Original research

New proposal for a multimodal imaging approach for the subclinical detection of hydroxychloroquine-induced retinal toxicity in patients with systemic lupus erythematosus

Abstract

Objective To compare multimodal structural and functional diagnostic methods in patients with systemic lupus erythematosus (SLE) treated with hydroxychloroquine, to identify the best complementary approach for detecting subclinical retinal toxicity.

Methods A cross-sectional, unicentric study was conducted on patients with SLE treated with hydroxychloroquine. Each patient underwent a comprehensive ophthalmic evaluation, comprising structural tests (spectral-domain optical coherence tomography (SD-OCT), en face OCT, en face OCT angiography (OCTA), fundus autofluorescence (FAF)) and functional tests (automated perimetry for visual field (VF) testing, multifocal electroretinography (mfERG)). A diagnosis of macular toxicity required the presence of abnormalities in at least one structural and functional test. The Kappa Concordance Index was used to assess the concordance among the different tests in detecting potential macular toxicity-associated alterations.

Results Sixty-six patients with SLE (132 eyes) were consecutively enrolled. Four (6.1%) patients developed subclinical hydroxychloroquine-induced retinal toxicity without visual acuity impairment. The proportion of abnormal results was 24% for both en face OCT and en face OCTA. Regarding functional analysis, VF was less specific than mfERG in detecting subclinical retinal toxicity (VF specificity 47.5%). En face OCT and en face OCTA structural findings showed better concordance, with a kappa index >0.8, and both identified the same cases of toxicity as FAF.

Conclusion Although structural OCT and VF are frequently used to screen for hydroxychloroquine-induced retinal toxicity, our findings suggest that a combination of mfERG, en face OCT and en face OCTA could improve the diagnostic accuracy for subclinical retinal damage. This study emphasises the importance of a multimodal imaging strategy to promptly detect signs of hydroxychloroquine-induced retinal toxicity.

What is already known on this topic

  • Structural optical coherence tomography (OCT) and visual field (VF) are the most commonly used tests for screening hydroxychloroquine-induced retinopathy.

What this study adds

  • New retinal imaging techniques (en face OCT/en face OCT angiography) show potential for detecting retinopathy from a single image.

How this study might affect research, practice or policy

  • This study provides valuable insights into multimodal imaging of subclinical hydroxychloroquine-induced retinal toxicity.

Introduction

Hydroxychloroquine is the most widely used drug for the treatment of systemic lupus erythematosus (SLE) because of its safety profile and survival benefits.1–4 Retinal toxicity is a known potential adverse effect of hydroxychloroquine, which can increase with longer exposure.3 5 6 Early detection of retinal toxicity features is essential to prevent irreversible macular damage.7

En face optical coherence tomography (en face OCT) is a novel retinal imaging technique that has demonstrated the ability to detect reflectivity changes in cases of photoreceptor defects and assess the extent of retinopathy in a single image.8 In addition, OCT angiography (OCTA) can detect decreased vascular density in the capillary plexus layer and has been proposed as a potential additional parameter of retinal toxicity.9 These emerging techniques offer superior visualisation of the macular region, enabling identification of subclinical disease and detection of possible toxic alteration before irreversible damage occurs.

The American Academy of Ophthalmology (AAO) and the United Kingdom Royal College of Ophthalmologists (RCO) have established guidelines for the screening of hydroxychloroquine retinal toxicity, which have evolved to include objective measures to improve early detection,4 10–12 but they have not yet incorporated these new techniques.10 13

Thus, the aim of the present study was to compare the quantitative analysis of the retina by en face OCT and en face OCTA with the tests currently recommended by the AAO and the RCO for the screening of hydroxychloroquine retinopathy and to investigate the sensitivity and specificity of different structural and functional tests in detecting early hydroxychloroquine retinopathy in patients with SLE.

Materials and methods

Study design and patient selection

Consecutive patients undergoing hydroxychloroquine treatment who attended the Department of Ophthalmology between December 2018 and March 2020 were included. All patients fulfilled four or more of the updated 1997 American College of Rheumatology (ACR) classification criteria for SLE14 and were aged 18 years or older. Patients with previous maculopathy, glaucoma or high-grade media opacities were excluded.

