Retinal Detachment Risk in Cytomegalovirus Retinitis Related to the Acquired Immunodeficiency Syndrome FREE

CYTOMEGALOVIRUS (CMV) disease is one of the most common opportunistic complications of the acquired immunodeficiency syndrome (AIDS).¹ In 71% to 85% of cases, it presents as retinitis.^{2 - 3} Prior to the availability of highly active antiretroviral therapy (HAART), estimates of CMV retinitis incidence rates among patients with AIDS ranged from 24.8% to 44.9%.^{1 ,3 - 5} Vision loss is common in patients with AIDS and CMV retinitis.⁶ Most vision loss results from either progressive retinal necrosis caused by CMV replication or from retinal detachment. Retinal detachment is a common complication of CMV retinitis, with a reported incidence rate of approximately 50% per patient per year or 33% per eye per year.^{6 - 8}

The ganciclovir implant is a surgically implanted reservoir of ganciclovir that provides higher intraocular levels of drug than systemically administered ganciclovir⁹ and longer-lasting control of retinitis.^{10 - 11} One question has been whether the use of ganciclovir implant therapy alters the rate of retinal detachment in eyes with CMV retinitis, possibly increasing risk because of surgical disturbance of the posterior segment or perhaps decreasing risk because of improved control of retinitis.

The introduction of HAART has reduced the incidence of CMV retinitis.¹² In some patients with CMV retinitis, HAART restores sufficient immunity to suppress retinitis without specific anti-CMV therapy. After confirmation that an immunologic response to HAART has occurred, these patients may safely discontinue specific anti-CMV therapy.^{13 - 16}

In this study, we evaluated the effects of ganciclovir implant therapy and of HAART on the incidence of retinal detachment in a large, prospectively followed cohort of patients with AIDS and CMV retinitis. Although initially we were interested in the effect of ganciclovir implant therapy, the near simultaneous introduction of HAART seemed likely to confound the analysis, leading us to evaluate the effects of both on retinal detachment risk.

All patients with CMV retinitis and AIDS seen by the Division of Ocular Immunology at the Wilmer Ophthalmological Institute, Baltimore, Md, were studied, from the first case on August 8, 1983, through September 1, 1998. Diagnoses of CMV retinitis were based on the characteristic clinical picture of necrotizing retinitis seen by indirect ophthalmoscopy through a dilated pupil.⁵ Diagnoses of AIDS were made according to standard Centers for Disease Control and Prevention criteria in effect at the time of each diagnosis.^{17 - 21}

Each patient's CMV retinitis management strategy was determined by the best medical judgment of the treating physicians from among the therapeutic options available at the time of illness. Ganciclovir implants became available to our population in November 1995. Prior to that date, only systemic therapies(intravenous ganciclovir sodium,^{22 - 23} foscarnet sodium,^{23 - 24} and/or cidofovir^{25 - 26} ) were used for primary management, except for 3 eyes (of 3 patients), which received implants under study protocols in 1993.

When available at the time of the patient's illness, antiretroviral therapy and prophylaxis against specific opportunistic infections were used according to the best medical judgment of treating physicians. HAART, defined as combination antiretroviral therapy including a protease inhibitor and/or nonnucleoside reverse transcriptase inhibitor, has been widely used in our population since early 1996. Prophylaxis for Pneumocystis carinii became widely used in our population in 1990, whereas prophylaxis for Mycobacterium avium complex infections became widely used in 1996.

A computerized database was kept prospectively for all patients with CMV retinitis seen at our center. Demographic data collected included age at the time of CMV retinitis diagnosis, race, gender, and risk factor(s) for human immunodeficiency virus infection. Data regarding clinical status at the time of CMV retinitis diagnosis included the date of diagnosis (for each affected eye), current absolute CD4⁺ T-lymphocyte count (when available), location of the CMV lesion(s), and lesion size. Location was characterized according to whether each of the following zones was involved: zone 1, which includes the area within 3000 µm of the foveal center plus the area within 1500 µm of the optic disc; zone 2, which extends from zone 1 to the equator; and zone 3, anterior to the equator.^{5 ,27} Lesion size was classified as involving "24% or less" or "25% or more" of the retina.⁵

Follow-up data regarding treatment and outcomes were collected over time, including the date at which each specific anti-CMV therapy was initiated, dates of all surgerical procedures, and date of most recent visit. Date(s) of retinal detachment diagnosis by ophthalmoscopy or ultrasonography was also recorded.

