Results are defined based on the consensus grade from 2 of 3 photographic graders relative to the in-clinic ophthalmologist. A, Eyes with true-positive results. B, Eyes with true-negative results. C, Eyes with false-positive results. D, Eyes with false-negative results.
Jirawison C, Yen M, Leenasirimakul P, Chen J, Guadanant S, Kunavisarut P, Patikulsila D, Watanachai N, Ausayakhun S, Heiden D, Holland GN, Margolis TP, Keenan JD. Telemedicine Screening for Cytomegalovirus Retinitis at the Point of Care for Human Immunodeficiency Virus Infection. JAMA Ophthalmol. 2015;133(2):198-205. doi:10.1001/jamaophthalmol.2014.4766
Cytomegalovirus (CMV) retinitis is a leading cause of blindness in many developing countries, likely the result of inadequate screening. Telemedicine screening for CMV retinitis instituted at the point of care for human immunodeficiency virus (HIV) infection may allow for earlier detection.
To determine the diagnostic accuracy of retinal photography in detecting CMV retinitis at the point of HIV care and to characterize the clinical manifestations of CMV retinitis detected through the screening program.
Design, Setting, and Participants
We enrolled 103 participants from a population of 258 patients with HIV and a CD4 level of less than 100/μL treated at an HIV clinic in Thailand from June 2010 through June 2012. We captured mosaic fundus photographs through a dilated pupil using a digital fundus camera. An experienced on-site ophthalmologist masked to the results of the fundus images subsequently examined each eye with indirect ophthalmoscopy and recorded the clinical findings on a standardized form. Three remote graders evaluated each image for CMV retinitis.
Fundus photography and indirect ophthalmoscopy.
Main Outcomes and Measures
Sensitivity and specificity of telemedicine relative to indirect ophthalmoscopy for diagnosis of CMV retinitis and clinical features of CMV retinitis lesions.
Sixteen patients (15.5%) were diagnosed as having CMV retinitis, of whom 5 (31%) had bilateral disease. Of the 21 eyes (10.2%) with CMV retinitis, 7 (33%) had visual symptoms. Retinitis lesions occupied less than 10% of the total retinal surface area in 13 of 21 eyes (62%) and did not involve the posterior pole (ie, zone 1) in 15 of 21 eyes (71%). Mean logMAR visual acuity in affected eyes was 0.41 (95% CI, 0.11-0.71; Snellen equivalent, 20/50 [95% CI, 20/25-20/100]). The mean sensitivity for the 3 remote graders in detecting CMV retinitis on fundus photography was 30.2% (95% CI, 10.5%-52.4%), and mean specificity was 99.1% (95% CI, 97.8%-100.0%). The CMV retinitis lesions missed by the remote graders (false-negative findings) were more likely to be small (P = .001) and located in the peripheral retina (P = .04).
Conclusions and Relevance
Patients undergoing screening at a clinic for HIV treatment had less extensive retinitis than patients in recent reports from an ophthalmology clinic. Retinal photography with the camera used in this study was not highly sensitive in detecting CMV retinitis but may identify disease with an immediate threat to vision. Improved accuracy will require a camera that can more easily image the peripheral retina.
Cytomegalovirus (CMV) retinitis is a leading cause of blindness in many developing countries with a high burden of AIDS.1- 5 The condition causes full-thickness retinal necrosis and irreversible blindness if left untreated.6,7 Therefore, detection of CMV retinitis before the patient loses vision is important. The standard of care is to perform screening for patients who have a CD4 cell count of less than 100/μL (to convert to ×109/L, multiply by 0.001) with indirect ophthalmoscopy every 3 months. However, screening is not performed routinely in many developing countries owing to limited ophthalmic care.8- 11 Patients often have severe manifestations of disease by the time they seek care.12,13 New strategies are needed to increase the number of patients who are diagnosed as having CMV retinitis before the disease causes vision loss. Because CD4 testing is the primary indication for screening, it would be most effective if screening were initiated by the clinician treating the patient for human immunodeficiency virus (HIV) infection.2,13,14
Telemedicine using retinal photography is a potential method to screen for CMV retinitis in a primary care setting. Prior studies15- 17 found that telemedicine has high sensitivity and specificity in detecting CMV retinitis. However, these studies recruited participants from ophthalmology centers. In practice, telemedicine screening would occur in HIV clinics. Whether the diagnostic performance of telemedicine would differ in an HIV clinic, where nonophthalmic personnel would obtain photographs and the disease would likely be less extensive, has not been ascertained. The purpose of the present study was 2-fold. First, we assessed the diagnostic accuracy of fundus photography in detecting CMV retinitis when used in the setting of an HIV clinic. Second, we determined whether the cases identified at the point of HIV care have less extensive disease than those reported in prior studies of ophthalmology clinic–based screening programs.
