Telemedicine Screening for Cytomegalovirus Retinitis Using Digital Fundus Photography

Introduction

Cytomegalovirus retinitis (CMVR) is the most common ocular infection affecting acquired immune deficiency syndrome (AIDS) patients and is an important cause of blindness among these patients. ^1,2 It is found in up to a third of AIDS patients in Southeast Asia. ¹ The incidence of CMVR is higher in patients with CD4⁺ counts <100 cells/μL. ³ In the absence of immune recovery from highly active antiretroviral therapy (HAART), it is associated with a 60% increase in mortality. ⁴ With modern-day HAART, CMVR incidence has declined 60–75%, ⁵ and patients with initial opportunistic infections are also recovering and surviving longer. ⁵ However, in patients who have commenced HAART, undiagnosed and untreated CMVR has an increased risk of developing immune reconstitution uveitis, a significant cause of visual morbidity among this group. ⁶ Thus underlying CMVR needs to be identified in both HAART-naive and -treated patients to prevent visual complications and irreversible visual loss.

Currently, the majority of 34 million HIV cases worldwide occur in low- to middle-income countries. ⁷ Many human immunodeficiency virus (HIV) patients still progress to AIDS with decreased CD4⁺ counts because of a lack of diagnostic facilities, financial constraints, and late diagnosis, in both developed and especially developing countries. ⁴ This increases their risk of complications from opportunistic infections such as CMVR. Therefore, there are large numbers of patients susceptible to CMVR requiring regular screening. In this modern day, with increased accessibility to HAART, there are increasing public programs worldwide for early HIV detection and treatment to prevent AIDS. However, similar screening programs for CMVR are uncommon. ⁸ Many developing countries lack adequate facilities and healthcare personnel trained in retinal disease to diagnose CMVR. As such, the vast majority of patients requiring regular CMVR screening will simply not receive it as specialist care is both limited and inaccessible.

We aimed to report the use of composite montage images derived from nine-field digital fundus photography (DFP) for CMVR screening. We report its sensitivity and specificity in detecting CMVR and establish the level of agreement with indirect fundus ophthalmoscopy, the gold standard for diagnosis of retinal pathologies.

Materials and Methods

We conducted an audit of our CMVR screening program for HIV patients referred to the Department of Ophthalmology at the Center for Communicable Diseases, the national center for HIV management, at Tan Tock Seng Hospital, Singapore. All patients with any of the following criteria were recruited for screening: (a) any visual symptoms and/or history of (b) CD4⁺ count <50 cells/μL, (c) AIDS defining illness, and (d) opportunistic infections. An AIDS-defining illness referred to the list of diseases published by the U.S. Centers for Disease Control and Prevention, which may be used to define AIDS. ⁹ An opportunistic infection was defined as an infection caused by a microorganism that did not ordinarily cause disease but was capable of doing so under impaired host immunity. ¹⁰ The audit was conducted over a 12-month period. Basic epidemiological data were recorded on a standardized form that included demographic characteristics (age, gender, ethnicity) and reason(s) for referral.

After pupil dilation with tropicamide (1%) eyedrops, patients underwent DFP by a qualified technician using a Zeiss (Oberkochen, Germany) FF450 fundus camera with a Kodak (Rochester, NY) DCS620 digital-back. Nine standardized fields of 50° each were captured and stored in raw .tiff format. Automated panoramic reconstruction of the nine images into one composite montage image of the retinal fundus was performed. The composite image was read by a retinal specialist blinded to the patient's history. Features suggestive of CMVR, as well as other features such as vitreous haze and HIV retinopathy, were noted. CMVR was diagnosed based on the characteristic clinical appearance of lesions, typically consisting of an area of retinal opacification (necrosis or edema) surrounded by granular infiltrates and a silvery-white border marking the edge of the active borders, with variable amounts of retinal hemorrhage and inflammatory vascular sheathing. ¹¹ Each fundus photograph was then classified as “active CMVR positive,” “active CMVR negative,” “CMVR suspicious,” or “unreadable/poor view.” Patients previously diagnosed with CMVR with inactive current residual lesions were not considered to have active CMVR.

