Risk of Vision Loss in Patients With Cytomegalovirus Retinitis and the Acquired Immunodeficiency Syndrome FREE

CYTOMEGALOVIRUS (CMV) disease is a major opportunistic complication of the acquired immunodeficiency syndrome (AIDS).¹ Retinitis accounts for 71% to 85% of CMV disease in these patients.^{1 - 3} Prior to the availability of highly active antiretroviral therapy (HAART), it was estimated that about 30% of patients with AIDS would develop CMV retinitis during their lifetime.^{2 ,4} However, the incidence of CMV retinitis has declined since the introduction of HAART.^{4 - 7}

The clinical course of CMV retinitis in patients with AIDS has also changed during the HAART era. Although the progression of CMV retinitis was previously universal in the absence of anti-CMV therapy, ^{8 - 9} with HAART many patients develop immune recovery sufficient to suppress CMV retinitis even after discontinuing anti-CMV therapy.^{10 - 13} Retinal detachment, which occurred in about one third of eyes with CMV retinitis per year in the pre-HAART era, ^{14 - 17} occurs 60% less frequently with HAART.¹⁸ However, patients who experience immune reconstitution with HAART are at risk for immune recovery uveitis (IRU), an inflammatory condition that can cause vision loss in eyes with CMV retinitis.^{19 - 21} Published estimates of the incidence rate for IRU have varied widely, from 10.9%²² to 83.0%²³ per person-year. Whether the overall effect of HAART on the visual acuity of patients with CMV retinitis and AIDS is beneficial or deleterious is unknown.

The ganciclovir implant, a surgically implanted reservoir of ganciclovir delivering high levels of the drug directly into the vitreous cavity for 6 to 8 months, was demonstrated to have greater efficacy for preventing CMV retinitis progression than intravenous ganciclovir during the pre-HAART era.^{24 - 25} However, in a trial conducted during the HAART era, times-to-retinitis progression was similar in groups randomized to parenteral cidofovir or implant plus oral ganciclovir therapy.²⁶ Visual acuity outcomes of treatment with the ganciclovir implant and with parenteral therapies have been similar^{24 ,26} in comparative trials, although these studies have had limited power to detect differences because of early crossover²⁴ or small sample size.²⁶ Trials comparing different systemic treatments for CMV retinitis^{27 - 29} have found no important differences in visual acuity outcomes.

To more fully characterize the effect of CMV retinitis and its treatment on the vision of patients with AIDS, we evaluated loss of visual acuity in a large cohort of patients with AIDS and CMV retinitis. We studied both the prevalence of visual impairment at the time of CMV retinitis diagnosis and the incidence of visual impairment during follow-up.

The cohort of patients with CMV retinitis at the Johns Hopkins University School of Medicine (Baltimore, Md) has been described previously.¹⁸ In this study, we report the visual acuity outcomes for the affected eyes of 648 consecutive patients with CMV retinitis evaluated at our center during the period from August 8, 1983 (first case), through March 31, 2000.

Cytomegalovirus retinitis was diagnosed based on its characteristic appearance visualized using indirect ophthalmoscopy.¹⁷ Use of anti-CMV and/or antiretroviral treatment was determined according to the best medical judgment of the treating physician from among the options available at the time of each patient's illness. In our population, ganciclovir implants became available in late 1995, and HAART became available around the same time.

At the time of CMV retinitis diagnosis, demographic data including age, race, gender, and risk factors for human immunodeficiency virus (HIV) infection were entered into a database. Data regarding clinical status at the time of CMV retinitis diagnosis included the date of diagnosis (for each affected eye), current absolute CD4+ T-cell count, location of the CMV lesion(s), and lesion size. Lesion size in each eye was characterized as involving either 24% or less or 25% or more of the retina.^{17 - 18} Lesion location in each eye was classified according to the system proposed by Holland et al³⁰ in which zone 1 is defined as the area within 1500 µm of the optic nerve or 3000 µm of the fovea, zone 2 extends from zone 1 to the vortex veins, and zone 3 is anterior to the vortex veins.

