Accuracy and Reliability of Telemedicine for Diagnosis of Cytomegalovirus Retinitis


To determine the feasibility of remote diagnostic screening for cytomegalovirus (CMV) retinitis among HIV patients in northern Thailand.


Prospective, observational cross-sectional study.


One hundred eighty-two eyes from 94 consecutive patients with HIV seen in 2008 and 2009 at a tertiary university-based medical center were photographed using a digital retinal camera. Individual and composite images were uploaded to a secure web site. Three expert graders accessed the electronic images and graded each image for signs of CMV retinitis. Results of remote expert grading were compared with on-site patient examination by local expert ophthalmologists.


On-site ophthalmologists diagnosed CMV retinitis in 89 (48.9%) of 182 eyes. Trained ophthalmic photographers obtained digital retinal images for all 182 eyes. As compared with the on-site examinations, the sensitivity for detecting CMV retinitis by remote readers using composite retinal images ranged from 89% to 91%. The specificity for detecting CMV retinitis by remote readers ranged from 85% to 88%. Intrarater reliability was high, with each grader achieving a κ value of 0.93. Interrater reliability among the 3 graders also was high, with a κ value of 0.86.


Remote diagnostic screening for CMV retinitis among HIV-positive patients may prove to be a valuable tool in countries where the burden of HIV exceeds the capacity of the local eye care providers to screen for ocular opportunistic infections.

The World Health Organization estimates that there are currently 33.4 million people living with HIV worldwide, with the most severely affected regions being sub-Saharan Africa, Southeast Asia, and India. In many of these countries, intensive national and international efforts have led to the development of programs for HIV diagnosis and treatment. In contrast, most of these programs have no systems in place for screening patients for CMV disease. As a consequence, CMV retinitis is emerging as a leading cause of blindness in regions of the world most severely affected by HIV and AIDS. Nevertheless, it is difficult to understand fully the magnitude of this problem because limited screening for CMV retinitis is taking place in these regions. In Southeast Asia, CMV retinitis may be present in as many as one third of patients with AIDS. At Chiang Mai University in northern Thailand, CMV retinitis is the second most common cause of blindness, just behind cataract.

Using indirect ophthalmoscopy, an experienced ophthalmologist can detect CMV retinitis in a highly sensitive and specific manner. Unfortunately, the countries with the highest prevalence of HIV infection often have the fewest ophthalmologists and therefore are unable to meet the existing and growing need for screening examinations. Because the number of at-risk patients is large and increasing in regions affected by the current HIV and AIDS pandemic, novel ways of screening for CMV retinitis are needed.

Herein, we describe the use of a commercially available, noncontact retinal camera to obtain retinal images for remote diagnosis of CMV retinitis by expert graders with extensive experience in diagnosing and treating CMV retinitis. We propose that telemedicine can be an effective strategy to screen for CMV retinitis in underserved areas. We chose to study the application of telemedicine for CMV retinitis in northern Thailand because of the excellent existing ophthalmic care and internet capabilities coupled with the high prevalence of disease and need for increased diagnostic capacity.


Study Design

We conducted a prospective study at the Ocular Infectious Diseases Clinic, Maharaj Nakorn Chiang Mai Hospital, a university-based tertiary medical center. From August 7, 2008, through April 9, 2009, subjects were interviewed and examined according to a prespecified protocol with a standardized data form. We enrolled all HIV-positive patients newly referred to the clinic in whom a clinical diagnosis of CMV retinitis was made, as well as those HIV-positive patients diagnosed with CMV retinitis at the clinic in the preceding month. In addition, from August 7, 2008, until January 7, 2009, we also enrolled all newly referred HIV-positive patients without clinical evidence of CMV retinitis.