Ganciclovir implant placement was performed using standard surgical techniques.⁹ The anti-CMV treatment status of each eye was classified into 1 of 3 categories: (1) systemic anti-CMV therapy only, with no ganciclovir implant surgery ever performed; (2) primary implant, placed within 30 days after diagnosis of CMV retinitis in the ipsilateral eye; or (3) secondary implant, placed more than 30 days after diagnosis of CMV retinitis in the ipsilateral eye. In the primary implant group, none received treatment with intravenous therapy prior to implantation. In the secondary implant group, all but 2 eyes of 2 patients received systemic treatment prior to ganciclovir implant therapy. Two other eyes developed retinal detachment during treatment with systemic therapy and subsequently received a ganciclovir implant more than 30 days after CMV retinitis diagnosis. Because these eyes developed retinal detachment prior to ganciclovir implant exposure, they were included with the systemic therapy group, censoring follow-up at the time of retinal detachment occurrence.

Patients were considered as treated with HAART if at any visit they were taking combination antiretroviral therapy including a protease inhibitor and/or a nonnucleoside reverse transcriptase inhibitor. Those receiving HAART were further classified as "HAART responders" if an immunologic response was observed during follow-up; otherwise, they were classified as "HAART nonresponders." For the purposes of this analysis, a HAART response was defined as an increase in CD4⁺ T-lymphocyte count by 50 or more cells/µL above the level at the time of CMV retinitis diagnosis to an absolute level of at least 100 cells/µL.

The data set was closed September 1, 1998. Logical inconsistencies and outliers were investigated and adjudicated based on chart review. Missing data were investigated by chart review and entered into the data set unless unavailable.

Demographic and clinical characteristics at the time of CMV retinitis diagnosis were compared among subgroups by calculating odds ratios (ORs) and applying the χ² test or Fisher exact test when expected counts in a category were less than 5. Direct comparisons of time to retinal detachment by demographic and clinical characteristics were performed by constructing Kaplan-Meier curves and applying the log-rank test.²⁸ For each characteristic, risk ratios with 95% confidence intervals were generated using the Cox proportional hazards model,²⁸ using indicator variables when necessary to make pairwise comparisons. Cox models were also used for multiple regression analyses of time to retinal detachment. Final multiple regression models were generated by sequentially removing demographic and clinical characteristic variables found not to be significantly associated with retinal detachment by likelihood ratio testing. Robust variance estimation and methods to adjust for correlation between eyes of the same patient were used for all Cox models.²⁹ STATA 5.0 software³⁰ was used to perform statistical analyses.

Eyes receiving ganciclovir implants more than 30 days after diagnosis of CMV retinitis were not included in the secondary implant group if retinal detachment occurred prior to implant placement. Therefore, the risk of retinal detachment for this group was calculated from the time of implant placement. For all other factors and for the systemic therapy and primary implant groups, the reported retinal detachment risk was calculated from the time of CMV retinitis diagnosis. As a result, 2 different follow-up time denominators were used for risk calculations, making it necessary to create 2 Cox models for calculation of adjusted risk ratios. The "from implant" model excludes the time between CMV retinitis diagnosis and ganciclovir implant placement in the secondary implant group, providing the best means of assessing the adjusted risk ratio for the secondary implant group with respect to the systemic therapy group. The "from diagnosis" model includes all follow-up time in all groups for optimal calculation of all other adjusted risk ratios.

To address the question of whether retinal detachment risk was increased in ganciclovir implant groups during early follow-up (corresponding to the early postoperative period for the implant groups), the "from implant" analysis was repeated, censoring follow-up time at 60 days. The "from implant" analysis was also repeated, pooling implant groups, to give a comparison of overall ganciclovir implant treatment with respect to systemic therapy only.

To assess whether inclusion of eyes managed prior to the availability of ganciclovir implants and HAART altered results, risk was compared in eyes that received systemic anti-CMV therapy diagnosed with retinitis before vs after the availability of the ganciclovir implant, and in patients who did not receive HAART diagnosed before vs after the availability of HAART. Additional sensitivity analyses were conducted to determine whether classifying ganciclovir implant status and follow-up time differently would alter results: (1) the primary implant group was compared with the other groups combined (censoring follow-up of the secondary implant group at the time of initial implant surgery); and (2) all implant eyes were compared with all nonimplant eyes, with follow-up time beginning at CMV retinitis diagnosis. To confirm that analysis by eye(adjusting for increased correlation between eyes of the same patient) rather than by patient did not alter results, the 2 sensitivity analyses as well as the "from diagnosis" and the "from implant" analyses were repeated using patients rather than eyes as the unit of observation.

