Rates and Predictors of Hydroxychloroquine Retinal Toxicity in Patients With Rheumatoid Arthritis and Systemic Lupus Erythematosus

Transcription

1 Arthritis Care & Research Vol. 62, No. 6, June 2010, pp DOI /acr , American College of Rheumatology SPECIAL ARTICLE: DRUG SAFETY IN THE RHEUMATIC DISEASES Rates and Predictors of Hydroxychloroquine Retinal Toxicity in Patients With Rheumatoid Arthritis and Systemic Lupus Erythematosus FREDERICK WOLFE 1 AND MICHAEL F. MARMOR 2 Objective. Hydroxychloroquine (HCQ) retinopathy is of concern because of the potential seriousness of visual loss and the medicolegal consequences of failure to detect toxicity. However, there have been limited demographic data on which to base recommendations for screening. We have studied the largest unselected series of patients to date to evaluate the risk of toxicity and the relevance of purported risk factors. Methods. We studied 3,995 patients with rheumatoid arthritis or systemic lupus erythematosus who had used HCQ, including 1,538 current users. We screened for self-reported toxicity, and followed up on positive cases with detailed interviews and specialist confirmation. We categorized cases as definite or probable if there was bull s eye maculopathy or visual field loss. Results. Of the lifetime users of HCQ, 6.5% discontinued therapy because of an eye problem, including 1.8% who reported HCQ retinal problems. However, definite or probable toxicity was documented in only 0.65% (95% confidence interval ). The risk of toxicity was low in the initial 7 years of exposure, and was approximately 5 times greater after 7 years of usage (or 1,000 gm total exposure). Toxicity was unrelated to age, weight, or daily dosage. Eye examinations were obtained annually by 50.5% and every 6 months by 40.4% of patients. Conclusion. HCQ toxicity remains uncommon, but increases markedly with the duration of therapy and exceeds 1% after 5 7 years. Toxicity was unassociated with age, daily dosage, or weight. These findings will aid the reformulation of screening guidelines. INTRODUCTION Despite advances in the therapy of rheumatoid arthritis (RA), antimalarial therapy, principally as hydroxychloroquine (HCQ), has remained a component of RA treatment for more than half a century, and remains the key therapy of systemic lupus erythematosus (SLE). Through all of this time, a central concern of HCQ treatment has been its potential for retinal toxicity. The mechanism of HCQ toxicity is not well understood. HCQ binds to melanin in the retinal pigment epithelium (RPE), and this binding may serve to contribute to or prolong toxic effects. However, no anatomic features of the epithelial layer correlate with the parafoveal pattern of 1 Frederick Wolfe, MD: National Data Bank for Rheumatic Diseases and University of Kansas School of Medicine, Wichita; 2 Michael F. Marmor, MD: Stanford University, Stanford, California. Address correspondence to Frederick Wolfe, MD, National Data Bank for Rheumatic Diseases, 1035 North Emporia, Suite 288, Wichita, KS fwolfe@ arthritis-research.org. Submitted for publication November 3, 2009; accepted in revised form February 4, RPE loss (bull s eye maculopathy) that is most characteristic of toxicity (1). Patients may notice paracentral scotomata with early toxicity, but toxicity may progress despite discontinuation of the drug, and can lead to more serious loss of central and peripheral vision (1), for which there is no known treatment. Thus, screening for the earliest signs of retinal toxicity has been the core of management for the condition. The decision to use HCQ and how its use should be monitored should be based on the actual risk of toxicity, its seriousness and reversibility, as well as the cost of toxicity monitoring. However, there have been little demographic and clinical data available on which to base screening policy. The largest published series of 1,207 unselected rheumatic patients showed 1 case of definite toxicity and 5 cases of possible toxicity (2), and the bulk of the literature on toxicity has come from patients referred because of symptoms or signs of potential retinal damage. Specialty society groups such as the American Academy of Ophthalmology and the American College of Rheumatology (ACR) have made recommendations for monitoring of HCQ therapy (1,3,4), but without complete epidemiologic knowledge of the rate of toxicity and associated risk factors. 775

2 776 Wolfe and Marmor The current state of knowledge and beliefs about HCQ toxicity has been summarized in a review article by Yam and Kwok (5). The validity and reliability of assumptions about HCQ toxicity and its assessments relative to socioeconomic costs have also been challenged in the analyses of Silman and Shipley (6). In an effort to resolve these uncertainties, we have evaluated the largest population of HCQ users to date, selected only by their willingness to participate in a national rheumatologic data bank. We have determined the risk of definite or possible toxicity within this population, and examined purported risk factors for the development of toxicity. We also provide estimates documenting current surveillance patterns among patients treated by US rheumatologists. Our data suggest a need to reevaluate aspects of current recommendations on screening for HCQ retinopathy. PATIENTS AND METHODS Table 1. Characteristics of 3,995 RA and SLE lifetime users of HCQ* Variable N Value Full cohort 3,995 Age, years 3, (61.8) Sex, % male 3, Education category, 3,995 % years Weight, kg 3, (72.7) Ideal weight, kg 3, (56.8) Over ideal weight, % 3, RA, % 3, Disease duration, 3, (16.8) years Duration of HCQ use, 3, (3.7) years HCQ dose, mg 2, (400.0) Current users 1,538 HCQ dose, mg 1, (400.0) HCQ 400 mg per 1, day, % HCQ dose, mean 1, SD mg/kg Dose 6.5 mg/kg, % 1, Dose 6.5 mg/kg of ideal weight, % 1, * Values are the mean SD (median) unless otherwise indicated. RA rheumatoid arthritis; SLE systemic lupus erythematosus; HCQ hydroxychloroquine. Participants in this study were part of the National Data Bank for Rheumatic Diseases (NDB) ongoing longitudinal observational study of rheumatic disease outcomes (7,8). Beginning in 1998, NDB participants were recruited from the practices of US rheumatologists and followed prospectively with semiannual, detailed, 28-page questionnaires. This program followed the Declaration of Helsinki, and was approved by the St. Francis Regional Medical Center Institutional Review Board. Recruitment was unrelated to use of HCQ. Participants are volunteers who have a variety of rheumatic diseases (Table 1). They are not compensated for their participation. From 6,307 patients (5,669 with RA, 638 with SLE) evaluated in 2008, we identified 3,995 (63.3% overall, 60.1% RA, and 92.2% SLE) who had received HCQ over the course of their illness. Of those, 1,538 patients (38.5%; 474 with lupus and 1,064 with RA) were still taking HCQ (Figure 1). The mean SD age in years and the percentage of males were and 5.2% for SLE and and 20.0% for RA, respectively. Participants completed a special questionnaire about visual problems (see Supplementary Appendix A, available in the online version of this article at interscience.wiley.com/journal/ /home). As the major purpose of the study was to evaluate the effect of antimalarial therapy on visual problems, we combined RA and SLE patients who had used HCQ into a single group for detailed analysis. For evaluation of dose by weight, we defined ideal body weight as 50 kg plus 2.3 kg per inch (2.54 cm) over 5 feet (152.4 cm) for men and 45.5 kg plus 2.3 kg per inch over 5 feet for women (9). Following completion of the January 2008 questionnaire, we instituted additional followup contact with patients who attributed visual problems to antimalarial treatment. With these patients, we confirmed the original questionnaire data, obtained additional details of symptoms, determined the type of physician who had been consulted for the antimalarial-associated visual problems, and determined the name and/or address of that physician if that physician was an ophthalmologist or optometrist. We then wrote to those specialists concerning their evaluation. Eye specialists were asked: 1) Did the patient have evidence of toxicity from these drugs (antimalarials)? 2) Was there loss of acuity or visual field? 3) Were there any retinal changes? and 4) Was there any evidence of bull s eye maculopathy? We also asked for textual descriptions Figure 1. Distribution of duration of hydroxychloroquine (HCQ) therapy in lifetime HCQ users (histogram) and cases (n 10) of HCQ toxicity at the time of diagnosis.

3 Hydroxychloroquine Retinal Toxicity 777 of the findings and copies of medical records, including visual fields. Patient population groups. Among patients who had received antimalarial therapy while being followed in the NDB, only 0.5% had received an antimalarial compound other than HCQ. These patients were not evaluated in this study. From the 3,995 patients with RA or SLE, we identified 298 who stated a physician told them that they had eye problems caused by HCQ ( Were you ever told by a physician that hydroxychloroquine had caused a problem with your eyes? ). Patients who answered affirmatively were selected for the followup interview, and those who did not were presumed to have no HCQ toxicity. Of the 298 patients, we were unable to obtain followup information on 83. Of the remaining 215 patients, 165 stated they had been seen by an eye specialist (22.4% by an optometrist, 77.6% by an ophthalmologist), while 50 had been seen by a rheumatologist or a primary care physician. We classified patients not seen by an eye specialist as not having HCQ ocular toxicity, under the assumption that physicians would refer patients for a specialist examination if they suspected eye toxicity. Because some patients did not remember the name or location of the eye specialist and not all specialists responded to our request for information, we were able to obtain evaluation data from only 84 of the 165 specialists. Determination of retinal toxicity. All specialist records and replies that implicated retinal toxicity were evaluated by a retina specialist (MFM) in the absence of information about dose or duration of use. The cases were classified as definite or probable toxicity or possible toxicity (see Supplementary Appendix B, available in the online version of this article at journal/ /home). Since the most distinctive feature of HCQ retinopathy is bull s eye maculopathy, evident in fundus changes and/or in central visual fields, we considered this as the most critical determinant of toxicity. This clinical finding is quite specific to HCQ toxicity, and is unlikely to result from other potential retinal diseases in this population such as vasculopathy (which shows very different clinical findings) and age-related maculopathy (which is typically central and does not follow a bull s eye pattern). Nonspecific retinal findings that might as well be on the basis of age or other disease were considered possible concomitants of toxicity, but not definitive. Unfortunately, there are no established criteria for the diagnosis of very early toxicity so that it is hard to rule out the possibility of toxicity in cases that were labeled toxic by the examining doctors. Cases reported to have bull s eye maculopathy. There were 9 such cases, but the supplemental data from 2 were not consistent with bull s eye maculopathy and were thus classed as probable. One showed only minimal macular pigment mottling in a photograph, and field loss turned out to be on the basis of prior optic neuropathy (possible case 9; Supplementary Appendix B, available in the online version of this article at com/journal/ /home). The other had completely normal central fields, unlikely in the presence of visible bull s eye maculopathy (possible case 6; Supplementary Appendix B, available in the online version of this article at home). The other cases are considered definite or probable. Cases not reported to have bull s eye maculopathy. Of 14 such cases, 11 had varying degrees of nonspecific maculopathy or normal fundi and/or central field loss without any bull s eye pattern, and these were all classified as possible. In most, the fields were unlikely to be from HCQ toxicity because of high