After CMV retinitis diagnosis, patients were asked to visit our clinic monthly. Between June 1999 and April 2000, the medical records of these patients were reviewed retrospectively, and the visual acuity scores and current antiviral therapy at each visit were abstracted. Retrospective data collection was facilitated by our clinic's use of a standardized flow sheet, which called for entry of visual acuity measurements and all medications at every visit, and by detailed clinical notes. Distance visual acuity was measured for purposes of clinical practice using either a Snellen or logarithmic³¹ chart. Pinhole distance visual acuity was also measured if the distance visual acuity was poor. Refraction was performed when indicated based on clinical judgment. For bedside examinations, a Jaeger chart was used, and the scores were converted into equivalent values. Because refraction was not performed in a standardized fashion at every visit, the best score among pinhole, distance, and Jaeger visual acuity was recorded if more than 1 was available.

Visual acuity scores were converted into log MAR scores according to the following formula:

log MAR = −log₁₀(visual acuity fraction).

Using this convention, each 0.1 decrement in the log MARvisual acuity score corresponds to the loss of 1 line on a logarithmic visual acuity chart.³¹ Standards for assigning log MARequivalent values do not exist for very low levels of visual acuity. To use as much of the available information as possible for patients with low levels of visual acuity, we chose to count visual acuity values worse than 2/200 in the affected eye, which were recorded as a fraction (eg, 1/200), as 1 line worse than 2/200. Notations of counting fingers, hand motions, light perception, and no light perception were counted as 2, 3, 4, and 5 lines worse than 2/200, respectively. Thus, each step along this ordinal scale was counted as a loss of 1 line of visual acuity.

The proportions of eyes of patients with a visual acuity measurement of 20/50 or worse and 20/200 or worse at the time of CMV retinitis diagnosis were evaluated for the prevalence study. For the incidence study, we evaluated time to loss of visual acuity events. For our primary incidence analysis, we defined events based on the number of lines of visual acuity (−0.1 steps on the log MAR scale, as defined previously) lost from the visual acuity score obtained at the time of CMV retinitis diagnosis: 3 lines, 6 lines, or 10 lines (corresponding to 2-, 4-, and 10-fold increases in the visual angle).³¹ Eyes were excluded from these analyses if the initial visual acuity measurement in the affected eye was 5/200 or worse (log MAR score≥1.6). Using this exclusion criterion, all included eyes could lose 6 lines and still have their visual acuity measured as a fraction. Therefore, only the loss of 10 lines of visual acuity event was affected by our method of interpolating visual acuity values not recorded as a fraction (eg, hand motions) on the visual acuity scale. Losses of visual acuity to the levels of 20/50 or worse or 20/200 or worse were also evaluated as alternative event definitions for the incidence study. Eyes with visual acuity measurements at or worse than each of these thresholds at the time of CMV retinitis diagnosis were excluded from the corresponding analysis. Because the outcomes based on changes in the visual angle are more analytically robust and the threshold-based outcomes are more intuitively understandable, we elected to report both.

Analyses of losses of visual acuity were complicated by the early transient postoperative vision loss that is expected following ganciclovir implant surgery. Such vision loss most often resolves within 28 days of ganciclovir implant surgery and nearly always within 42 days.³² Therefore, we elected to count only losses of visual acuity that persisted across consecutive visits spanning 28 days or more as events, to avoid misclassification of expected transient postoperative vision loss as a vision loss event. As a sensitivity analysis, we repeated our analyses counting only loss of visual acuity events persisting across a 42-day span.

We defined HAART as treatment at any time during observation with 3 or more antiretroviral agents, including at least 1 protease inhibitor or non-nucleoside reverse transcriptase inhibitor. The eyes of patients who did not receive HAART made up the no-HAART group, whereas the eyes of patients who did receive HAART made up the HAART group. Because immune recovery sufficient to control CMV retinitis is more accurately predicted by the CD4+ T-cell count than HIV viral load, ^{10 - 13} an immune recovery response to HAART was defined as an increase in CD4+ T-cell count to a level greater than 100 cells/µL and at least 50 cells/µL higher than the level at the time of CMV retinitis diagnosis.^{18 ,22} Sufficient CD4+ T-cell count information was obtained for the patients who received HAART to determine whether or not a HAART response had occurred during follow-up. The HAART group was further subdivided into HAART responder and HAART nonresponder groups according to whether or not an immune recovery response to HAART was observed during follow-up.