A total of 591 consecutive patients with AIDS who had CMV retinitis affecting 911 eyes were evaluated at our center as of September 1, 1998. Of these, 48 patients (69 eyes) never returned for follow-up. Nine other patients had follow-up visits after diagnosis of CMV retinitis in their first eye, but no follow-up after subsequent diagnosis in the second eye (9 eyes). Twenty-four patients (33 eyes) did not receive specific anti-CMV therapy, either because no such therapy existed at the time of their illness or by choice. For 5 patients, the date of subsequent CMV retinitis diagnosis in the second eye was missing(5 eyes). Eight patients had retinal detachment already present at the time of CMV retinitis diagnosis in their only affected eye (8 eyes), while 14 patients already had retinal detachment in 1 of 2 affected eyes by the time of CMV retinitis diagnosis (14 eyes) but remained at risk for detachment in the second eye. A final cohort of 773 eyes of 511 patients for analyses of time to retinal detachment with the eye as the unit of observation (90.3% of treated eyes at risk of retinal detachment) remained. For analyses of time to retinal detachment with the patient as the unit of observation, 497 patients were at risk of a first retinal detachment (not including the 14 with unilateral detachment already present at the time of diagnosis, who were no longer at risk).

Comparing the demographic and clinical characteristics listed in Table 1 of the final cohort vs the groups of patients with no follow-up, with no treatment, and with preexisting retinal detachment, the only significant difference was seen in patients without follow-up who were more likely to have an initial CD4⁺ T-lymphocyte count less than 12 cells/µL (OR, 4.5; P = .01). Comparing the eye-specific clinical characteristics listed in Table 1, the only significant difference between any of these groups and the final cohort was that eyes with preexisting retinal detachment were more likely to have lesions involving 25% or more of the retina (OR, 6.0; P<.001).

Patient-specific demographic and clinical characteristics did not significantly differ between the systemic therapy, primary implant, and secondary implant groups, except that eyes of patients who received HAART were much more likely to receive primary (OR, 30.3; P<.001) and secondary(OR, 30.3; P<.001) implants than systemic anti-CMV therapy (Table 1). Comparing eye-specific clinical characteristics, the lesion size at the time of retinitis diagnosis was similar in all anti-CMV treatment groups (data not shown). However, by the time of initial implant surgery, secondary implant eyes had become significantly more likely to have larger lesions than the other groups (P = .008). The lesion location distribution did not differ between anti-CMV treatment groups (P = .35). Except for higher usage of implants in eyes of patients who received HAART (see above), neither patient- nor eye-specific characteristics differed significantly between HAART groups (data not shown).

Fifty eyes (of 35 patients) in the primary implant group underwent 76 implant surgerical procedures. Of these, 31 received 1 implant, 13 received 2, 5 received 3, and 1 eye received 4 during follow-up. Forty-nine secondary implant eyes (of 39 patients) underwent 71 implant procedures, with 32, 12, and 5 eyes having 1, 2, and 3 implants, respectively. The median time from CMV retinitis diagnosis to initial implant surgery was 10 days for the primary implant group (range, 1-27 days) and 218 days (range, 33-1244 days) for the secondary implant group. Median follow-up time was similar for the systemic therapy (0.42 year) and primary implant (0.52 year) groups. For the secondary implant group, distribution of follow-up time was also similar when started from the time of initial implant surgery (median, 0.49 year), but substantially longer when started from the date of CMV retinitis diagnosis (median, 1.26 year).

Seventy-seven patients with a total of 116 eyes affected by CMV retinitis were treated with HAART. Thirty-one of these patients (45 eyes) were observed to have an immunologic response to HAART (as defined previously). The median follow-up time was longer for patients receiving HAART (0.81 year) than patients who did not receive it (0.41 year), primarily because of increased survival in the HAART responder subgroup.

Among the 773 eyes followed for the incidence of retinal detachment, 177 retinal detachments were observed (23%). The estimated median time to retinal detachment for all eyes was 1.96 years. The retinal detachment rates associated with patient- and eye-specific demographic and clinical characteristics are presented in Table 2 and Table 3, respectively. The final Cox multiple regression models for time to retinal detachment by eye are given in Table 4.