Ethical approval was obtained from the Committee on Human Research at the University of California, San Francisco, and the institutional review board of the Nakornping Hospital, Chiang Mai, Thailand. This study adhered to the tenets of the Declaration of Helsinki. All participants gave written informed consent before enrollment. No patients were harmed during data collection.
From June 18, 2010, through June 15, 2012, we enrolled patients with a CD4 cell count of less than 100/μL from the HIV clinic at Nakornping Hospital, a tertiary medical center. We intended to enroll all consecutive patients, which proved to be logistically difficult. Patients who were pregnant, who were younger than 18 years, or who carried a previous diagnosis of CMV retinitis were excluded. We trained an HIV clinic staff member (S.G.) with no previous experience in ophthalmology or fundus photography to use a digital fundus camera (sensor resolution, 10.1 megapixels) (Topcon TRC-NW6S; Topcon). In a single session before enrolling participants, we trained the staff member to use the internal fixation settings to obtain a standard set of 9 overlapping 45° fundus images. Her skills were deemed to be adequate by the end of this session. We directly monitored all photography for the first 6 weeks and periodically thereafter. This staff member took all photographs during the study. Images of both eyes from each patient were captured after dilating each eye with tropicamide, 1%. A composite mosaic image covering an 85° retinal field was created using software for image alignment and mosaic construction (i2kRetina; DualAlign LLC). The photographer was masked to all clinical and demographic information about the patient.
A fellowship-trained retina specialist with extensive experience diagnosing and treating CMV retinitis (C.J.) subsequently examined each eye for CMV retinitis using indirect ophthalmoscopy while masked to the findings of the fundus photography. Clinical characteristics of retinitis lesions were recorded on a standardized form. For each eye with CMV retinitis, a detailed drawing of the lesion was made on a template that showed important retinal landmarks and the 3 zones used to denote CMV retinitis location.18 The in-clinic ophthalmologist categorized retinitis lesion size as a percentage of total retinal surface area and assessed the degree of opacity of the lesion border against standard photographs from a previously described system.18 Study participants were offered repeated examinations and photography every 3 months until their CD4 cell count increased to 100/μL or greater. Eyes that developed CMV retinitis at any point in the study subsequently underwent censoring.
The mosaic image for each eye was uploaded to a secure server and transferred to 3 fellowship-trained retina specialists with expertise in diagnosing CMV retinitis (P.K., D.P., and N.W.). After completing a training session, these remote graders used a standardized form to assign to each image a diagnosis of CMV retinitis present, absent, or unknown. Experts also determined the gradability of each image. Gradability was defined as good if the image was in focus and covered the full 85° field; acceptable if the image was slightly out of focus or did not cover the full 85° field, but a definitive diagnosis could be made; and poor if the image was so out of focus or covered such a small degree of the retina that a definitive diagnosis could not be made. Eyes underwent independent evaluation in random order. Graders received no demographic information and were masked to the clinical diagnosis that was made on indirect ophthalmoscopy and by the other graders.
We calculated the sensitivity and specificity of each grader relative to the in-clinic examination as the reference standard. The purpose of telemedicine screening is to identify all potential cases of CMV retinitis for referral to an ophthalmologist. Therefore, we considered CMV retinitis present and unknown status to be positive diagnoses. Sensitivity and specificity were calculated using the individual eyes as the unit of analysis; we also performed a secondary analysis in which we considered the patient as the unit of analysis. For eye-level analysis, eyes were excluded if the on-site ophthalmologist could not determine with certainty the presence or the absence of CMV retinitis. For patient-level analysis, the study participant was excluded if the on-site ophthalmologist diagnosed both eyes as unknown status or diagnosed one eye as unknown status and the other as CMV retinitis absent. Participants were considered to have a positive diagnosis for CMV retinitis if at least 1 eye had CMV retinitis and to have a negative diagnosis if neither eye had CMV retinitis.
Positive and negative predictive values were determined using the prevalence of CMV retinitis in this study population. We used the Cohen κ coefficient to measure intragrader agreement for each remote grader using 50 randomly selected and presented repeated images and to measure intergrader agreement among all 3 graders. To account for the nonindependence of 2 eyes from the same patient and multiple images of the same eye on different visits, we used the bootstrap method to calculate 95% CIs for diagnostic test statistics with resampling at the patient level (9999 repetitions). We calculated P values using the Fisher exact test for categorical variables and the Wilcoxon rank sum test for continuous variables. We used commercially available software (Stata, version 13; StataCorp) to perform all statistical analysis.
