Clinical images (left to right): fundus photography, spectral-domain optical coherence tomography (OCT), and 10-2 white target visual fields (pattern deviation plot and threshold plot). Levels of retinal toxicity (top to bottom): A, Normal fundus. B, Mild retinal toxicity, with distinctive parafoveal thinning of the outer retina (arrowhead) and fragments of a ring scotoma. C, Moderate retinal toxicity, with marked outer retinal thinning on both sides of the fovea (arrowheads) and a prominent ring scotoma (but still no pigmentary changes visible in the fundus). D, Severe retinal toxicity, with bull’s eye maculopathy on the fundus image, disruption of the retinal pigment epithelium on spectral-domain OCT, and severe visual field defects.
A, Receiver operating characteristic (ROC) curve of the prediction of retinal toxicity by real body weight compared with ideal body weight. The difference in the area under the ROC curves is significant (P = .03). B and C, Effect of body habitus on the rate of retinal toxicity, comparing a daily use cutoff level of 5.0 mg/kg real body weight (B) to a cutoff level of 6.5 mg/kg ideal weight (C). Body habitus is indicated by body mass index (BMI, calculated as weight in kilograms divided by height in meters squared).The lines show adjusted predictions after logistic regression analysis of BMI and retinal toxicity with 95% CIs.
A, Kaplan-Meier curves showing the cumulative risk of hydroxychloroquine retinal toxicity at 3 use levels. B, Smoothed hazard estimates showing yearly risk of toxic retinopathy at 3 use levels. C, Interaction of use and duration of use in the prevalence of retinal toxicity.
eTable. Patient Characteristics of Inclusion Versus Exclusion Groups
eFigure 1. Comparison of Dosing by Weight to Dosing by Ideal Body Weight
eFigure 2. Effect of Kidney Function on the Risk of Retinal Toxicity
Melles RB, Marmor MF. The Risk of Toxic Retinopathy in Patients on Long-term Hydroxychloroquine Therapy. JAMA Ophthalmol. 2014;132(12):1453–1460. doi:10.1001/jamaophthalmol.2014.3459
Hydroxychloroquine sulfate is widely used for the long-term treatment of autoimmune conditions but can cause irreversible toxic retinopathy. Prior estimations of risk were low but were based largely on short-term users or severe retinal toxicity (bull’s eye maculopathy). The risk may be much higher because retinopathy can be detected earlier when using more sensitive screening techniques.
To reassess the prevalence of and risk factors for hydroxychloroquine retinal toxicity and to determine dosage levels that facilitate safe use of the drug.
Design, Setting, and Participants
Retrospective case-control study in an integrated health organization of approximately 3.4 million members among 2361 patients who had used hydroxychloroquine continuously for at least 5 years according to pharmacy records and who were evaluated with visual field testing or spectral-domain optical coherence tomography.
Hydroxychloroquine use for at least 5 years.
Main Outcomes and Measures
Retinal toxicity as determined by characteristic visual field loss or retinal thinning and photoreceptor damage, as well as statistical measures of risk factors and prevalence.
Real body weight predicted risk better than ideal body weight and was used for all calculations. The overall prevalence of hydroxychloroquine retinopathy was 7.5% but varied with daily consumption (odds ratio, 5.67; 95% CI, 4.14-7.79 for >5.0 mg/kg) and with duration of use (odds ratio, 3.22; 95% CI, 2.20-4.70 for >10 years). For daily consumption of 4.0 to 5.0 mg/kg, the prevalence of retinal toxicity remained less than 2% within the first 10 years of use but rose to almost 20% after 20 years of use. Other major risk factors include kidney disease (odds ratio, 2.08; 95% CI, 1.44-3.01) and concurrent tamoxifen citrate therapy (odds ratio, 4.59; 95% CI, 2.05-10.27).
Conclusions and Relevance
These data suggest that hydroxychloroquine retinopathy is more common than previously recognized, especially at high dosages and long duration of use. While no completely safe dosage is identified from this study, daily consumption of 5.0 mg/kg of real body weight or less is associated with a low risk for up to 10 years. Knowledge of these data and risk factors should help physicians prescribe hydroxychloroquine in a manner that will minimize the likelihood of vision loss.