All eyes were counted as receiving 1 of 3 anti-CMV treatment strategies beginning on the date of CMV retinitis diagnosis. Eyes receiving a ganciclovir implant within 28 days of CMV retinitis diagnosis made up the primary-implant group, as in our prior study of retinal detachments.¹⁸ Secondary-implant therapy was defined as initial systemic therapy or deferral followed by ganciclovir implant treatment more than 28 days after the CMV retinitis diagnosis. Systemic therapy was defined as other treatment strategies (eg, systemic ganciclovir, foscarnet sodium, or cidofovir). Repeated intravitreal injections were not used as a primary therapy in our population. For sensitivity analysis, primary-implant therapy was compared with systemic therapy, counting secondary-implant eyes in the systemic therapy group and censoring follow-up time for secondary-implant eyes at the time of the initial implant placement.

For the prevalence study, we calculated proportions and compared odds ratios (ORs) using the χ² statistic or with the Fisher exact test when expected counts in a category were less than 5. Multiple logistic regression analysis was used to adjust the ORs for potential confounding. For the incidence study, Kaplan-Meier curves were constructed and used to calculate the percentage of eyes with loss of visual acuity events at 6 and 12 months.³³ Overall comparisons of time to loss of visual acuity events used the log-rank test. Cox regression analysis was used to obtain relative risks adjusted for potential confounding.³⁴ All logistic regression and Cox regression models used robust variance estimation with an independent working-correlation structure to account for correlation between eyes in bilaterally affected patients.³⁵ Stata 6.0 statistical software (Stata Corp, College Station, Tex) was used for all analyses.

From October 8, 1983, through March 31, 2000, 648 consecutive patients with AIDS and CMV retinitis affecting 971 eyes were evaluated at the Division of Ocular Immunology of the Wilmer Eye Institute at Johns Hopkins University. Information from the time of CMV retinitis diagnosis was available for all of these eyes except for 31 eyes (3.2%) of 21 patients who missed the measurement of visual acuity within 1 week of initial CMV retinitis diagnosis. In most cases, these patients were obtunded at the time of retinitis diagnosis. Neither the prevalence nor the incidence of visual impairment could be evaluated in these eyes, so they were excluded. All other eyes were included in the prevalence study.

One hundred three eyes (10.6%) of 95 patients had already experienced severe vision loss by the time of CMV retinitis diagnosis (visual acuity measurement of 5/200 or worse) and therefore could not be evaluated in the incidence study. Fifty-six eyes (5.8%) of 42 patients had no follow-up at our center after the CMV retinitis diagnosis was made. An additional 135 eyes (13.9%) of 106 patients did not have 2 or more follow-up visits spanning 28 days after CMV retinitis diagnosis, making the follow-up insufficient to meet our loss of visual acuity event definitions; this occurred mostly because of death shortly after CMV retinitis diagnosis. For 39 eyes (4.0%) of 27 patients, the medical records could not be located at the time of retrospective review for visual acuity scores. The remaining 607 eyes (of 426 patients) were longitudinally evaluated for the incidence of loss of 3, 6, or 10 lines of visual acuity. This group represents 72.5% of those who had a visual acuity measurement better than 5/200 at the time of CMV retinitis diagnosis.

Among the eyes excluded from the incidence study, 230 eyes (23.7%) were known to be at risk for loss of visual acuity events, 31 (3.2%) had unknown initial visual acuity status, and 103 (10.5%) had already experienced severe vision loss (5/200 or worse). Comparison of the eyes included and excluded from the incidence study according to lines of visual acuity lost demonstrated that the eyes of patients with a history of injection drug use, large CMV lesions, and involvement of zone 1 of the retina at the time of CMV retinitis diagnosis were more likely to be excluded (P<.05 for all 3 comparisons), primarily because a higher proportion in these groups had an initial visual acuity measurement of 5/200 or worse. Eyes of patients for whom the initial CD4+ T-cell count was missing, usually from early in the epidemic before this information was regularly used in clinical management, were also more likely to be excluded (P<.05 for all 3 comparisons), primarily because of inadequate follow-up (often a result of early mortality).