Diagnosis was confirmed with dilated indirect ophthalmoscopy by another trained ophthalmologist after each patient was asked for visual symptoms. Indirect ophthalmoscopy is the gold standard for diagnosis of any retinal pathology, including CMVR. A detailed retinal drawing was made for each eye, with the location of disease categorized into three zones: Zone 1 encompassed an area one disc diameter from the edge of the optic nerve or two disc diameters from the center of the fovea, Zone 2 extended from the edge of Zone 1 approximately to the equator as marked by a circle identified by the vortex vein ampullae, and Zone 3 extended anteriorly from the edge of Zone 2 to the ora serrata. ⁴ Each fundus was again classified as “active CMVR positive,” “active CMVR negative,” or unreadable/poor view. CMVR-positive patients, defined as having CMVR in at least one eye, would exit the audit and be treated appropriately according to protocol. ¹¹ Patients who were CMVR-negative and had “unreadable/poor” fundi would continue to return for screening of both eyes at 3-month intervals, for resolution of media opacity and screening of the fellow eye.

Descriptive analysis was performed with continuous variables summarized either as means with standard deviations or as medians with ranges. The independent-samples t test, Fisher's exact test, and the Mann–Whitney test were used for comparison between the CMVR-positive and non-CMVR patients' ages, gender, and CD4⁺ count, respectively. Using indirect ophthalmoscopy as the gold standard, sensitivity, specificity, and kappa statistics for DFP were calculated. Data analysis was performed using IBM SPSS Statistics (version 19; IBM Corp., New York, NY).

Results

Discussion

Telemedicine diagnosis with fundus photography is particularly suited for diagnosis of retinal diseases such as CMVR as diagnosis is clinical, based on typical fundus appearances. Telemedicine has already been shown to be effective in screening for diabetic retinopathy and retinopathy of prematurity. ^12,13 With increased Internet connectivity and telecommunication worldwide, telemedicine is able to reach out to patients even in remote locations. Furthermore, telemedicine is cost-effective for both doctor and patient, as manpower and traveling costs of screening visits are reduced. These are important, as many HIV patients are financially challenged and lack accessibility to specialist care. As such, examining at-risk HIV patients for CMVR in underserved areas remotely via telemedicine with DFP is a novel solution of reaching out to patients who would otherwise not have access to specialist ophthalmological care. This study investigated telemedicine's validity against traditional gold-standard methods for CMVR screening before institution of possible screening programs.

Successful implementation of such a screening program in rural areas depends on feasibility of application. Retinal cameras are very common, and technicians are easily trained to perform retinal fundus photography with nine-field views obtained by having the patient look in different directions.

Cost-effectiveness is also important for telemedicine to be widely used in developing countries. One of the main factors is manpower cost associated with the evaluating reviewers. In assessing the feasibility of telemedicine screening, we used expert graders (consultant retinal specialists) to avoid any limitation due to observer expertise. With feasibility now demonstrated from our study, we propose that the evaluation of photographs can be performed by trained retinal graders whom we feel will be able to perform reading of fundus photographs as it requires identification of defined visual characteristics; this identification can be trained. The high rate of CMVR identification was promising given that, unlike the clinical reviewers, the photo reviewers did not receive any patient information. This suggests that even when remotely screening for CMVR, photo reviewers need not have detailed clinical information and that fundus photographs alone are adequate for CMVR diagnosis. Without the apparent need to evaluate patient information, the graders do not require specialist medical knowledge, and thus the sourcing and training of non-expert technician graders for evaluating fundus photographs are simplified.

Telemedicine allows mobility of trained technicians, decreasing specialist care manpower requirements. The screening process can be further streamlined with the help of defined enrollment criteria for the screening program to identify and screen only those patients most at risk. This then allows ophthalmologists to focus their time and energies on patients with significant pathology.