All patients provided written informed consent before participating in the study. This study meets all five of the CODE-EHR minimum framework standards for the use of structured healthcare data in clinical research.

Clinical and demographic data

Demographic variables, such as sex, ancestry and age at disease diagnosis, were reported. Clinical and immunological parameters were determined according to the European League EULAR 2019 SLE classification criteria.15 Significant comorbidities, patient factors that increase the likelihood of hydroxychloroquine toxicity3 10 13 16 and medication details during the inclusion period were recorded. Antimalarial encompassed the type of drug, dose, adverse effects, start and stop dates, and reasons for discontinuation. Chronic kidney disease (CKD) was considered if any of the CKD stages 3–5 were present and defined according to previous studies and guidelines.5 17

The SLE Disease Activity Index was measured using the Safety of Estrogens in Lupus Erythematosus National Assessment-SLEDAI (SELENA-SLEDAI-2K) and clinical SLEDAI-2K (considering only clinical variables from SLEDAI-2K).18 Cumulative damage was measured using the SLICC/ACR Damage Index for SLE (SDI).19

Patients were categorised according to the duration of hydroxychloroquine treatment as follows: (1) <5 years and (2) ≥5 years.

Ophthalmic examinations

Each patient underwent a detailed ophthalmic examination of both eyes, including measurement of best-corrected visual acuity on the Snellen scale, slit-lamp biomicroscopy, intraocular pressure measurement by Goldman applanation tonometry and fundus examination under pharmacological mydriasis. All patients also underwent the following structural and functional ophthalmic tests and were analysed for the characteristics mentioned in the following tests:

Structural tests

All spectral-domain optical coherence tomography (SD-OCT) images were obtained using a Zeiss CIRRUS 5000 device (Carl Zeiss Meditec, Dublin, California, USA).

Qualitative tests
  1. Cross-sectional SD-OCT B-scan. The macular cube scan was obtained using a 512×12 raster scan pattern covering a 6×6 mm area and a B-scan 5-HD line programme centred on the fovea to detect imaging features, including the presence of moth-eaten photoreceptor inner/outer segment appearance, perifoveal ellipsoid zone loss, defects in the outer nuclear layer or retinal pigment epithelium resembling a flying saucer sign.20

  2. C-scan en face SD-OCT and en face OCTA of the macular area were analysed to detect overall alterations in the ellipsoid zone. The ellipsoid layer was considered abnormal if diffuse or localised areas of uneven hyporeflectivity were detected and if they corresponded with thinning or changes in the ellipsoid parafoveal zone on the cross-sectional SD-OCT B-scan.21

  3. Fundus autofluorescence (FAF) was performed using an OPTOMAP 200 device (Optos Panoramic Retinal Exam, Dunfermline, Scotland, UK). Images were classified as normal or abnormal areas based on autofluorescence (perifoveal hypoautofluorescence, hypoautofluorescent mottled appearance surrounded by hyperautofluorescent zones or hyperautofluorescent pericentral ring).22

Quantitative tests
  1. SD-OCT. Macular thickness parameters were measured by acquiring a series of 128 horizontal scan lines, each consisting of 512 A-scans. The optical nerve fibre layer thickness was measured using a 6×6 mm cube with 200 horizontal scan lines, each consisting of 200 A-scans. The choroidal thickness was measured manually using a 5HD-OCT programme with an enhanced depth imaging display.

  2. OCTA macular cubes were obtained using a 6×6 mm Angioplex scan protocol. OCTA parameters assessed included vessel density (mm/mm2), vascular perfusion (%) and foveal avascular zone, which included area, perimeter and circularity, and were calculated using built-in software.