Hydroxychloroquine sulfate is used by physicians in many specialties for the long-term treatment of lupus erythematosus, rheumatoid arthritis, and other autoimmune conditions and is being considered for wider applications, such as the management of diabetes mellitus. While hydroxychloroquine has few systemic adverse effects, long-term use may lead to irreversible and potentially blinding retinal toxicity.1 This adverse effect has been considered rare (estimated occurrence in 0.5%-2.0% of long-term users2- 4), but the risk may in fact be considerably greater. Most existing data about the prevalence of hydroxychloroquine retinal toxicity come from studies2,3 based primarily on short duration of use and on diagnosis at an advanced stage of visible retinal damage. However, retinopathy can be detected much earlier by central visual field testing and modern techniques, such as spectral-domain optical coherence imaging (SD-OCT).5
Because hydroxychloroquine distributes poorly in fatty tissues,6 it was suggested earlier7,8 (and reinforced by current American Academy of Ophthalmology screening guidelines9) that dosage should be calculated by ideal body weight to reduce the theoretical risk of overdosing obese patients. A daily dose of 6.5 mg/kg of ideal body weight has been generally recommended based on studies7,8 that found few cases of severe retinal toxicity below 6.5 mg/kg of real body weight, but this value has never been evaluated to our knowledge.
We studied a large population of long-term hydroxychloroquine users whose retina was evaluated with sensitive diagnostic techniques. The data suggest that hydroxychloroquine retinopathy is not rare and that alternative dosing criteria and awareness of specific risk factors may enhance the safe use of this drug.
Kaiser Permanente Northern California (KPNC) is an integrated health organization with a diverse population of approximately 3.4 million members. The organization has used electronic medical records for more than 20 years, and digital ophthalmic images have been reviewable on all patients since 2009. After KPNC institutional review board approval, we queried the pharmacy database for patients (n = 3482) taking hydroxychloroquine as of January 1, 2009, for a minimum of 5 continuous years, with a maximum gap in therapy of less than 1 year. Because we are reporting only on aggregate data obtained by medical record reviews, informed consent from individual patients was not deemed necessary by the institutional review board.
Inclusion criteria were a reliable central visual field examination or SD-OCT (2361 patients [67.8%]), techniques that can demonstrate retinopathy before any visible fundus change. Excluded were 654 patients (18.8%) who were screened only by fundus examination (photography or ophthalmoscopy) and 348 patients (10.0%) who had no evidence of screening. Also excluded were 119 patients (3.4%) with prior chloroquine use or significant retinal comorbidity, such as macular degeneration or diabetic retinopathy. Demographic characteristics of the included and excluded populations were similar (eTable in the Supplement).
A sequence of findings with visual field testing and SD-OCT, from mild to severe, is shown in Figure 1. Fields could use white or red targets,10 and all were performed using standard equipment (Humphrey perimeter; Carl Zeiss Meditec). The SD-OCT recordings were performed with a standard instrument (Spectralis; Heidelberg Engineering). Retinal toxicity was judged by characteristic damage on visual field testing or SD-OCT (Figure 1) and was confirmed to be unequivocal by both of us. For visual field testing, toxicity meant partial or full ring scotomas mainly involving the parafoveal region.10,11 For SD-OCT, this meant predominantly parafoveal thinning of the outer retina and loss of photoreceptor outer segment marker lines (ellipsoid zone and interdigitation zone).5,12,13
Most patients (2020 [85.6%]) had started taking hydroxychloroquine after the computerized pharmacy system was implemented, and their medication use was calculated from the number of tablets dispensed. The remainder (341 [14.4%]) started taking hydroxychloroquine a mean of 4.5 years before joining KPNC or before implementation of the pharmacy system. Their use for the additional years was calculated from prescribed amounts and was adjusted by their mean compliance rate when tracked by the pharmacy database. Throughout this article, dosage is expressed as use (ie, consumption rather than prescribed dosage) relative to real body weight unless explicitly stated as ideal body weight. Ideal body weight was calculated using a simplified formula: for women, 100 lb for the first 5 ft of height, plus 5 lb for every inch of height over 5 ft; and for men, 110 lb for the first 5 ft of height, plus 5 lb for every inch of height over 5 ft (for women, 45 kg for the first 1.5 m of height, plus 2.3 kg for each 2.5 cm of height over 1.5 m; and for men, 50 kg for the first 1.5 m of height, plus 2.3 kg for each 2.5 cm of height over 1.5 m).14
We estimated the mean glomerular filtration rate (GFR) during the course of hydroxychloroquine therapy using the 4-variable Modification of Diet in Renal Disease equation: GFR = 175 × Serum Creatinine Level−1.154 × Age−0.203 × (0.742 if Female) × (1.210 if Black).15 A patient was considered to have kidney disease from a notation of stage 3, 4, or 5 disease on their problem list or if their mean GFR was less than 60 mL/min per 1.73 m2. Significant liver disease was determined by a diagnosis of chronic hepatitis in the patient’s medical record or by the mean liver enzyme levels (aspartate aminotransferase and alanine aminotransferase) being more than twice the normal upper limit.