error scores, associated peripheral loss, variability on retesting, etc. However, 2 cases showed central fields with a distinct paracentral (bull s eye) pattern of sensitivity loss (definite or probable cases 8 and 10; Supplementary Appendix B, available in the online version of this article at wiley.com/journal/ /home), and 1 additional case had clinic notes describing crescentic parafoveal depigmentation associated with central loss on a less sensitive field test (definite or probable case 9; Supplementary Appendix B, available in the online version of this article at home). Overall, the population of 84 patients who had documented eye doctor examinations included 10 cases of definite/probable toxicity and 13 cases of possible toxicity (see Supplementary Appendix B, available in the online version of this article at com/journal/ /home), which were then analyzed relative to parameters of demographics and medication usage. Statistical methods. Missing data were noted for the following study variables: age, sex, education (0%), weight (1.2%), ideal weight (0.9%), HCQ dose (39.3%), and duration of HCQ therapy (0%). For the 298 patients who reported toxicity, data were missing for weight (1.3%), ideal weight (1.0%), HCQ dose (46.3%), eye specialist determination of toxicity (53.0%), and type of physician or eye specialist determining toxicity (30.5%). As the NDB collects data on current medications for all participants, the missing dose data entirely reflected use of HCQ by patients before they participated in the NDB. We calculated age as the age at last use of HCQ in the NDB. If HCQ was discontinued prior to entry into the NDB, we used age at NDB entry. In order to determine the rate of HCQ toxicity, it was necessary to estimate the rate of toxicity in cases where patients reported toxicity, but for which we lacked specialist reports. As noted above, some patients did not remember the name or location of the eye specialist, not all specialists responded to our request for information, and some patients were unable to tell us the type of specialist or physician they had consulted. There were two possible approaches to estimating toxicity in the missing cases. First, using the observed cases and specialist/nonspecialist distribution to extrapolate to the unobserved cases. This method, however, does not incorporate any measure

4 778 Wolfe and Marmor Table 2. Toxicity and eye examinations associated with HCQ therapy* Variable Value Patient-reported data, % Physician stated HCQ toxicity 7.5 Physician stopped HCQ for eye problem 6.5 Physician stopped HCQ because on it too long 5.1 Eye specialist validated toxicity Prevalence per 1,000 patients, rate (95% CI) (extrapolated estimate) Definite or probable HCQ toxicity 6.5 ( ) (7.3) Possible HCQ toxicity 10.4 ( ) (9.5) Incidence per 1,000 patient-years, rate (95% CI) Definite or probable HCQ toxicity 1.1 ( ) Possible HCQ toxicity 1.7 ( ) Frequency of eye examinations for toxicity, % Every 6 months or less 40.4 Every year 50.5 Every 2 years 5.3 Every 3 years 0.6 Every 4 years 0.4 Every 5 years or more 0.7 Never 2.1 * HCQ hydroxychloroquine; 95% CI 95% confidence interval. See Patients and Methods section for details. of the uncertainty that is inherent in extrapolation, and does not make use of age, sex, weight, ideal weight, or duration or dose of HCQ therapy. A second approach to missing data is to use imputation. We used multiple imputation (MI) of missing data in order to provide confidence intervals that expressed the uncertainty about the estimates. The MI method used was imputation with chained equations, with bootstrap sampling of the non-missing observations (10). Bootstrap sampling has the advantage of robustness since the distribution of predictor variables is no longer assumed to be multivariate normal. In these analyses we created 200 imputed data sets. Imputation of toxicity was performed on cases reported as positive for toxicity by patients, and included age, sex, education, duration of therapy, dose, type of physician or eye specialist determining toxicity, weight, and ideal weight in the imputation process. To call attention to possible differences in point estimates estimated by extrapolation and by MI, we have also included extrapolated point estimate results in Table 2. Except as indicated in the results, we report the results of analyses using imputed data. As part of sensitivity and validation analyses, we also analyzed the data using casewise deletion for missing data and found agreement in results between methods. Among cases of specialist-reported toxicity, 4 of 10 cases had missing dose data. Therefore, analyses that rely on dose data should be considered to have limited validity. In analyses that used multiply imputed data sets, we modeled the univariate and multivariable relationship between toxicity and duration of therapy with logistic regression and the association of toxicity with other variables using Cox regression. Figure 2 shows Kaplan-Meier estimates of the probability of definite/probable toxicity over the duration of therapy. To more accurately represent and smooth the data, we included results from all of the 200 imputations in this figure and also in a hazard rate estimate in Figure 3. This method resulted in 5,214 failures for 799,000 observations, or an average failure rate of 26 per imputation. There is no way to determine appropriate standard errors for these survival data, so they are shown as point estimates without confidence intervals in the figures. Predictors of the frequency of eye examinations were analyzed using ordered logistic regression. Time at the start of HCQ exposure was determined from the patients reports of time on HCQ. The NDB tracks all treatment time for patients in the NDB, and inquires about Figure 2. The estimated probability of absence of definite or probable retinal toxicity as a function of hydroxychloroquine exposure time using the Kaplan-Meier survivor function. Data represent results from 200 multiply imputed data sets. Exact survivor function results are at 5 years, at 10 years, at 15 years, at 20 years, at 25 years, at 30 years, and at 35 years.