Among eyes in the primary implant group, the retinal detachment rate was two thirds that observed with systemic therapy (relative hazard [RH], 0.64; P = .25; Figure 1), but adjustment for other variables brought the risk ratio close to unity (RH, 1.12; P = .79). In the secondary implant group, with time-at-risk starting from date of first implant placement, the retinal detachment rate was three quarters of that observed in the systemic therapy group (RH, 0.76; P = .46), with the risk ratio also close to unity after adjusting for other variables (RH, 1.06; P = .88). The apparent difference between simple and adjusted risk ratios was primarily due to higher utilization of HAART in patients whose eyes received implants. The adjusted overall comparison for either kind of ganciclovir implant treatment vs systemic therapy showed no significant difference in risk (RH, 1.09; P = .77). However, CMV lesions involving zone 3 (RH, 1.80; P = .03 [adjusted RH, 1.77; P = .047]), and lesions involving greater than 25% of the retina (RH, 1.37; P = .048 [adjusted RH, 1.46; P = .03]) were associated with increased risk of retinal detachment.

Patients treated with HAART were substantially less likely to develop retinal detachment than patients who did not receive HAART (RH, 0.40; P = .003; Figure 2). Subset analysis demonstrated that patients observed to have an immunologic response to HAART had less than one third the retinal detachment rate of patients who did not receive HAART (RH, 0.30; P = .006; Figure 3). This effect remained statistically significant after adjustment for other variables (adjusted RH, 0.36; P = .02). The subset receiving HAART but not observed to have an immunologic response had a retinal detachment rate half that of the group that did not receive HAART (RH, 0.52; P = .048); however, this difference became less marked and statistically nonsignificant after adjusting for other variables (adjusted RH, 0.65; P = .29). Patients with bilateral disease were also more likely to have 1 or more retinal detachments than patients with unilateral disease (RH, 1.78; P<.003 [adjusted RH, 1.62; P = .01]). However, the individual eyes of these patients were not observed to have significantly higher risk of retinal detachment than affected eyes of patients with unilateral retinitis (adjusted RH, 1.28; P = .20).

Final Cox multiple regression models (Table 4) are reported both "from implant" (for comparing the secondary implant vs systemic therapy groups) and "from diagnosis" (for all other comparisons)(see definitions in the "Statistical Analysis" section). Although older age⁷ and CD4⁺ T-lymphocyte count⁷ previously have been reported as significantly associated with retinal detachment risk in CMV retinitis, significant associations were not observed in our population. A trend toward lower retinal detachment rates in nonwhite vs white subjects was observed in our population, but was not statistically significant. Inclusion or noninclusion of these 3 variables in the final models did not substantially alter other associations, so they were omitted.

Results of the sensitivity analyses performed showed no important differences from the results reported above. In particular, among eyes treated with systemic therapy only, there was no significant difference in risk between those diagnosed with retinitis before vs after ganciclovir implants became available (adjusted RH, 0.77; P = .53). Neither was there a significant difference among patients who did not receive HAART diagnosed before vs after HAART became available (adjusted RH, 1.36; P = .54). Inclusion in the multiple regression model of an indicator variable for whether CMV retinitis diagnosis was made prior to or after ganciclovir implants or HAART became available did not alter other associations. Comparison of retinal detachment risk in the pooled implant groups to the systemic therapy group, censoring follow-up after 60 days, showed no significant increase in retinal detachment risk during the early postoperative period (RH, 1.29; P = .39).

Nonrandomized studies of treatment outcomes, while having the advantage of being able to study a more "real world" population than will typically enroll in a randomized clinical trial, lack design features to control selection bias. Our primary results were robust to sensitivity analysis exploring this and other issues. We also adjusted for the potentially confounding effects of known risk factors for retinal detachment present at the time of CMV retinitis diagnosis to account for differences between groups in pretreatment retinal detachment risk. Nevertheless, interpretation of our results must bear in mind concern about potential selection bias. Although studies at tertiary care facilities such as ours can be subject to referral biases, we estimate that our center provides ophthalmological care for more than 80% of all patients with AIDS and CMV retinitis in our metropolitan area. The demographic makeup of our cohort includes relatively large numbers of African American and female patients, and of patients who had used intravenous drugs, reflecting the distribution of risk factors among patients with advanced AIDS in Baltimore. We were able to follow up a high percentage of treated patients with CMV retinitis at risk for retinal detachment. Follow-up over time was relatively successful, with the date of death known or a follow-up visit having occurred within 30 days of the closing date for 87% of patients in the cohort.