Demographic and clinical characteristics according to the antiretroviral and anti-CMV treatment groups are given in Table 1. Ninety-three patients (126 eyes) received HAART during follow-up. Of these, 46 patients (64 eyes) developed an immune recovery response, whereas 47 patients (62 eyes) were nonresponders. The remaining 333 patients(471 eyes) did not receive HAART, primarily because it was not available at the time of their illness. Fifty-five eyes received primary-implant therapy, whereas 59 eyes received secondary-implant therapy. During follow-up of eyes in the primary-implant group, 30 (55%) received only 1 implant, 13 (24%) received 2, 9 (16%) received 3, and 3 (5%) received 4 implants. Among eyes in the secondary-implant group, the distribution of the number of implants received was not significantly different from the primary-implant group (P = .37), with 41 (69%), 11 (19%), 5 (8%), and 2 eyes (3%) receiving 1, 2, 3, and 4 implants, respectively. The 493 eyes in the systemic therapy group never received implant therapy.

Patients who received HAART were more likely to undergo ganciclovir implant treatment for CMV retinitis than patients who did not receive HAART(P<.001). Patients who received secondary-implant therapy for CMV retinitis were less likely to be men who had sex with men(MSM) than patients in the systemic therapy group (P =.009). Eyes that received primary-implant treatment were significantly more likely to have zone 1 disease at the time of CMV retinitis diagnosis than eyes that received systemic therapy (P = .02) or secondary-implant therapy (P<.001).

In our patients, poor vision was often present in 1 or more eyes by the time of CMV retinitis diagnosis. Among affected eyes, the initial visual acuity measurement was 20/50 or worse in 33% and 20/200 or worse in 17% at the time of CMV retinitis diagnosis. Of the 237 patients whose CMV retinitis was bilateral at the time of initial diagnosis, 22% and 7.9%, respectively, had an initial visual acuity measurement of 20/50 or worse or 20/200 or worse in the better eye. Among 400 patients whose CMV retinitis was unilateral at the time of diagnosis, 35% and 17% of involved eyes had an initial visual acuity measurement of 20/50 or worse and 20/200 or worse, respectively.

The relationships of demographic, clinical, and treatment characteristics to poor visual acuity at the time of CMV retinitis diagnosis are given in Table 2. Patients with minority backgrounds had initial visual acuity measurements of 20/50 or worse (adjusted OR = 1.59; P = .01) or 20/200 or worse (adjusted OR = 1.53; P = .07) more frequently than white patients. Compared with eyes of patients who did not admit to injection drug use, the eyes of injection drug users were more likely to have initial visual acuity scores of 20/50 or worse (adjusted OR = 1.61; P = .04) or 20/200 or worse (adjusted OR = 1.95; P = .007). In contrast, the prevalence of initial visual acuity scores of 20/50 or worse(adjusted OR = 0.53; P = .008) or 20/200 or worse(adjusted OR = 0.57; P = .03) was lower for eyes of MSM than for eyes of other patients. Although most MSM were white and gave no history of injection drug use, the ORs for this characteristic remained significantly lower than 1 after adjustment for potential confounding by the other factors. Women had loss of visual acuity at the time of CMV retinitis diagnosis more frequently than men, but this association was due to confounding by other factors.

Eyes with initially large CMV retinitis lesions and eyes with initial involvement of zone 1 of the retina more frequently had reduced visual acuity at the time of CMV retinitis diagnosis. These characteristics usually indicate advanced disease. When adjustment for initial lesion size and initial lesion location was performed via multiple logistic regression analysis, the associations of most demographic factors with reduced visual acuity at the time of CMV retinitis diagnosis became nonsignificant (data not shown). There were 2 exceptions: MSM had a lower prevalence of visual acuity of 20/50 or worse (adjusted OR= 0.59; P = .05), and injection drug use remained associated with a higher prevalence of visual acuity of 20/200 or worse (adjusted OR = 1.90; P = .02) at the time of CMV retinitis diagnosis.