5 Hydroxychloroquine Retinal Toxicity 779 Figure 3. The smoothed estimate hazard for duration of hydroxychloroquine (HCQ) therapy and cumulative dose of HCQ. At the mean daily dosage of HCQ used in the study (344 mg), 1,000 gm of HCQ will be used in 7.8 years. treatment times prior to NDB entry at the time patients join the study. In the survey questionnaire, we also asked patients to report the total time they had used HCQ. In instances where currently reported time was less than NDB recorded time, we used NDB recorded time. We also telephoned patients who reported long durations of HCQ and interviewed them to determine if their reports were correct. In general, we were able to verify patients reports. Patients data were not further analyzed (i.e., were censored) beyond the time of HCQ discontinuation if they had discontinued HCQ prior to study completion (n 2,447 [61.3%]) or if they were still receiving HCQ but did not have toxicity at the end of the study (n 1,538 [38.5%]). In separate analyses by logistic and Cox regression, we found no statistical suggestion of an increased or separate association of toxicity with lupus, and we therefore pooled RA and lupus patient data. However, only 17% of patients had lupus, and only 1 case of toxicity was noted in lupus patients insufficient cases to appropriately evaluate toxicity in lupus. Data were analyzed using Stata, version 10.1 (11). All tests were 2-tailed, and significance was set at RESULTS Patient and treatment characteristics. The duration of HCQ therapy among lifetime HCQ users had a mean SD of and a median of 3.7 years (Table 1), and 25% of patients used HCQ for at least 9.5 years (Figure 1). Among the 1,538 current HCQ users, 27.7% used 400 mg of HCQ per day, 70.7% used 400 mg, and 1.6% used 400 mg of HCQ per day. These relationships were very similar among the prior users who did report the dose (see Supplementary Appendix B, available in the online version of this article at /home). As 6.5 mg/kg ideal weight has been considered a cutoff for a safe daily dosage of HCQ (1,5,12), we examined dose and weight relationships for the current users (Table 1). The mean SD dose in mg/kg was , and only 12.6% of patients exceeded 6.5 mg/kg. However, population averages may be misleading because virtually all of the patients fell into 1 of 2 fixed-dosage groups: either 400 or 200 mg/day. For the 400 mg/day users, the mean SD dose in mg/kg was , and for 200 mg/day users, it was For the 400 mg/day users, 16.3% exceeded 6.5 mg/kg. When ideal weight instead of actual weight was used, 53.6% exceeded the 6.5 mg/kg mark, with 77.0% using 400 mg per day and 0.7% using 200 mg per day. The relationship between weight and HCQ dose for all of the study subjects is shown graphically in Figure 3. In that figure, weights 61.5 kg represent HCQ doses 6.5 mg/kg at 400 mg per day. The mean ideal weight for study patients was 57.4 kg compared with 76.7 kg for actual weight. For ideal weights, the mean SD mg/kg was for 400 mg users, for 200 mg users, and overall. With respect to other medical factors assumed to be related to HCQ toxicity, none of the patients with identified HCQ toxicity in this study reported current liver or renal comorbidity compared with 3.9% with renal disease and 2.0% with liver disease among patients without toxicity. Considering published American Academy of Ophthalmology risk categories other than duration of use (age 60 years, dose 6.5 mg/kg of ideal body weight), 81.5% of all current HCQ users can be considered to be in a high-risk category. HCQ and HCQ toxicity. Of the 3,995 RA and SLE lifetime users of HCQ (Table 2), 7.5% reported that a physician had told them that HCQ had caused an eye problem and 1.8% reported HCQ-attributed retinal problems. HCQ had been discontinued by 6.5% of patients because of an eye problem, and by 5.1% because they had been on HCQ too long. The mean SD and median durations of HCQ among those who had taken it too long were and 5.0 years, respectively. HCQ toxicity. In our cohort, definite or probable HCQ toxicity occurred in 6.5 (95% confidence interval [95% CI] ) per 1,000 patients, and possible toxicity occurred in 10.4 (95% CI ) per 1,000 patients (Table 2). The characteristics of the individual patients with toxicity are shown in Supplementary Appendix B (available in the online version of this article at wiley.com/journal/ /home). According to the eye specialist data, 90% of patients we classified as having definite or probable toxicity had some loss of visual acuity or visual field, and 92% of those with possible toxicity had such loss. Bull s eye retinal lesions were reported in 70% of definite or probable cases and 15% of possible cases. The classification of these cases, relative to the eye specialist report, was described in the Patients and Methods section and Supplementary Appendix B (available in the online version of this article at wiley.com/journal/ /home). Effect of treatment duration on toxicity. Figure 1 shows the distribution with time of the 10 observed cases with

6 780 Wolfe and Marmor definite or probable toxicity. Figure 2 shows the data using 200 multiply imputed data sets in a Kaplan-Meier life table survivor analysis, and Figure 3 displays a hazard rate analysis. Because of the question of when HCQ toxicity surveillance should begin in clinical practice and at what intervals monitoring should continue, we were particularly interested in when the risk for toxicity begins and whether and how the risk increases with time. Figure 3 suggests that there is a low rate of toxicity during the first few years of usage, which increases sharply after approximately 6 years or a cumulative dose of 800 gm and continues to rise for approximately 12 years. The rate increases progressively for approximately 12 years. From Figure 2, it is possible to obtain a more accurate estimate of progression. Point estimates of the probability of toxicity displayed in Figure 2 are: 0.29% at 5 years, 0.33% at 7 years, 1.0% at 10 years, 2.1% at 15 years, and 3.1% at 20 years. We also assessed the risk of toxicity with formal statistical