We enrolled 103 of 258 eligible individuals (ie, patients treated in the HIV clinic during the study period with a CD4 cell count of <100/μL as documented by the hospital laboratory), 30 of whom underwent multiple screens. Individuals who enrolled were more likely to be male (63 participants [61.2%] compared with 71 nonparticipants [45.8%]) but were otherwise similar to those who did not enroll. The mean (SD) age was 37.5 (9.1) years for participants vs 37.8 (8.4) years for nonparticipants, and the mean (SD) CD4 cell count was 29.5/μL (18.6/μL) in participants vs 29.5/μL (19.2/μL) in nonparticipants. In total, we performed 277 separate eye examinations of 205 eyes. Table 1 shows the clinical and demographic characteristics of the study population.
The study-site ophthalmologist diagnosed CMV retinitis by indirect ophthalmoscopy in 21 distinct eyes (10.2% [95% CI, 6.5%-15.2%]) from 16 distinct study participants (15.5% [95% CI, 9.1%-24.0%]). Cytomegalovirus retinitis was detected at the initial screening examination in 17 eyes of 13 participants and at the second screening examination (3 months after the initial screening examination) in 4 eyes of 3 participants. The on-site ophthalmologist could not give a definitive diagnosis in 5 eyes of 4 participants. All 5 eyes were excluded from the eye-level analysis, and 3 of the 4 participants were excluded from the person-level analysis. The participant not excluded from the patient-level analysis had CMV retinitis in one eye and a diagnosis of unknown status in the contralateral eye.
Five of 16 participants with CMV retinitis (31%) had bilateral disease at the time of diagnosis. Of the 21 eyes diagnosed as having CMV retinitis, the mean logMAR visual acuity was 0.41 (95% CI, 0.11-0.71; Snellen equivalent, 20/50 [95% CI, 20/25-20/100]). Seven of 21 eyes (33%) in 6 of 16 participants (38%) had visual symptoms at the time of diagnosis. The extent of retinitis was less than 10% of the total retinal surface area in 13 of 21 eyes (62%); only 6 of 21 eyes (29%) had lesions measuring 25% or more of the total retinal surface area. Lesions extended into zone 1 in 6 eyes (29%) and spared zone 1 in 15 (71%).
The 3 retina specialists remotely graded the 272 eligible images. Photographic quality was high, with 262 of 272 photographs deemed to be of acceptable or good quality by at least 2 of the 3 graders (Figure). The diagnostic accuracy of each grader relative to indirect ophthalmoscopy is shown in Table 2. Among remote graders, the mean sensitivity was 30.2% (95% CI, 10.5%-52.4%) and mean specificity was 99.1% (95% CI, 97.8%-100.0%) for diagnosis of CMV retinitis in individual eyes. Assuming the study prevalence of 10.2%, mean positive and negative predictive values were 78.7% (95% CI, 61.7%-89.4%) and 92.6% (95% CI, 91.4%-93.6%), respectively. The intrarater κ statistic was 1.00 for all graders; the interrater κ statistic was 0.92 (95% CI, 0.74-1.00). Sensitivity was slightly higher and specificity slightly lower in the patient-level analysis (Table 2).
Table 3 shows the characteristics of the 21 eyes diagnosed as having CMV retinitis by indirect ophthalmoscopy, stratified by whether CMV retinitis was detected by at least 2 remote graders. The 15 eyes for which the graders did not diagnose CMV retinitis (ie, false-negative findings) had smaller (P = .001) and more peripheral (P = .04) lesions with less border opacity (P < .001). To assess which lesion characteristics were most important, we stratified false-negative findings by the most posterior retinitis location (Table 4). Of particular interest were 2 eyes from 1 patient in which the in-clinic ophthalmologist documented retinitis in zone 1, a location that should have been detected on photography. Both of these eyes had lesions that occupied less than 10% of the total retina surface area and 1+ border opacity (faint border whitening not obscuring the details of the underlying choroid or consisting of satellites only). We retrospectively reviewed the medical records of this patient; at subsequent clinic visits, the in-clinic ophthalmologist determined that neither eye had CMV retinitis. The false-negative montage images also underwent evaluation by 3 unmasked uveitis experts (D.H., G.N.H., and T.P.M.). Of the 15 images reviewed, they found potential abnormalities in 3 eyes, each of which had been classified as having CMV retinitis in or anterior to zone 2 by the in-clinic ophthalmologist. One of the eyes had a possible small focus of retinitis at the image border, and 2 eyes had lesions thought unlikely to be CMV retinitis but requiring close follow-up.