Study visits took place on only 1 occasion for each enrolled subject. At this time, an attending ophthalmologist examined each enrolled patient with an indirect ophthalmoscope through a dilated pupil. A detailed retinal drawing was performed for each examined eye, with location of disease categorized by zone, as described previously. A trained ophthalmic photographer used a Topcon TRC-NW 6S digital fundus camera (Topcon, Tokyo, Japan) to capture 10 images per eye (1 image of the external segment and 9 overlapping 45-degree fundus images). Automated composite images covering an 85-degree retinal field were created for each photographed eye using software included with the camera, and both individual and composite images then were transferred to a secure online server via a file transfer protocol. Three expert graders (D.H., G.N.H., T.P.M.) evaluated each eye in 3 separate rounds of grading: first, the composite images; second, the single central image, including the macula and optic nerve; and third, the set of all individual images. Each expert grader was a uveitis specialist with extensive expertise in diagnosing and treating CMV retinitis. Diagnosis of the presence or absence of CMV retinitis was made according to the clinical experience of each expert grader; zones were assigned using the same guidelines as used by the indirect ophthalmoscopist. Eyes were analyzed independently, in random order. No clinical information was provided to the remote expert graders. In each round of grading, images were graded for gradability, and the presence or absence of CMV retinitis, vitreous haze, and cotton wool spots. Gradability was defined as good (focused images including the entire 85-degree retinal field), acceptable (images did not include the entire 85-degree retinal field, but the presence or absence of CMV retinitis still could be determined, or images moderately out of focus), or poor (images included such a small degree of retinal field that no determination of the presence or absence of CMV retinitis could be made, or images were extremely out of focus). Remote expert grading then was compared with on-site examinations in Thailand, which served as the gold standard. After sensitivity and specificity were calculated, eyes that were classified incorrectly (false positive or false negative) by all 3 remote expert graders were identified for further analysis. Two authors (A.H.S., J.D.K.) analyzed the corresponding fundus images and retinal drawings of these eyes to determine reasons for misclassification.

Statistical Analysis

Descriptive data were analyzed as medians with interquartile range or as proportions. The sensitivity and specificity of remote diagnosis by expert graders was calculated separately for each grader using the on-site examination as the gold standard. Eyes were excluded if the on-site ophthalmologist could not determine with certainty whether CMV retinitis was, or was not, present. The goal of remote screening by remote expert graders was to identify all possible cases of CMV retinitis for referral. Therefore, for the purposes of our analysis, any eyes graded by the remote expert graders as either “CMV retinitis present” or “cannot determine” were classified as a CMV retinitis diagnosis. Negative and positive predictive values were calculated using the prevalence of CMV retinitis observed at the Ocular Infectious Diseases Clinic during the course of the study, including those not enrolled in the study. Exact binomial confidence intervals were calculated. Intrarater reliability was estimated for each grader with a κ statistic by comparing grades of 50 duplicate composite images. Interrater reliability was estimated for all 3 graders together, using a κ statistic. All analyses were conducted with Stata software version 10 (Stata Corp, College Station, Texas, USA).


Ninety-four patients were enrolled during the 8-month study period. Characteristics of these patients are summarized in Table 1 . Ophthalmologists at Chiang Mai University diagnosed 59 (62.8%) of the 94 patients with CMV retinitis in at least 1 eye. They were able to determine definitively whether CMV retinitis was present or absent in 182 (96.8%) of the 188 eyes. Of these, 89 eyes (48.9%) had CMV retinitis. Indirect ophthalmoscopy could not determine definitively the presence or absence of CMV retinitis in 6 eyes because of seclusio pupillae (n = 3, 50%), dense vitreous haze or hemorrhage (n = 2, 33%), or cataract (n = 1, 17%); these 6 eyes were excluded from further analysis. At least 1 image was captured for each of the 182 eyes in the study. All captured images from these 182 eyes were uploaded successfully to an online server and were accessed by the 3 remote expert graders ( Figure ).


Characteristics of 94 HIV-Infected Patients from an Ocular Infectious Diseases Clinic in Thailand Who Participated in a Study to Determine the Reliability of Telemedicine Diagnosis of Cytomegalovirus Retinitis

Characteristic Median (IQR) or Proportion
Median age (IQR), years 38 (33 to 45)
Female, % 53.2%
Self-reported CD4 count a
Median (IQR) 52 (17.5 to 173.5)
CD4 < 50, % 48.4%
CD4 <100, % 62.5%
On antiretroviral therapy, % 76.6%
≥ 1 eye with CMV retinitis, % 62.8%
≥ 1 eye with visual symptoms, % 77.7%

CMV = cytomegalovirus; IQR = interquartile range.

a Only 64 patients had available CD4 data.


Representative fundus images of cytomegalovirus retinitis from a tertiary medical center in northern Thailand, captured with the Topcon TRC-NW 6S digital fundus camera. Images captured disease in zone 1 (Top left, Bottom left, Bottom right) and zone 2 (Top left, Top right, Bottom left, Bottom right).