The relationship of demographic, clinical, and treatment characteristics to the incidence of loss of visual acuity following CMV retinitis diagnosis is indicated in Table 3. Vision loss with time was common, with 42%, 30%, and 23% of the 607 eyes at risk losing 3, 6, and 10 lines of visual acuity, respectively, within the first year of CMV retinitis diagnosis. Among the 467 affected eyes with visual acuity measurements better than 20/50 at the time of CMV retinitis diagnosis, 34% had values of 20/50 or worse 1 year after CMV retinitis diagnosis. Among the 573 eyes with initial visual acuity measurements better than 20/200 at the time of CMV retinitis diagnosis, 24% had values of 20/200 or worse at 1 year. Combining the incidence and prevalence of visual acuity loss, 56% of all eyes with CMV retinitis reached 20/50 or worse and 37% reached 20/200 or worse 1 year after CMV retinitis diagnosis.

Among the 84 patients who had visual acuity values better than 20/50 OU at the time of diagnosis of an initially bilateral CMV retinitis, 10% lost visual acuity in both eyes to the level of 20/50 or worse by 6 months after the diagnosis of retinitis, and 19% by 12 months. For the 111 patients who had visual acuity measurements better than 20/200 OU at the time of diagnosis of an initially bilateral CMV retinitis, 4.8% lost visual acuity in both eyes to the level of 20/200 or worse by 6 months after the diagnosis of retinitis, and 7.2% by 12 months. Combining the incidence and prevalence of bilateral loss of visual acuity, 37% of patients with initially bilateral CMV retinitis had a visual acuity value of 20/50 or worse in the better eye 1 year after CMV retinitis diagnosis, and 15% had a value of 20/200 or worse in the better eye at 1 year.

The relative risks of visual acuity loss according to demographic, clinical, and treatment characteristics, each adjusted for potential confounding by the other characteristics, are given in Table 4. The eyes of patients who received HAART had a substantially lower incidence of loss of visual acuity events compared with the eyes of patients who did not receive HAART (Figure 1). The eyes of patients observed to have immune recovery had the greatest benefit (Figure 2), with the adjusted relative risk of visual acuity loss ranging from 0.11 to 0.21 with respect to the eyes of patients who did not receive HAART and from 0.22 to 0.38 with respect to the eyes of HAART nonresponders, depending on the loss of visual acuity event evaluated. For all visual acuity loss events evaluated, the adjusted relative risk in the eyes of HAART nonresponders was about half that of eyes of patients who did not receive HAART, but this difference was statistically significant only for the loss of 3 lines (the analysis with the highest statistical power).

Initial (primary) and late (secondary) implant therapies were confounded heavily by the use of HAART, reducing our study's power to evaluate the outcomes of anti-CMV treatment. No significant differences were observed between the groups (Figure 3). Sensitivity analysis, in which events were counted only if a loss of visual acuity was observed to persist for a 42-day span or longer, resulted in slightly lower risk estimates for losses of visual acuity in the implant groups compared with the systemic therapy group (data not shown), but these findings did not differ qualitatively from the results of the analysis based on events persisting for 28 days or more. Sensitivity analysis counting secondary-implant eyes in the systemic therapy group and censoring follow-up at the time of initial implant placement also showed no significant difference between groups.

Among other potential risk factors for loss of visual acuity during follow-up, the eyes with lesions initially involving the posterior pole (zone 1) of the retina had a 1.5- to 2-fold higher risk of visual acuity loss than the eyes without initial zone 1 involvement after adjustment for confounding. Involvement of zone 3 was not associated with an increased risk of visual acuity loss (data not shown). The eyes of patients with bilateral retinitis at the time of diagnosis had an approximately 25% lower risk of loss of visual acuity than those with unilateral involvement after adjustment for potential confounding. Eyes that had large lesions at CMV retinitis diagnosis had a higher risk of visual acuity loss that reached 20/50 or worse, but there was no significant association for the other loss of visual acuity events evaluated. Although female gender, minority background, younger age, MSM exposure, and injection drug use were all significantly associated with an increased or decreased risk of loss of visual acuity in 1 or more analyses, these associations all became nonsignificant after adjustment for confounding. Initial CD4+ T-cell count was not associated with loss of visual acuity, so this variable was not included in multiple regression analyses.

