. Further studies to determine the generalizability of these findings, as well as their usefulness and potential cost savings in the clinical practice setting, seem to be warranted.

Accepted for Publication: February 4, 2019.

Corresponding Author: Neil M. Bressler, MD, Wilmer Eye Institute, Retina Division, Johns Hopkins University, 600 N Wolfe St, Maumenee 752, Baltimore, MD 21287 (nmboffice@jhmi.edu).

Published Online: April 11, 2019. doi:10.1001/jamaophthalmol.2019.0565

Author Contributions: Drs Chaikitmongkol and Khunsongkiet had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Chaikitmongkol, Khunsongkiet, Patikulsila, Winaikosol, Choovuthayakorn, Ingviya, Bressler.

Acquisition, analysis, or interpretation of data: Chaikitmongkol, Kong, Sachdeva, Chavengsaksongkram, Dejkriengkraikul, Winaikosol, Watanachai, Kunavisarut, Ingviya, Bressler.

Drafting of the manuscript: Chaikitmongkol, Kong, Khunsongkiet, Patikulsila, Dejkriengkraikul, Kunavisarut, Ingviya.

Critical revision of the manuscript for important intellectual content: Kong, Sachdeva, Chavengsaksongkram, Winaikosol, Choovuthayakorn, Watanachai, Ingviya, Bressler.

Statistical analysis: Chaikitmongkol, Ingviya.

Obtained funding: Bressler.

Administrative, technical, or material support: Kong, Khunsongkiet, Patikulsila, Sachdeva, Chavengsaksongkram, Winaikosol, Choovuthayakorn, Watanachai, Kunavisarut.

Conflict of Interest Disclosures: Dr Chaikitmongkol reported receiving research grants from Bayer and ThromboGenics and travel expenses from Allergan, Bayer, and Novartis. Dr Khunsongkiet reported receiving travel expenses from Bayer. Dr Patikulsila reported working as a consultant and receiving honoraria and travel expenses from Bayer and Novartis and receiving honoraria and travel expenses from Alcon. Dr Chavengsaksongkram reported receiving travel expenses from Bayer. Dr Choovuthayakorn reported receiving honoraria and travel expenses from Alcon, Allergan, Bayer, and Novartis. Dr Watanachai reported receiving honoraria and travel expenses from Alcon, Allergan, Bayer, and Novartis. Dr Kunavisarut reported receiving honoraria and travel expenses from Bayer and Novartis. Dr Bressler reported holding a patent on a system and method for automated detection of age-related macular degeneration and other retinal abnormalities and receiving grants from Bayer, Genentech/Roche, Novartis, and Samsung Bioepis outside the submitted work. No other disclosures were reported.

Funding/Support: This work was supported in part by Research Committee, Faculty of Medicine, Chiang Mai University; the James P. Gills Professorship, the China Research Council; and unrestricted research funds to the Johns Hopkins University School of Medicine Retina Division for Macular Degeneration and Related Diseases Research.

Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Disclaimer: Dr Bressler is editor of JAMA Ophthalmology, but was not involved in the editorial evaluation or decision to accept this article for publication.

Meeting Presentation: This paper was presented at the Annual Meeting of the Association for Research in Vision and Ophthalmology; May 1, 2018; Honolulu, Hawaii.