Functional tests

  1. Standard Humphrey 10-2 automated perimetry (VF) was performed using a Humphrey 10.2 standard perimeter (Carl Zeiss Meditec) measuring the temporal and nasal 10° at a total of 68 points separated by 2°. The test was repeated if the initial results had low reliability because of fixation loss, false positives or negatives. Acceptable thresholds included fixation loss of less than 20%, false-positive errors of less than 20% and false-negative errors of less than 20%. The VF was considered abnormal if a partial or complete ring defect was found between 2° and 6° with central sparing.23

  2. Multifocal electroretinography (mfERG) was recorded using wire electrodes (Hansen Ophthalmic Laboratories, Iowa City, Iowa, USA) and the RETIscan system (Roland Consult, Wiesbaden, Germany) according to the guidelines of the International Society for Clinical Electrophysiology of Vision (ISCEV) for visual electrodiagnostic procedures.24 The mfERG data collected included the pseudoresponse amplitude of the second (R2) and third (R3) sector means, the morphology of the R2 and R3 waves, and the ratios of R2 to the fifth sector (R2/R5) and R3 to the fifth sector (R3/R5).25 26

Diagnosis of macular toxicity

Macular toxicity was diagnosed according to current guidelines,10 13 which require participants to have changes in at least one structural test and one functional test. Macular toxicity was confirmed by an altered mfERG, which included a reduction in central ring amplitude in R2 or R3, a depression of the P1 wave, and an R2/R5 or R3/R5 ratio of less than 1, among low amplitudes of the perifoveal ring.27

Statistical analysis

Absolute and relative frequencies were used to describe categorical variables. Quantitative variables were described using the median and IQR. Fisher’s exact test or Mann-Whitney U test was used to compare defined groups based on the time of exposure to hydroxychloroquine and to compare the R2/R5 and R3/R5 ratios of mfERG analyses relative to en face OCT or en face OCTA stratified by clinical outcome (normal or altered). The concordance between the different tests, in terms of detecting possible changes, was evaluated using the Kappa Concordance Index, with a value greater than 0.8 considered to be near perfect concordance.28 A two-sided type I error of 0.05 was used to determine statistical significance. SPSS V.27 (IBM) was used for all statistical analyses.

Results

Baseline demographics and disease characteristics

Sixty-six patients (132 eyes) were included in the study (online supplemental figure S1). The baseline characteristics of the entire series are summarised in table 1.

Table 1
|
Cohort characteristics

No statistically significant differences were observed between patients with different hydroxychloroquine treatment durations in terms of age, sex, ethnicity, comorbidities, laboratory parameters, previous treatments and ophthalmological results. Compared with patients treated with hydroxychloroquine for ≥5 years, those treated for a shorter period exhibited a higher prevalence of antiphospholipid syndrome (p=0.022).

Hydroxychloroquine was administered at a median daily dose of 200 mg to all 66 patients. The median (IQR) duration of hydroxychloroquine treatment was 5 (2; 11) years, the median (IQR) cumulative dose of hydroxychloroquine was 396 (144; 792) g, and the median (IQR) dose per kilogram of real body weight was 3.5 (2.96; 4) mg/kg/day. Ten (15.15%) patients exceeded the risk dose of 1000 g.

Some patient factors previously implicated in increasing the likelihood of hydroxychloroquine toxicity were not significantly different between the two groups. None of the patients had a previous retinal disease or were receiving tamoxifen, the median (IQR) body mass index was 22.9 (21.6; 24.2) kg/m2, and 6 (9.1%) patients had CKD. The SLEDAI-2K was <4 in 42 (63.3%) patients, clinical SLEDAI was <4 in 59 (89.4%) patients and SDI was >0 in 32 (48.5%) patients.

Characteristics of patients with hydroxychloroquine retinal toxicity

Four patients (6.1 %) had hydroxychloroquine retinal toxicity (online supplemental table S1). All patients were diagnosed in the subclinical phase, as none reported any subjective complaints. After discontinuing hydroxychloroquine in patients with abnormal mfERGs, all patients had normal mfERG results 6 months later.

Ophthalmological findings

The median (IQR) visual acuity and intraocular pressure of the included eyes and the median spherical equivalent are shown in table 1.

Structural tests

Qualitative test
  1. Cross-sectional SD-OCT B-scan. All patients with hydroxychloroquine retinal toxicity showed localised defects involving the ellipsoid and outer neurosensory retinal layers in the juxtafoveal region, which corresponded topographically with the en face OCT and OCTA image alterations (figure 1).