Our requirement that a loss of visual acuity be sustained across successive visits spanning an interval of at least 28 days caused us to exclude 16.3% of eyes with initial visual acuity values better than 5/200, the main reason for exclusion. We imposed this requirement to avoid counting early transient postoperative vision loss in eyes that had undergone implants as equivalent to permanent irreversible vision loss. However, the variables of interest other than type of anti-CMV treatment used did not require such a restriction to obtain valid adjusted relative risk values. To explore whether the exclusion of these eyes altered the conclusions for the other potential risk factors, we also conducted sensitivity analysis without requiring that loss of visual acuity be sustained. Carrying out this procedure resulted in adjusted relative risk values similar to those reported in Table 4.

Our results demonstrate that CMV retinitis in patients with AIDS is a condition associated with a high degree of visual morbidity. By the time the CMV retinitis diagnosis was made, 1 in 13 patients with initially bilateral CMV retinitis was legally blind in both eyes, and nearly 1 in 5 had bilateral visual impairment (20/50 OU or worse). As expected, the proportion of eyes with visual impairment at the time of diagnosis was higher than the number of patients with bilateral visual impairment, with 1 in 6 having an initial value of 20/200 or worse and one third with an initial measurement of 20/50 or worse. These cases of vision impairment represent a substantial burden of disease that is potentially avoidable. The magnitude of this problem presumably decreased during the late 1990s because of the reduced incidence of CMV retinitis in the era of HAART.^{5 - 7} However, because the number of patients with AIDS continues to rise in the United States, ³⁶ many of whom fail to achieve complete suppression of HIV replication despite the use of HAART, ^{37 - 40} the number of patients losing vision from CMV retinitis may increase in the future. Therefore, effective strategies are needed to prevent vision loss due to CMV retinitis. Our observation that loss of visual acuity commonly occurs among HAART nonresponders despite anti-CMV treatment suggests that prevention would be preferable to early detection.

An important observation from the prevalence study was that minority background and injection drug use were associated with a substantially higher rate of visual impairment at the time of initial CMV retinitis diagnosis. However, these factors were less predictive of visual acuity loss during follow-up(while receiving health care). Adjustment for indicators of more advanced disease at the time of retinitis diagnosis eliminated most of these associations, suggesting that patients with these characteristics may have accessed health care less promptly than others. As the AIDS epidemic in the United States increasingly evolves toward racial minorities and injection drug users, systems to improve access to care for these groups are needed.

The incidence study demonstrated that loss of visual acuity with time in cases of CMV retinitis is common despite treatment. However, our study also found a large reduction in the incidence of visual impairment in patients who received HAART, particularly when immune recovery was observed. We have previously reported that HAART reduces retinal detachment risk in cases of CMV retinitis.¹⁸ In the same population, the effect of HAART in reducing the risk of loss of visual acuity was even stronger. Presumably, improved immunity resulted in fewer retinitis progressions that involved the macula and optic disc as well as fewer retinal detachments, explaining the additional benefit.

As noted previously, the incidence of IRU in our cohort²² was much lower than in another reported series.²³ Because vision loss from IRU can sometimes be severe, ⁴¹ it is possible that the benefit of HAART in preventing loss of visual acuity would be less in a population with a high rate of IRU. However, the available literature^{21 ,41 - 43} and clinical experience suggest that IRU generally causes mild to moderate rather than severe vision loss, whereas severe vision loss from progressive retinitis and retinal detachment is common in patients who do not receive HAART. Therefore, HAART may reduce the risk of visual acuity loss even in populations with high rates of IRU. It is possible that some patients who develop severe IRU will be exceptions to this generalization and may require different clinical management. However, the proportion of patients for whom HAART will cause enough visual disability to outweigh its systemic benefit is probably small.