Comparisons between groups were performed using t test for continuous measures and χ2 test for categorical measures, and all reported probability values are 2-sided. Odds ratios were derived using logistic regression analysis.
Our results show that the prevalence of hydroxychloroquine retinopathy is much higher than previously recognized and depends on risk factors such as daily dose, duration of use, and kidney disease. The results also suggest the need to revise the way that dosage is calculated to minimize risk. We identified a new risk factor of concurrent tamoxifen citrate use. These data have important implications for medical and ophthalmologic practice to maximize the availability of hydroxychloroquine to patients while avoiding retinal toxicity.
Table 1 lists the clinical characteristics of our patient population. Of 2361 patients who had taken hydroxychloroquine continuously for at least 5 years and who had 10-2 visual fields or SD-OCT, 177 (7.5%) showed clear signs of retinal toxicity (Figure 1). None of the primary medical indications for hydroxychloroquine therapy were significantly associated with an increased prevalence of retinal toxicity. A multivariable logistic regression analysis was performed (Table 1) on those factors that showed significant differences by univariate analysis, along with age (which has been postulated to increase risk). Daily use, duration of use, concurrent tamoxifen therapy, kidney disease, and lower weight were correlated with retinal toxicity, but age and sex were not. Additional univariate logistic regression analysis showed the effect of factors that have frequently been used to estimate the risk of retinal toxicity for individual patients (Table 2). Of 177 patients diagnosed as having retinal toxicity, 98 had color fundus photographs available for review, and only 31 (31.6%) showed a visible bull’s eye depigmentation (Figure 1, bottom), confirming the higher sensitivity of our screening techniques.
We calculated receiver operating characteristic curves (Figure 2A) to assess the sensitivity and specificity of real vs ideal body weight in predicting retinal toxicity and found that real body weight is a better predictor of retinal toxicity (receiver operating characteristic curve area, 0.78 for real body weight vs 0.75 for ideal body weight; P = .03). Because current dosing recommendations advise 6.5 mg/kg of ideal body weight,9 it is important to have a comparable value in real body weight to aid in the interpretation of population data. If one looks at the distribution of patients comparing use by real body weight with use by ideal body weight (eFigure 1 in the Supplement), 6.5 mg/kg of ideal body weight corresponds approximately to 5.0 mg/kg of real body weight along the regression line (patients were typically approximately 25%-30% heavier than ideal body weight). This value is also realistic in terms of rheumatologic practice because most of our patients (1828 of 2361 [77.4%]) were in fact consuming less than 5.0 mg/kg of real body weight. Furthermore, the prevalence of retinal toxicity relative to real body weight is essentially independent of body habitus (Figure 2B), whereas the risk is much higher in thin individuals using ideal body weight (Figure 2C). Because of these findings, we have used real body weight for all subsequent presentations in this article and 5.0 mg/kg of real body weight as a division between judicious and excessive use.