  2. En face OCT and en face OCTA. We found a higher rate of pathological en face OCTA (20% in RE and 15.4% in LE) than en face OCT (16.7% in both eyes) (online supplemental table S2). There were no statistically significant differences between patients with different durations of hydroxychloroquine treatment (online supplemental table S3). The abnormal structural SD-OCT, en face SD-OCT and en face OCTA findings are shown in figure 2.

  3. FAF. Only a small proportion of the patients (7.9%) had FAF changes (online supplemental table S3). Of these, only those diagnosed with hydroxychloroquine-induced retinopathy showed changes in FAF in both eyes (figure 2). The results of this additional test were not statistically significant between patients with different hydroxychloroquine treatment durations.

Figure 1
Figure 1

Multimodal imaging using structural and en face spectral-domain optical coherence tomography. High-definition sagittal sections of the macular region show localised involvement of the ellipsoid and outer neurosensory retinal layers in the juxtafoveal region (asterisks) (A and E). Reflective infrared en face images (B and H) show changes due to thinning of the outer layers of the neurosensory retina (arrows and double asterisks) and retinal pigment epithelium. The en face section at the level of the ellipsoid (C and G) shows concentric alterations (donut-like; arrows) that prevent the foveal area. The en face section, at the level of the retinal pigment epithelium-choriocapillaris (D and F), shows the involvement of the outermost layers (arrows). Right eye (A–D), left eye (E–H).

Figure 2
Figure 2

Multimodal image from retinography and structural spectral-domain optical coherence tomography. Colour retinography (A and E) showed very mild retinal pigment epithelium alterations, which may have been more apparent under near-infrared retinography (B and F), which showed concentric involvement of the juxtafoveal in the macular area (donut-like; arrows). Fundus autofluorescence retinography (C and G) shows concentric hypoautofluorescence and hyperautofluorescence changes characteristic of retinal pigment epithelium involvement (arrows). High-definition sagittal sections (D and H) of the macular region show localised involvement of the ellipsoid and outer neurosensory retinal layers in the parafoveal region (asterisks). Right eye (A–D), left eye (E–H).

Quantitative tests

Quantitative data from SD-OCT and OCTA are shown in online supplemental table S4. None of the morphological parameters showed statistically significant differences between the treatment duration groups.

Functional tests

  1. VF. Abnormal VF results were observed in 54.5% of the patients (online supplemental table S2), with no statistically significant differences between subjects treated for 5 years (47.1%) and those treated for longer periods (62.5%).

  2. mfERG. Quantitative mfERG data for both eyes are shown in online supplemental table S5. Only 6.2% of the mfERG recordings were abnormal (online supplemental table S3), and the results did not differ with respect to the time for hydroxychloroquine use. Of the four patients with abnormal mfERGs, three showed a decrease in R2 response amplitude and one showed morphological changes in the bilateral perifoveal ring, indicating susceptibility to maculopathy.

Ten patients, representing 15.2% of the total, had normal results on standard techniques (SD-OCT and VF) but abnormal results on en face OCT and en face OCTA.

Assessing the accuracy of ophthalmic tests in detecting subclinical retinal toxicity

The sensitivity of both en face OCT and OCTA in detecting ocular abnormalities was 100%, and their specificity were 80.3% and 80%, respectively. The sensitivity and specificity of FAF were 100% and 98.3%, respectively. Although the sensitivity for detecting subclinical retinal toxicity of VF was 100%, its specificity was lower (47.5%).

A high degree of agreement was found between mfERG and en face OCT (86.9%) and between mfERG and en face OCTA (85.9%) (online supplemental table S6).

Online supplemental table S7 provides information on the concordance between different tests for detecting changes in en face OCT, en face OCTA, FAF, VF and mfERG. The highest agreement was observed for the combination of mfERG and FAF (98.4%), with a kappa index of 0.880 (p<0.001), and the combination of en face OCT and en face OCTA (95.4%), with a kappa index of 0.873 (p<0.001). In contrast, the lowest agreement was observed for the combination of mfERG and VF (50%), with a kappa index of 0.099 (p=0.069).