Remote Expert Grading

With regard to gradability, remote expert graders determined that 81.9% to 92.9% of the composite images were satisfactory or good-quality images. Detection of CMV retinitis by remote expert grading had a sensitivity of 88.8% to 91.0% and specificity of 84.9% to 88.2% when the composite image was compared against the gold standard of indirect ophthalmoscopy. The sensitivity and specificity were similar when graders scored only a single central image ( Table 2 ). When graders scored a set of individual 45-degree images for each eye instead of a single composite image, sensitivity decreased and specificity increased ( Table 2 ). Intrarater reliability was high (κ, 0.93 for each grader). Interrater reliability among the 3 graders also was high (κ, 0.86). The prevalence of newly diagnosed CMV retinitis in eyes examined at the clinic during the study period was estimated to be 35.6% (89/250). Assuming this prevalence, the positive predictive value ranged from 76.9% to 80.8% and the negative predictive value ranged from 93.4% to 94.5%.


Sensitivity and Specificity of Remote Cytomegalovirus Retinitis Diagnosis, as Compared with a Gold Standard of Indirect Ophthalmoscopy, in a Study of 182 Eyes Examined at an Ocular Infectious Diseases Clinic in Thailand

Grader 1 Grader 2 Grader 3
Gradability 87.9% (82.2 to 92.3) 92.9% (88.1 to 96.1) 81.9% (75.5 to 87.2)
Composite photograph 88.8% (80.3 to 94.5) 89.9% (81.7 to 95.3) 91.0% (83.1 to 96.0)
Central photograph 85.4% (76.3 to 92.0) 91.0% (83.1 to 96.0) 91.0% (83.1 to 96.0)
Individual photographs 79.8% (69.9 to 87.6) 82.0% (72.5 to 89.4) 84.3% (75.0 to 91.1)
Composite photograph 88.2% (79.8 to 93.9) 88.2% (79.8 to 93.9) 84.9% (76.0 to 91.5)
Central photograph 91.4% (83.8 to 96.2) 90.3% (82.4 to 95.5) 86.0% (77.3 to 92.3)
Individual photographs 91.4% (83.8 to 96.2) 91.4% (83.8 to 96.2) 88.2% (79.8 to 93.9)
Intrarater reliability a 0.93 (0.80 to 1.00) 0.93 (0.80 to 1.00) 0.93 (0.79 to 1.00)
Positive predictive value 80.6% (71.4 to 87.9) 80.8% (71.6 to 88.0) 76.9% (67.3 to 84.5)
Negative predictive value 93.4% (88.2 to 96.8) 94.0% (89.0 to 97.2) 94.5% (87.8 to 96.7)

95% confidence intervals shown for all estimates.

a κ statistic for diagnosis of cytomegalovirus retinitis comparing 50 duplicate composite images.


There were 16 eyes for which at least 1 remote expert grader diagnosed CMV retinitis, but the on-site examiners did not diagnose CMV retinitis by indirect ophthalmoscopy (false positives). In 10 of these 16 eyes, CMV retinitis was diagnosed by all 3 remote expert graders. The images from 7 of these 10 eyes were rated as poor quality by 2 or more of the remote expert graders. In the remaining 3 eyes, a post hoc analysis revealed either retinal scarring consistent with prior CMV retinitis (2 eyes) or a small patch of CMV retinitis in an eye with multiple cotton wool spots (1 eye). There were 12 eyes for which at least 1 grader failed to diagnose CMV retinitis, but a diagnosis was made by indirect ophthalmoscopy (false negatives). In 7 of these eyes, none of the remote expert graders made the diagnosis of CMV retinitis. Four of these 7 eyes had CMV retinitis in zone 3 only, and the remaining 3 eyes were scored by the remote expert graders as having cotton wool spots.

Zone 3 Retinitis

CMV retinitis confined to zone 3 would not be detected by the camera used in this study. Indirect ophthalmoscopy classified 14 of 182 eyes as having CMV retinitis localized to zone 3 only. Nevertheless, remote expert graders failed to diagnose CMV retinitis in only 4 of these eyes. Review of the images of the remaining 10 eyes revealed that 2 had significant vitreous haze, 1 had a small focus of CMV retinitis in zone 1 that had been classified as a cotton wool spot by indirect ophthalmoscopy, 1 had an area of scarring in zone 2, and 6 were misclassified by the on-site ophthalmologist, as evidenced by retinal drawings that clearly show disease in zone 1 or 2.

Only gold members can continue reading. Log In or Register to continue

Jan 16, 2017 | Posted by in OPHTHALMOLOGY | Comments Off on Accuracy and Reliability of Telemedicine for Diagnosis of Cytomegalovirus Retinitis
Premium Wordpress Themes by UFO Themes