Kaplan-Meier curves in Figure 3A show the cumulative risk in the population for 3 different ranges of dosage per kilogram. Patients with a mean daily use exceeding 5.0 mg/kg had approximately a 10% risk of retinal toxicity within 10 years of treatment and an almost 40% risk after 20 years. Patients using an intermediate amount of 4.0 to 5.0 mg/kg had risk of less than 2% within the first 10 years of use but almost 20% risk after 20 years. These medication use values are based on pharmacy dispensing information and were on average approximately 20% lower than the prescribed dosage because of variable patient compliance.
The smoothed hazard estimates in Figure 3B show the incremental risk of retinal toxicity (annual risk) that a patient without retinal toxicity faces in each ensuing year. For use of 5.0 mg/kg or less, this annual risk is less than 1% in the first decade of use but rises to almost 4% after 20 years and is 2 to 3 times higher at use exceeding 5.0 mg/kg.
Figure 3C shows more directly the continuous interaction of use and duration of use in determining the prevalence of retinal toxicity. No dosage is completely safe, but regulation of either factor will greatly reduce the risk from the other.
The kidneys are the main mechanism for clearance of hydroxychloroquine,8 and decreased renal function leads to higher serum concentration.16 Kidney disease markedly increases the risk of retinal toxicity (Table 2), and eFigure 2 in the Supplement shows the relationship between retinal toxicity and the GFR. A drop in kidney function by 50% leads to an approximate doubling of the risk of retinopathy. Although hydroxychloroquine is partially cleared by the hepatic system,7 we found no increase in the risk of retinal toxicity from liver disease.
Patients who had concurrent tamoxifen therapy for breast cancer were at greatly increased risk of the development of retinal toxicity (Table 2), and retinal toxicity correlated with greater cumulative tamoxifen intake (P = .03). In contrast, patients treated for estrogen receptor–positive breast cancer with concurrent anastrozole did not appear to have a similar increase in risk. None of the patients who took hydroxychloroquine and tamoxifen concurrently showed crystalline deposits or macular edema that is characteristic of tamoxifen retinopathy,17 but they all had parafoveal outer retinal damage and were classified as having hydroxychloroquine retinopathy.
We found that 7.5% of long-term hydroxychloroquine users screened with modern techniques showed evidence of retinal toxicity. This prevalence is approximately 3 times higher than previously reported,2- 4 but the risk to an individual depends on dosage and duration of use. Prior studies2,3 included patients with shorter duration of use and depended mainly on the development of bull’s eye maculopathy to detect retinal toxicity. In contrast, most of our patients diagnosed as having retinal toxicity were detected before bull’s eye maculopathy was visible.
A limitation of our study is that approximately 30% of long-term hydroxychloroquine users initially identified were excluded because of a lack of sensitive screening studies or for comorbid retinopathy. However, the excluded group showed demographic characteristics similar to those of the included group (eTable in the Supplement), and we do not believe a major difference would existin their prevalence of retinopathy.
Our data show that, although no hydroxychloroquine use is completely safe, the risk of retinal toxicity can be kept low with careful dosage adjustment and with shorter periods of use. However, the risk rises markedly with concurrent kidney disease, and the prevalence can exceed 50% with use above 5.0 mg/kg and with duration beyond 20 years.