We found no substantial difference in retinal detachment risk for the eyes with CMV retinitis treated using ganciclovir implants (whether as initial or secondary treatment) compared with systemic therapy only. This result agrees with the retinal detachment results of previous large-scale clinical trials.^{10 - 11} Indeed, the unpublished Kaplan-Meier curves for observed time to retinal detachment in the Roche Ganciclovir Study Group³¹ and the Ganciclovir Implant Study Group³² trials bear a striking similarity to that from our study (Figure 1), including the slight dip observed in the early postoperative period for implant-treated eyes. This transient early increase did not approach statistical significance. These results suggest that placement of a ganciclovir implant into the posterior segment using standard surgical techniques does not substantially increase retinal detachment risk in eyes with CMV retinitis. Whether this conclusion can be generalized to eyes with other disease processes amenable to treatment using implantable reservoirs of medication remains to be seen.

We also found that use of HAART was associated with a 60% reduction of retinal detachment risk in eyes of patients with CMV retinitis. Use of HAART had the greatest effect on risk of retinal detachment of all variables evaluated, larger than that associated with well-established risk factors such as large lesion size^{6 ,8 ,25} and anterior lesion location.^{8 ,27} Because not all patients treated with HAART received HAART throughout follow-up, the extent of benefit may be underestimated. The greatest reduction in risk was found in patients with observed immune recovery, suggesting that the benefit may be conferred by improved immunity.

Several mechanisms may explain the reduced risk of retinal detachment in eyes of patients with AIDS and CMV retinitis treated with HAART. HAART has been reported to affect the clinical course of CMV retinitis, in that patients who develop immune recovery usually regain the ability to suppress retinitis progression without specific anti-CMV therapy.^{13 - 15} Therefore, it seems reasonable that the reduced risk of retinal detachment in eyes of our patients who received HAART was the result of improved immune control over CMV replication, which more successfully suppressed retinitis activity, preventing progression to larger lesion sizes. Both current activity of retinitis⁶ and larger lesion size^{6 ,8 ,25} are known risk factors for retinal detachment in patients with AIDS and CMV retinitis. Another protective factor may have been increased inflammatory activity within the retina of patients treated with HAART. It has been suggested that such inflammation may cause a strong adhesion between the retinal pigment epithelium and the neurosensory retina,^{33 - 34} which may be protective vs retinal detachment. Finally, it has also been suggested that the attached posterior hyaloid may apply traction on susceptible necrotic retina in this setting.^{33 - 34} Indeed, Martin et al,³⁵ observed that every retinal detachment in their implant trial occured concurrently with a posterior vitreous separation. Perhaps the different pattern of inflammation related to HAART-mediated immune improvement alters the course of vitreous detachment, changing retinal detachment risk. Our study is unable to discriminate which of these factors may have been the primary protective mechanism(s) in our patients with CMV retinitis who were treated with HAART.

As in prior studies, larger lesion size^{6 ,8 ,25} and involvement of the anterior retina near the vitreous base^{8 ,36} were associated with increased risk of retinal detachment. Importantly, neither of these factors nor use of HAART significantly modified retinal detachment risk in eyes treated with implants vs systemic therapy only. Therefore, neither large lesion size nor anterior lesion location seem to be a contraindication to implant therapy. Higher risk of retinal detachment was observed in patients with bilateral retinitis, but not in individual eyes of these patients, presumably because patients with bilateral disease have 2 eyes at high risk rather than 1.

Our study found no statistical or substantial increase in retinal detachment risk when CMV retinitis was treated with ganciclovir implants compared with systemic therapy only. However, the use of HAART in patients with AIDS and CMV retinitis was associated with a substantially lower rate of retinal detachment, presumably due to improvement of immune competence against CMV.

Accepted for publication June 16, 2000.

This study was supported in part by grants EY00386 and EY07127 (Dr Kempen) and EY10268 (Dr Jabs) from the National Eye Institute, National Institutes of Health, Bethesda, Md.

We thank Alfred Sommer, MD, MHS, for his helpful advice regarding this manuscript, and Judith Southall for valuable secretarial assistance in its preparation.

Corresponding author and reprints: John H. Kempen, MD, MPH, MHS, 550 N Broadway, Suite 700, Baltimore, MD 21205 (e-mail: jkempen@jhmi.edu).