analyses. Adjusted for age, sex, dose weight, ideal weight, and education, the odds ratio (OR) for definite or probable toxicity for a 5-year increase in duration of HCQ exposure was 1.4 (95% CI ). The OR for 7 years of HCQ use compared with 7 years of HCQ use was 5.1, and the OR for a cumulative dose of 1,000 gm compared with 1,000 gm of HCQ was 4.5 (95% CI ). There was no significant effect of duration of HCQ use on the patients with possible HCQ toxicity cases (P 0.500), reinforcing doubts about whether these cases actually represent HCQ retinopathy (see Discussion section). Effect of daily treatment dosage on toxicity. Most patients in the study simply took a daily dosage of 400 or 200 mg. We did not detect any relationship between daily HCQ dosage and definite or probable toxicity either in the imputed (hazard ratio [HR] 1.8, 95% CI ; P 0.480) or raw data analyses (HR 0.7, 95% CI ; P 0.498). However, only 6 of the 10 probable or definite toxic cases had dose data. We also found no significant associations between possible HCQ toxicity and dose. Effect of weight on toxicity. Heavier patients were more likely to be prescribed 400 mg of HCQ than 200 mg (OR 1.11, 95% CI ) per 5 kg weight difference, and for those above the median weight (72.7 kg) compared with those below that weight, the OR was 1.8 (95% CI ). Weight was not associated with the risk of toxicity in univariate (HR 1.0, 95% CI ; P 0.091) or multivariable analyses (HR 1.0, 95% CI ; P 0.116). Ideal weight was also unassociated with toxicity in univariate (HR 1.0, 95% CI ; P 0.194) and multivariable analyses (HR 1.0, 95% CI ; P 0.648). There was also no association of weight or ideal weight with possible retinal toxicity. It has been suggested that doses 6.5 mg/kg of ideal weight and HCQ dose are risk factors for toxicity (1,5,12). Although dose data have limitations in this study, it is virtually impossible to exceed the 6.5 mg/kg cut point while taking 200 mg per day. At 400 mg per day, only patients with an ideal weight of 61.5 kg will exceed that cut point. We were, therefore, able to examine the cut point effect. In a univariate model, 6.5 mg/kg of ideal weight was not associated with a statistically significant increased risk of retinal toxicity (HR 2.1, 95% CI ; P 0.270), and in the multivariable model, there was no relationship with toxicity (HR 1.1, 95% CI ; P 0.888). Similarly, no association was noted with possible toxicity. Effect of age on toxicity. We examined the effect of age in univariate and multivariable analyses. There was no significant association of age and toxicity. As age 60 years has been reported to be associated with toxicity, we examined its effect. The univariate HR for age 60 years was 2.3 (95% CI , P 0.213), and was 1.9 (95% CI , P 0.350) in multivariable analysis. The mean SD age of patients with probable or definite toxicity was years compared with for those without toxicity, but the age at which toxicity was discovered might actually be lower because 4 of the 10 cases with probable or definite toxicity had stopped the drug prior to entering the database, when their age was first recorded. Frequency of eye examinations and eye examination predictors among current HCQ users. Examinations to detect HCQ toxicity occurred at 6-month intervals in 40.4% of current HCQ users and at yearly intervals in 50.5% (Table 2). Examination intervals less frequent than yearly were noted by 0.7%, and 2.1% reported never having a toxicity examination (Table 2). To understand the associations between eye examination frequency and demographic and treatment factors, we regressed the frequency of examinations on various combinations of age, sex, education level, RA diagnosis, dose of HCQ, and duration of HCQ therapy using ordered logistic regression (omitting the never category of eye examinations). In the full model, neither HCQ dose nor duration of therapy was associated with eye examination interval (P 0.05), but a 10-year reduction in age was associated with a 17% increase in the probability of more frequent eye examinations (OR 1.17, 95% CI ); males were 40% more likely to have less frequent examinations (OR 1.40, 95% CI ). There was no association between education and diagnostic category (RA or SLE) and the frequency of eye examinations in this model. Reduced models (omitting one or more covariates) produced similar results. Simplifying the frequency of eye examinations to every 6 months compared with 1 year or greater and analyzing the same variables by logistic regression did not appreciably change the results. These data suggest that patients obtain eye examinations that are independent of the intensity and/or duration of HCQ therapy. DISCUSSION The results of this study show that definite or probable HCQ retinal toxicity occurred at a rate of 6.5 cases per 1,000 HCQ users in a large unselected population, but the rate was less than 3 per 1,000 within the first 5 years of usage, and rose toward 20 per 1,000 (2%) between 10 and

7 Hydroxychloroquine Retinal Toxicity years of continuous usage. Thus, HCQ is an infrequent event, even for long-term users; but the risk becomes increasingly significant beyond 5 7 years given that the outcome may be irreversible blindness. Doses 6.5 mg/kg of ideal body weight were not significantly associated with toxicity, and age was not associated with toxicity. Although retinal toxicity is easy to recognize when advanced to the point of a classic bull s eye maculopathy, there are no established criteria or diagnostic procedures that define or recognize early retinopathy. Mild changes in vision, visual field, or macular pigmentation can raise concern over HCQ toxicity, but may also result from other causes, including age-related maculopathy, cataract, optic nerve dysfunction, or vasculopathy from lupus. In the current study, we defined two levels of toxicity based on review of examining specialists records and opinions: definite or probable toxicity and possible toxicity. We have restricted the definition of definite or probable toxicity to cases with evidence for bull s eye damage on either