We propose the use of real body weight for dosage calculation because it correlates better with retinal toxicity than ideal body weight and allows estimations of risk that are independent of body habitus. A recent study18 found that serum levels of hydroxychloroquine also correlate better with real body weight. The empirical use limit of 5.0 mg/kg of real body weight that we suggest should be sufficiently high to provide medical relief for most patients because it is in fact equivalent to the current dosing recommendation of 6.5 mg/kg of ideal body weight for patients of ordinary habitus9 (eFigure 1 in the Supplement), and most of our patients were being maintained medically on less than 5.0 mg/kg of hydroxychloroquine sulfate. It is important to emphasize that our data represent pills dispensed, and many of our patients who were prescribed at a dosage of 6.5 mg/kg of ideal body weight were using a lower amount because of imperfect compliance. Dosing by real body weight will be simpler to calculate than by ideal body weight, and the main effect clinically will be a reduction of dosage for thin patients (many of whom may be receiving high dosages by the criteria of this study). However, it will be incumbent on physicians to consider compliance in the prescription of this drug. Maintaining daily use at 5.0 mg/kg or less would keep both the cumulative risk and annual risk of retinal toxicity low, especially for the first 10 years of use. Because hydroxychloroquine takes several months to reach stable blood levels,19 dosing can be adjusted to weight by omitting or splitting tablets on certain days of the week. In theory, it would be ideal if hydroxychloroquine dosing could be guided by blood levels, but studies19,20 have found wide variations, and the results suggest that blood concentrations in an individual do not correlate closely with dosage, weight, or clinical effectiveness. However, blood levels may aid in judging noncompliance or the effects of kidney disease.16,18,21
Our prevalence data apply to the overall population of long-term hydroxychloroquine users, and risk rises markedly after 10 years of use. However, in rheumatologic practices, many patients benefit from the use of the drug for much longer periods, and it is important to know the annual risk as they stay on the drug regimen. The smoothed hazard estimates (Figure 3B) show that a patient who shows no signs of retinal toxicity at a given point in time and is not overdosed will have a risk of developing retinal toxicity during the ensuing year of approximately 1% after 10 years of use and an annual risk of less than 4% after 20 years of use. These data should serve to reassure medical specialists that the drug can be prescribed safely for extended periods with understanding of the ocular risks and with effective screening.
If we extrapolate the use of long-term hydroxychloroquine at KPNC to the entire US population, approximately 350 000 patients should receive annual eye screening by current guidelines.9 Therefore, screening for hydroxychloroquine retinal toxicity is an important economic and patient safety issue. Retinal toxicity from hydroxychloroquine use cannot be completely prevented, but effective screening should recognize retinal toxicity before symptoms or significant risk of central visual field loss appear (ie, before the appearance of bull’s eye maculopathy). Screening requires the use of tests, such as 10-2 visual fields and SD-OCT (and other modern techniques, such as autofluorescence imaging and multifocal electroretinography22,23), to demonstrate early retinal damage. We have shown that 10-2 visual fields are sometimes more sensitive than SD-OCT in revealing retinal toxicity.24 However, SD-OCT is more specific and objective, and we suggest the use of both tests when available. With effective screening, retinal toxicity can be recognized at an early stage when patients are typically asymptomatic and disease is unlikely to progress.25
Our results confirm the basic principles of screening recommended by the American Academy of Ophthalmology.9 However, we propose the use of real body weight rather than ideal body weight to calculate daily dose, along with possible adjustment for patient compliance. Our data show that age is not a risk factor, but our findings emphasize the importance of kidney disease, which raises the effective blood level of hydroxychloroquine.16 Unexpectedly, we also found a strong relationship between retinal toxicity and tamoxifen use. Tamoxifen is a retinal toxin in its own right, although most cases of tamoxifen retinopathy were reported early in the history of the drug when higher dosages were prescribed.26 Our data indicate that chronic low-dosage administration of tamoxifen has an adverse synergism with hydroxychloroquine, and the effect is related to the cumulative dose.
These data suggest that hydroxychloroquine retinopathy is more common than previously recognized, especially at high daily intake and with long durations of use or in the presence of kidney disease or concurrent tamoxifen therapy. Daily use of 5.0 mg/kg of real body weight or less is associated with a low risk for up to 10 years of use. We anticipate that these data will help physicians develop prescribing patterns that maintain patients on this valuable medication while minimizing the risk of retinal toxicity.
Submitted for Publication: May 1, 2014; final revision received July 11, 2014; accepted July 12, 2014.
Corresponding Author: Ronald B. Melles, MD, Kaiser Permanente, Redwood City Medical Center, 1150 Veterans Blvd, Redwood City, CA 94063 (firstname.lastname@example.org).
Published Online: October 2, 2014. doi:10.1001/jamaophthalmol.2014.3459.
Author Contributions: Dr Melles had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: All authors.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: All authors.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: All authors.
Administrative, technical, or material support: Marmor.
Study supervision: Melles.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.
Additional Contributions: Eric Jorgenson, PhD, Division of Research, Kaiser Permanente Northern California, and Frederick Wolfe, MD, National Data Bank for Rheumatic Diseases, gave advice regarding statistical methods.
Correction: This article was corrected on October 30, 2014, to fix an error in Table 2.