The general quality of our photographs was good as evidenced by the fact that the same 16 eyes were unreadable on both DFP and indirect ophthalmoscopy. Thus clinical examination did not aid in clarifying any of the unreadable photos. We also found that a single composite nine-field image per patient was adequate to determine CMVR presence, without needing to additionally evaluate each of the nine-field images separately. Evaluation of a single composite image, which can achieve up to a 95–105° field of view, saves time and possibly reduces reviewer fatigue from reading multiple images for 1 patient. We propose that a single high-resolution composite image without under- or overlap may be adequate for CMVR screening.

In our study, we showed that DFP was an effective method of screening for CMVR with 100% sensitivity and 99.9% specificity compared with indirect ophthalmoscopy. There was also substantial agreement between both methods (kappa value of 0.739). ¹⁴ This is in agreement with a study by Ausayakhun et al. ⁸ in Northern Thailand showing that telemedicine can be an accurate and reliable way of screening for CMVR. It is noteworthy that the level of agreement between DFP and indirect fundus ophthalmoscopy rose from 0.739 to 0.987 when “CMVR-suspicious” patients were excluded. Indirect ophthalmoscopy is the gold standard for CMVR diagnosis, and thus no case was classified as “suspicious” on indirect ophthalmoscopy. In our study, all cases found to be suspicious by DFP were diagnosed as “CMVR-negative” after indirect ophthalmoscopy. There was almost perfect agreement between the two methods for “CMVR positive/negative” and “unreadable” fundi, ¹⁴ and the differences in kappa values were due to the “suspicious” lesions caused by artifacts and illumination limitations on DFP. Therefore, we conclude that DFP has a high level of agreement with indirect ophthalmoscopy.

The reliability of our study was increased with a large study population. Furthermore, 34.0% had been followed up for at least 6 months, and almost half (48.9%) underwent two or more screenings. Patients at risk of CMVR with risk factors such as low CD4⁺ counts or history of AIDS-defining illness or opportunistic infection may develop CMVR over time, and thus multiple screening visits allowed numerous opportunities to capture CMVR, if any. However, we acknowledge our study had some limitations. One possible limitation was observer bias as the ophthalmologist who took the history also conducted the ophthalmic examination for the same patient, and thus the diagnosis of CMVR may have been influenced by the knowledge of a patient's CD4⁺ count, presence of symptoms, and AIDS-defining illnesses. However, we believe that this mirrors the clinical setting more closely. Second, because the clinical reviewers were trained ophthalmologists, the influence of a clinical history on clinical examination should be minimal. Third, it is theoretically possible that DFP may miss small lesions within Zone 3. However, only 2% of newly diagnosed CMVR patients have lesions confined to Zone 3 at diagnosis. ¹⁵ CMVR progresses at an average rate of 24 μm/day in patients who were not on anti-cytomegalovirus treatment or HAART. ¹⁶ Even if small lesions in Zone 3 were present and undetected on DFP, we believe that our screening time interval of 3 months was sufficient to detect the lesions prior to development of significant visual loss secondary to macula or optic disc involvement, the major cause of CMVR-related visual loss. ⁴ Nonetheless, the advent of ultra-wide-field imaging such as with the Optomap^® imaging system (Optos PLC, Dunfermline, Scotland, United Kingdom) may increase the peripheral view. Mudvari et al. ¹⁷ reported that the Optomap captured 48% greater retinal area and 40% greater CMVR area compared with standard nine-field photography when imaging for CMVR.

We have demonstrated that, compared with indirect ophthalmoscopy by trained ophthalmologists, composite nine-field digital fundus photography is a useful, sensitive, and specific screening tool for CMVR detection in HIV patients and that there is a high level of agreement between both techniques. Based on our results, only patients who were positive or suspicious for disease, or those with unreadable photographs, will need to be recalled for clinical review and treatment, saving much time for most patients and reducing clinic workload for ophthalmologists. It is hoped that this will be adopted by many communities worldwide, especially in rural areas or places where specialty care to detect and treat early CMVR is limited, hence reducing the incidence of permanent visual loss from CMVR.