retinal examination or in the visual field. Bull s eye maculopathy is highly characteristic of HCQ toxicity, but is not found in lupus or rheumatoid vasculopathies and would be atypical for age-related macular degeneration. Thus, there should be few false-positives. Our estimates may be on the low side (given than we may have missed some mild cases of toxicity). On the other hand, we suspect that most of our possible cases (derived from eye examination records indicating that the patient had any evidence of toxicity, but for which drug-related retinal changes could not be verified; see the Patients and Methods section) have causes other than HCQ. There was no relationship between possible toxicity and either duration of use or cumulative drug dose, factors that are clearly relevant to the development of definite or probable retinopathy both in this study and in a large body of literature. Review of the 4 largest series of HCQ users (2,5,13 16) indicates that HCQ retinopathy is rare, ranging from less than 0.1% to 0.5% (mostly on the low side). The largest previous series by Levy et al identified 1,207 patients who took HCQ between 1991 and 1993 in multiphysician practices (2). They found 1 case of definite toxicity and 5 cases of what they called probable (but what we would classify as possible) toxicity. Depending on how one accepts their definition of toxicity, this amounts to a prevalence range from 0.8 to 5.0 per 1,000 patients. For clear toxicity, their estimate is well below our own of 6.5 of 1,000. However, the characteristics of the subjects in the 2 studies were different: while similar in weight, daily dosage, and dose/ kg, the subjects in the study by Levy et al were younger (46.9 years versus 58.4 years), and had on average used HCQ for only 3.3 years relative to a mean of 6.0 years in our sample. Since risk rises progressively with duration of use (Figures 2 and 3), this may account for our higher prevalence of documented toxicity. Our finding that risk rises to 1% and more after 5 7 years of use has not previously been documented in a large series, and has important implications for the public health impact of the disorder and for cost benefit analyses with respect to screening. A considerable body of literature and the 2002 screening recommendations of the American Academy of Ophthalmology (1,2) have emphasized the importance of daily dosage, in particular a dose below 6.5 mg/kg. However, our data showed no relationship between HCQ retinal toxicity and daily dosage by either measured weight or ideal body weight. It is important to examine both the history of this widely-used recommendation, and possible implications of our results. The recommendations for the determination of weight and, therefore, for the use of 6.5 mg of HCQ per kg of body weight as a guide to appropriate dosing of HCQ, are inconsistent in the literature with respect to how weight is calculated. Some studies specifically refer to ideal body weight (5,12,17), whereas others simply use the term weight. For example, Fraenkel and Felson state Several studies have shown that retinal toxicity is exceedingly rare in patients taking HCQ at 6.5 mg/kg lean body weight/ day (17). However, a number of those studies make no mention of lean body weight. In their review of 47 studies, Yam and Kwok indicate the importance of ideal body weight, but do not distinguish between measured weight and ideal weight in their table of studies (5). The American Academy of Ophthalmology recommendations that were subsequently endorsed by the ACR (position statement approved on February 22, 2006) (18) do not mention ideal weight, but note that high body fat or obesity is an added risk factor. Yam and Kwok comprehensively reviewed the literature of HCQ toxicity with respect to general beliefs about causal factors (5). They noted that the concept of safety of HCQ at doses 6.5 mg/kg came from large case series and case reports. Mackenzie found no cases of HCQ toxicity among 900 patients treated with 6.5 mg/kg of HCQ for a mean of approximately 7 years (12). Bernstein reported no cases of HCQ toxicity in the world s literature between 1960 and 1989 at doses of less than 6.5 mg/kg (19). In the series of 1,207 patients by Levy et al, the definite case had received more than 6.5 mg/kg. However, Yam and Kwok report 14 cases that occurred at less than the cutoff of 6.5 mg/kg (5), and they cite up to 21 cases that have been reported elsewhere (9). Lyons and Severns (20) recently examined 62 referral cases with a high likelihood of toxicity and found no relationship between daily dosage and toxicity. Thus, the literature is divergent on the issue of dose cutoff. One reason for a relationship in earlier rather than later literature is that rheumatologists are now aware of the concern and few individuals get very high doses of HCQ that might accelerate toxicity. Current recommendations emphasize the importance of asthenic build or ideal body weight, since HCQ does not distribute in fatty tissue, but little of the published literature has ideal weight values. The classic recommendations for staying under 6.5 mg/kg were made by measured weight, and nearly 90% of our subjects used less than 6.5 mg/kg of measured weight (Figure 4). Thus, it is not surprising that daily dosage/kg was not a significant factor relative to cumulative dose. Where does this leave the prescribing physician? The 6.5 mg/kg value is no longer defensible as safe, insofar as cases occur at lower doses, and we show a low but steadily rising risk over the initial years of use independent of dose. A safe level relative to ideal body weight, if it exists, has not been determined in any large series. Fur-