Two eyes without CMV retinitis on indirect ophthalmoscopy were classified as having CMV retinitis by remote graders (ie, false-positive findings). One image was rated as having poor gradability and diagnosed as unknown status by all of the graders; the other was found by the 3 unmasked uveitis experts to have a nonspecific inactive chorioretinal scar.
In this study, we report the results of a CMV retinitis telemedicine screening program that was instituted at the point of HIV care. In contrast to previous studies from ophthalmology referral centers in Thailand that have found CMV retinitis in as many as one-third of patients with HIV, we diagnosed CMV retinitis in 15.5% of those patients undergoing screening.20 Although this finding should be interpreted with caution because only 39.9% of patients with a CD4 cell count of less than 100/μL underwent screening, these results likely provide a more accurate estimate of the burden of CMV retinitis in the Thai population with HIV. The retinitis detected in this point-of-care screening program was less extensive than previous reports from nearby Thai ophthalmology clinics, suggesting that our program detected CMV retinitis earlier.12 Remote CMV retinitis screening by telemedicine did not perform as well in the HIV clinic setting as it had in a previous study from a tertiary ophthalmology referral clinic,15,17 but the program may be useful for identifying patients with immediately vision-threatening disease.
Fundus photography, as implemented in a single HIV clinic in Thailand, detected approximately one-third of CMV retinitis cases. This performance is inferior to that of telemedicine screening for diabetic retinopathy, for which the sensitivity typically exceeds 85%.21,22 Most of the missed lesions were small and located in the peripheral retina outside the photographable range of the camera, whose internal fixation targets do not image zone 3 or the anterior portion of zone 2. Two small lesions missed by the remote graders but classified as being in zone 1 by the in-clinic ophthalmologist were later determined not to be CMV retinitis by the treating physician, suggesting the reference diagnosis was misclassified in these cases. Thus, fundus photography with the study camera detects most vision-threatening lesions in the posterior pole but misses disease confined to the peripheral retina. Obtaining a wider view of the retina would likely increase the sensitivity of telemedicine CMV retinitis screening. This view could be accomplished with a camera that has a wider-angle lens or by using an external fixation light instead of the internal fixation light used in the present study. Neither of these limitations is easily overcome. Fundus cameras with wider fields of view are cost-prohibitive for resource-constrained settings, and obtaining peripheral retinal images with an external fixation target is technically challenging and may not be feasible in the hands of personnel with minimal training. A potential alternative would be to train HIV clinicians to use indirect ophthalmoscopy to diagnose CMV retinitis.23
Even an imperfect screening test could increase the number of patients with CMV retinitis detected by the health care system. Fundus photography detected cases that were more likely to have larger lesions, zone 1 involvement, and poorer visual acuity. Telemedicine may therefore identify patients in urgent need of referral to an ophthalmologist due to immediately sight-threatening disease.6,24 However, untreated peripheral lesions may grow into the posterior pole and may increase the risk for retinal detachment.25- 28 Patients would therefore need to undergo rescreening periodically for previously missed lesions that progress into the viewable range of the camera. This process may be feasible at the early stages of highly active antiretroviral therapy, when most cases of CMV retinitis in Thailand are diagnosed, because patients see their physician every few weeks for routine HIV care.12 Nevertheless, a more sensitive screening test that reliably identifies early retinitis clearly would be preferable.