Because of a high degree of confounding by HAART, our incidence study had limited power to compare ganciclovir implant therapy with systemic treatment for CMV retinitis. After adjusting for confounding, no significant differences were observed in the risk of loss of visual acuity. The agreement of these results with those from previous trials^{24 ,26} suggests no large differences in visual acuity outcomes between implant and systemic therapies for the treatment of CMV retinitis. However, the limited power of these 3 studies leaves the question open regarding whether small to moderate differences in visual outcomes may exist. We did observe a few cases in which transient loss of visual acuity after ganciclovir implant surgery persisted for more than 28 days, demonstrating that in clinical practice, visual recovery after this surgery is occasionally protracted.

The only nontreatment factor that was consistently associated with increased incidence of loss of visual acuity, after adjustment for confounding, was involvement of the posterior pole of the retina (zone 1) at the time of CMV retinitis diagnosis. Because patients with posterior pole involvement and initially preserved visual acuity are at risk for irreversible vision loss from even small amounts of retinitis progression, they must be treated carefully. A panel of experts has recommended considering the use of ganciclovir implant therapy for such eyes because of its greater efficacy in preventing retinitis progression.⁴⁴ Unfortunately, our data were not sufficient to evaluate whether this strategy was more successful than alternatives for eyes with immediately vision-threatening CMV retinitis.

Methodologic concerns exist for our studies of the prevalence and incidence of visual acuity loss in patients with AIDS and CMV retinitis. One weakness is that we measured visual acuity for the purpose of clinical care rather than using a standardized measurement protocol. Our practice has been to perform refraction at any visit when visual acuity is abnormal or visual complaints exist, and we took additional steps to reduce visual acuity measurement error from inadequate refraction (see "Methods" section). Nevertheless, our approach may have been more prone to measurement error than a standardized approach. This problem is of less concern for the incidence study than the prevalence study; consecutive measurement errors in the same direction for visits 28 days apart or more would be needed to create a false event for the former. To evaluate the potential effect of measurement error in our study, note that visual acuity errors would usually be small; large errors would have provoked refraction and retesting. Therefore, measurement error would have affected the events consisting of a smaller degree of vision loss (eg, loss of 3 lines) more than those consisting of a several-fold increase in the visual angle. However, the patterns of association were quite similar across all events for both the prevalence and incidence studies, suggesting that visual acuity measurement error did not greatly affect our study.

Other concerns include the lack of randomization as a mechanism to prevent selection bias as well as the possibility that the distribution of cases at a tertiary center such as ours may not be generalizable to an average population of patients with AIDS and CMV retinitis. We attempted to address potential selection bias by adjustment for possible risk factors for loss of visual acuity. Our results with respect to implant vs systemic therapy for CMV retinitis were similar to those observed in randomized trials of initial therapy with a ganciclovir implant–based regimen compared with parenteral ganciclovir²⁴ or cidofovir.²⁶ Regarding the representativeness of the population studied, we have estimated that at least 80% of patients with CMV retinitis in our metropolitan region receive their eye care at our center.¹⁸ Our population included relatively large numbers of African Americans, women, and injection drug users, which reflects the demographic characteristics of the AIDS epidemic in our region and may reflect the future of the national epidemic. Although patients admitting to a history of injection drug use as well as those with large lesions and posterior pole involvement at diagnosis were more likely to be excluded from the incidence study (primarily because they had already experienced severe vision loss prior to retinitis diagnosis), these groups were not differentially excluded from the prevalence study.

In summary, reduced visual acuity at the time of initial diagnosis was common in our patients with AIDS and CMV retinitis. Patients with minority backgrounds and injection drug users appeared to access care less promptly than other groups. Vision loss was also common during follow-up despite clinical treatment of these diseases. Highly active antiretroviral therapy was associated with a large reduction in the risk of loss of visual acuity, particularly when immune recovery was observed. No difference in the risk of visual acuity loss was observed between eyes treated with the ganciclovir implant and those that received systemic anti-CMV therapy.

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

Submitted for publication June 11, 2001; accepted May 24, 2002.

This study was supported in part by grants EY00386 and EY07127 (Dr Kempen) and EY10268 and EY00405 (Dr Jabs) from the National Eye Institute, Bethesda, Md.

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