8 782 Wolfe and Marmor Figure 4. Relationship between hydroxychloroquine (HCQ) daily dosage in mg and weight in kg (A) and ideal weight in kg (B). The vertical line at 61.5 kg indicates 6.5 mg/kg cutoff at 400 mg of HCQ per day. The vertical line at 30.8 kg indicates 6.5 mg/kg cutoff at 200 mg of HCQ per day. A small amount of random noise has been added to the individual points to make them visible in cases of overlap. thermore, the relationship of daily dosage to effective dose is unclear. A recent report has recommended dosing on the basis of blood levels rather than weight, because more than 1,000 ng/ml of HCQ was found to be necessary to control SLE recurrences. The blood levels in that study did not correlate with dose/kg among a population of patients receiving 400 mg/day (21). On the other hand, higher doses must accumulate faster than lower ones, and in theory, higher doses might accelerate the process of cellular toxicity. Our study showed that most current patients receive 400 mg daily irrespective of weight, and suggests that this dosage is reasonable in clinical practice insofar as the published guidelines for dose/kg do not have a proven relationship to the risk of toxicity. At the same time, it seems prudent to use low doses whenever possible, to use a reduced dose for small individuals, and to be aware of the cumulative duration and dose levels that we have found to be associated with increased risk. We hope that screening guidelines will be updated to take these new data into account. The American Academy of Ophthalmology recommendations are currently being reevaluated. The association of age with HCQ toxicity derives in part from one study of 47 patients, where none of the 9 patients younger than 60 years of age developed toxicity in contrast to 12 who were age 60 years (5,22). In our current report, 54% of all of the patients were age 60 years, as were 30% with definite or probable toxicity and 39% with possible toxicity. In this series, we did not have complete data on the age at which toxicity began (see the Patients and Methods section), so it is likely more patients than we report here were younger than 60 years of age. Bernstein considered the risk of toxicity to be almost zero for patients using HCQ for less than 10 years (19). However, there are many case reports with shorter durations of use (23). Yam and Kwok cited 47 cases with a mean duration of use of 7 years (5). They also identified nine patients who received less than the recommended dosage and developed retinopathy with therapy for less than 10 years and in three after less than 5 years of treatment. Thus, there is no question that toxicity may occur early in certain patients. However, isolated case reports give no measure of incidence, and they may appear in the literature because they are unusual and considered reportable. Our study provides statistical estimates of the probability of toxicity over time and the rate at which the toxicity risk increases with continued usage (Figures 2 and 3). The probability is very low before 5 7 years ( 0.33%). By 15 years, the rate is 2.1%. In addition, there is a suggestion in our data that there is a rapid increase toward 7 years (or a cumulative dose of 1,000 gm). We acknowledge that our estimates may be approximate, but it is hard to discount the fact that 90% of our toxic cases had 10 or more years of exposure. Our estimates are consistent with the findings of Lyons and Severns (20) from a large series of referral examinations in terms of rising risk after sufficient duration and cumulative dose. If one accepts a risk of 1% or more as sufficient to justify screening for a serious disease, then our data suggest a rationale for the commencement of regular screening after approximately 5 years of HCQ therapy. The problems of determining the extent to which HCQ is associated with retinal toxicity and the cost/benefit of screening have been reviewed by Silman and Shipley (6). Our present data show that retinal toxicity is not so rare after continued use of HCQ. And although there are no established criteria for recognizing early or reversible stages of toxicity, there are many published cases in the ophthalmology literature to document that early recognition minimizes the eventual visual loss. Bull s eye maculopathy can be annoying in terms of broken words and lost details off center, but it is not nearly as disabling as the irreversible loss of central vision that will result if the toxicity continues unchecked. While many cases of HCQ retinopathy do progress for a year or so after the cessation of therapy, because of the slow clearance of the drug, the end result usually reflects the status at the time of recognition. Early bull s eye maculopathy is typically asymptomatic, and if patients are not seen until they report visual acuity loss, the visual prognosis is poor. Screening with fundus examination and central visual field testing remains the standard for early recognition of toxic maculopathy, and will usually document pathology before patients report visual loss and at a stage where central vision will be retained if the drug is stopped. Sophisticated new techniques such as multifocal electroretinography, fundus autofluorescence photography, and ocular coherence tomography can demonstrate bull s eye damage anatomically as well as functionally, and may be of value for certain cases, although their relative sensitivity and specificity for early HCQ damage are still being evaluated (24). The 2002 guidelines of the American Academy of Ophthalmology for screening defined higher risk to include more than 5 years of use, age 60 years, and a dose above 6.5 mg/kg. By these criteria, 91.6% of current users of HCQ in our study would be in the higher risk category, which raises the question about the usefulness of such a designation. Our present data suggest that regular annual screening for all patients would have a basis in risk after approx-