The CMV retinitis diagnosed at this HIV clinic was less extensive compared with disease diagnosed in previous studies performed at ophthalmology clinics, likely because the diagnosis occurred earlier. Patients in Thailand with CMV retinitis must be referred to an ophthalmologist sometimes several hundred kilometers away for diagnosis and treatment.9 This situation likely leads to disease progression before diagnosis, permanent vision loss, and decreased quality of life.29 This study provides evidence that integrating screening for CMV retinitis into routine clinical care for HIV allows for earlier detection. Compared with the study conducted at the nearby ophthalmology referral center, we found less bilateral retinitis (31% vs 46.2%), fewer visual symptoms (33% vs 90.8%), better median visual acuity (20/50 vs 20/80), smaller lesion sizes (29% of lesions occupied more than one-fourth of the total retinal surface area in the present study vs 56.7%), and less zone 1 involvement (29% vs 61.8%).12 Notably, the disease profile of the patients in the HIV clinic in this study was closer to that observed in Western populations.19,30- 32
Our goal was to screen all patients with HIV and a CD4 cell count of less than 100/μL using an established protocol to recruit participants; we enrolled only 39.9% of eligible patients, resulting in a relatively small sample size. Although nonparticipation may limit the generalizability of the study if CMV retinitis affected participants and nonparticipants differently (eg, in terms of location or severity), we found no evidence suggesting that this was the case because CD4 cell counts were similar in both groups. We did not record why prospective study participants did not enroll in the study. However, we retrospectively solicited the impressions of clinic personnel, who suggested the following 3 major reasons for nonparticipation: (1) some patients, especially those unaccompanied by a caretaker, did not want to undergo pupil dilation owing to concerns about driving home after dilation; (2) some patients, especially those without visual symptoms, did not want to spend the extra time needed to participate; and (3) on busy days, clinic staff often did not have time to obtain informed consent. These 3 factors must be addressed before screening is adopted as a clinic policy.
Our research had several other limitations. We did not follow up study participants after diagnosis and cannot comment on the long-term visual outcomes or complication rates of our patients. Based on the results of a prior study,15 we trained the photographer to use the internal fixation lights to create mosaic images of the posterior retina. This strategy limited the ability of the camera to detect far peripheral disease, which likely reduced the sensitivity of the test. Finally, patients in the study were recruited from an urban medical center, where residents likely have better access to medical care than those living in rural areas.33,34 Retinal photography may have a higher sensitivity in detecting CMV retinitis if instituted at a rural clinic, where patients are more likely to present at a later stage of their disease.
Instituting photographic CMV retinitis screening with the systems and protocols described in this article is feasible at the primary care level and could lead to earlier diagnosis. Retinal imaging with the camera used in this study did not have a high sensitivity for detecting CMV retinitis because cases with less extensive and more peripheral disease were missed. However, our data show that telemedicine can identify lesions that are an immediate threat to vision and require urgent attention by an ophthalmologist. A camera that can image the entire retina more easily would increase the accuracy of telemedicine in detecting CMV retinitis among at-risk patients and would be even more valuable in an HIV clinic.
Submitted for Publication: August 5, 2014; final revision received September 23, 2014; accepted September 27, 2014.
Corresponding Author: Jeremy D. Keenan, MD, MPH, Medical Sciences Bldg S334, Campus Box 0412, 513 Parnassus Ave, San Francisco, CA 94143 (firstname.lastname@example.org).
Published Online: November 20, 2014. doi:10.1001/jamaophthalmol.2014.4766.
Author Contributions: Dr Keenan 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. Dr Jirawison and Mr Yen contributed equally to this work.
Study concept and design: Jirawison, Leenasirimakul, Chen, Patikulsila, Ausayakhun, Heiden, Margolis, Keenan.
Acquisition, analysis, or interpretation of data: Jirawison, Yen, Leenasirimakul, Chen, Guadanant, Kunavisarut, Watanachai, Holland.
Drafting of the manuscript: Yen, Guadanant, Kunavisarut, Watanachai, Keenan.
Critical revision of the manuscript for important intellectual content: Jirawison, Yen, Leenasirimakul, Chen, Patikulsila, Ausayakhun, Heiden, Holland, Margolis.
Statistical analysis: Yen, Keenan.
Obtained funding: Margolis.
Administrative, technical, or material support: Jirawison, Leenasirimakul, Chen, Guadanant, Kunavisarut, Patikulsila, Watanachai, Ausayakhun, Heiden, Holland, Margolis.
Study supervision: Jirawison, Leenasirimakul, Margolis.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Dr Holland has served on advisory boards for the following companies: Genentech, Inc, Novartis International AG, Santen, Inc, and Xoma (US) LLC. Dr Margolis has pending intellectual property with the University of California, San Francisco, describing a mobile phone camera for retinal imaging, which at the current time has no financial value. No other disclosures were reported.
Funding/Support: This study was supported by the Gladstone Institute of Virology and Immunology Center for AIDS Research and the Research Evaluation and Allocation Committee, University of California, San Francisco; by grant K23EY019071 from the National Eye Institute; and by That Man May See, the Littlefield Trust, the Peierls Foundation, and the Doris Duke Charitable Foundation through a grant supporting the Doris Duke International Clinical Research Program at the University of California, San Francisco. Mr Yen is a Doris Duke International Clinical Research Fellow.
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.