9 Hydroxychloroquine Retinal Toxicity 783 imately 5 years of use, and that other factors are relative ones that should weigh into a physician s judgment as to how often or how carefully to screen. Neither age nor a dose above 6.5 mg/kg was related to toxicity in our study, but a very elderly patient or one receiving very high doses might merit more frequent screening or more extensive testing at certain visits. Present practices appear to err on the safe side, insofar as 40.4% of our patients were examined at 6-month intervals and 50.5% were examined at yearly intervals, without consideration of dose or duration of therapy. We do not have data on what drives the increased rate, although it may come from the manufacturer s legally protective recommendation of quarterly examinations or it could be driven by other economic forces. A similar nonadherence to guidelines has been found in the UK (25). Our data have many limitations. As with many other studies of HCQ toxicity, the eye specialists who examined our patients did not have a standard definition of toxicity nor a standardized way to evaluate the retina. Since HCQ toxicity is uncommon, determining predictors of toxicity can be problematic (although the large sample size and the inherent controls in the current data set have allowed us to test predictors statistically, something that was not available in almost all other studies). We had missing data, owing to the inability to get reports from all eye specialists, so that it was necessary to use MI to express the inherent uncertainty of our estimates. The confidence intervals for risk of toxicity were wide, such as 3.1 to 9.3 per 1,000 patients relative to our estimate of 6.5 per 1,000. Most other studies have not reported confidence intervals, but from the published data available these studies must also have wide intervals. HCQ retinal toxicity is an important problem, with major human and economic consequences ranging from blindness to the cost of screening. Our data show that the risk of HCQ toxicity is real, and the rate is higher than has generally been believed (1). We hope that these new estimates of risk, relative to clinical parameters such as age, dose, cumulative dose, and duration of use, will provide the tools for clinical societies to improve screening policies and guidelines and for individual physicians to make better judgments about the management of the disorder. AUTHOR CONTRIBUTIONS All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Wolfe had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study conception and design. Wolfe, Marmor. Acquisition of data. Wolfe. Analysis and interpretation of data. Wolfe, Marmor. REFERENCES 1. Marmor MF, Carr RE, Easterbrook M, Farjo AA, Mieler WF. Recommendations on screening for chloroquine and hydroxychloroquine retinopathy: a report by the American Academy of Ophthalmology. Ophthalmology 2002;109: Levy GD, Munz SJ, Paschal J, Cohen HB, Pince KJ, Peterson T. Incidence of hydroxychloroquine retinopathy in 1,207 patients in a large multicenter outpatient practice. Arthritis Rheum 1997;40: Marmor MF. New American Academy of Ophthalmology recommendations on screening for hydroxychloroquine retinopathy [letter]. Arthritis Rheum 2003;48: Saag KG, Teng GG, Patkar NM, Anuntiyo J, Finney C, Curtis JR, et al. American College of Rheumatology 2008 recommendations for the use of nonbiologic and biologic disease-modifying antirheumatic drugs in rheumatoid arthritis. Arthritis Rheum 2008;59: Yam JC, Kwok AK. Ocular toxicity of hydroxychloroquine [review]. Hong Kong Med J 2006;12: Silman A, Shipley M. Ophthalmological monitoring for hydroxychloroquine toxicity: a scientific review of available data. Br J Rheumatol 1997;36: Wolfe F, Michaud K. A brief introduction to the National Data Bank for Rheumatic Diseases. Clin Exp Rheumatol 2005;23: S Nadareishvili Z, Michaud K, Hallenbeck JM, Wolfe F. Cardiovascular, rheumatologic, and pharmacologic predictors of stroke in patients with rheumatoid arthritis: a nested, case control study. Arthritis Rheum 2008;59: Easterbrook M. Screening for antimalarial toxicity: current concepts. Can J Ophthalmol 2002;37: Royston P. Multiple imputation of missing values. Stata J 2004;4: Stata statistical software: release College Station (TX): Stata Corporation; Mackenzie AH. Dose refinements in long-term therapy of rheumatoid arthritis with antimalarials. Am J Med 1983;75: Wang C, Fortin PR, Li Y, Panaritis T, Gans M, Esdaile JM. Discontinuation of antimalarial drugs in systemic lupus erythematosus. J Rheumatol 1999;26: Morand EF, McCloud PI, Littlejohn GO. Continuation of long term treatment with hydroxychloroquine in systemic lupus erythematosus and rheumatoid arthritis. Ann Rheum Dis 1992;51: Grierson DJ. Hydroxychloroquine and visual screening in a rheumatology outpatient clinic. Ann Rheum Dis 1997;56: Mavrikakis I, Sfikakis PP, Mavrikakis E, Rougas K, Nikolaou A, Kostopoulos C, et al. The incidence of irreversible retinal toxicity in patients treated with hydroxychloroquine: a reappraisal. Ophthalmology 2003;110: Fraenkel L, Felson DT. Rheumatologists attitudes toward routine screening for hydroxychloroquine retinopathy. J Rheumatol 2001;28: American College of Rheumatology. American College of Rheumatology position paper: screening for hydroxychloroquine retinopathy URL: practice/clinical/position/hydroxyp2.pdf. 19. Bernstein HN. Ocular safety of hydroxychloroquine. Ann Ophthalmol 1991;23: Lyons JS, Severns ML. Using multifocal ERG ring ratios to detect and follow Plaquenil retinal toxicity: a review. Review of mferg ring ratios in Plaquenil toxicity. Doc Ophthalmol 2009;118: Costedoat-Chalumeau N, Amoura Z, Hulot JS, Hammoud HA, Aymard G, Cacoub P, et al. Low blood concentration of hydroxychloroquine is a marker for and predictor of disease exacerbations in patients with systemic lupus erythematosus. Arthritis Rheum 2006;54: Johnson MW, Vine AK. Hydroxychloroquine therapy in massive total doses without retinal toxicity. Am J Ophthalmol 1987;104: Warner AE. Early hydroxychloroquine macular toxicity. Arthritis Rheum 2001;44:

10 784 Wolfe and Marmor 24. Kellner S, Weinitz S, Kellner U. Spectral domain optical coherence tomography detects early stages of chloroquine retinopathy similar to multifocal electroretinography, fundus autofluorescence and near-infrared autofluorescence. Br J Ophthalmol 2009;93: Samanta A, Goh L, Bawendi A. Are evidence-based guidelines being followed for the monitoring of ocular toxicity of hydroxychloroquine? A nationwide survey of practice amongst consultant rheumatologists and implications for clinical governance. Rheumatology (Oxford) 2004;43:346 8.