Discussion

In this cross-sectional, consecutive, unicentric study, we performed a qualitative analysis comparing the conventional structural (SD-OCT and FAF) and functional (VF and mfERG) tests with the addition of novel techniques (en face OCT and en face OCTA) to evaluate the best complementary approach to detect subclinical retinal toxicity. Unlike most studies analysing tests to detect hydroxychloroquine-induced retinal toxicity in patients with SLE,29 30 our study focused on individuals who were less likely to experience retinal toxicity, as most patients (84.85%) had a cumulative dose of hydroxychloroquine of less than 1000 g. In our study, the incidence of hydroxychloroquine-induced retinal toxicity was 3.88 cases per 100 persons per year.

Our series of patients with hydroxychloroquine retinopathy showed changes in both structural and functional tests. All patients with hydroxychloroquine retinal toxicity had perifoveal ellipsoid zone defects, as seen on cross-sectional OCT B-scan images, with uneven reflectivity on en face OCT and OCTA, which was confirmed with mfERG alterations.

In our study, the structural tests, en face OCT and en face OCTA, showed a high sensitivity for detecting macular photoreceptor layer alterations, which were confirmed or rejected by mfERG and FAF, which proved to be more specific. On the other hand, the AAO recommends FAF as a structural and objective test for the screening of hydroxychloroquine retinopathy, especially in cases of reduced VF, as it may provide additional information for the diagnosis and monitoring of hydroxychloroquine retinopathy.10 Throughout our study, FAF results changed simultaneously with mfERG, but did not show more sensitivity than en face OCT and en face OCTA.

Regarding quantitative macular measurements, we have recently reported the differences between patients with SLE and controls in SD-OCT and OCTA parameters.9 Patients with SLE showed significant thinning of the central macular thickness and retinal nerve fibre layer thickness compared with healthy controls. Furthermore, patients with SLE had significantly decreased vascular density and vascular perfusion when compared with controls, while the area of the foveal avascular zone did not differ between them.9 In the present analysis, all quantitative results from SD-OCT and OCTA showed no statistically significant differences between the two groups of hydroxychloroquine duration for any of the parameters analysed. However, three patients with hydroxychloroquine retinopathy had macular thickness measurements that were lower than the mean for the entire series. Because macular thickness can vary widely between individuals and can be influenced by several factors,31 further research is needed to assess the sensitivity of SD-OCT compared with VF or mfERG, as prominent changes are sometimes observed on SD-OCT even before VF loss is detected.11 32

Regarding functional tests, although many patients with SLE treated with hydroxychloroquine showed changes in VF, the specificity of VF for detecting early retinal toxicity in our study was only 47.5%, making it an unreliable standalone test for detecting macular toxicity. Patient cooperation and response and the need to repeat the test for reproducibility further limit its usefulness in detecting macular toxicity.33

Although mfERG is considered the gold standard for the diagnosis of macular toxicity,25 27 the diagnosis of macular toxicity should be made in combination with altered structural tests; nevertheless, the mfERG is very helpful in confirming hydroxychloroquine toxicity in controversial cases. The high concordance of mfERG with the other tests was significant, especially with en face OCT, en face OCTA and FAF, as 100% of the cases diagnosed by mfERG had abnormalities in the other tests. Therefore, a multimodal approach combining several objective tests, including mfERG, and structural tests, such as en face OCT and en face OCTA, may be helpful in detecting subclinical macular toxicity in patients with SLE receiving hydroxychloroquine (figure 3).

Figure 3
Figure 3

Multimodal assessment of hydroxychloroquine-induced retinal toxicity. Spectral-domain optical coherence tomography (OCT) en face sections (en face OCT) at ellipsoid level showed central and parafoveal changes in both eyes, with typical ‘bull’s eye’ macular appearance. Fundus autofluorescence (FAF) revealed a hypoautofluorescent semi-circular lesion in the temporal parafoveal region in both eyes. Functional perimetry with a 10.2 central visual field (VF 10.2) revealed localised paracentral scotomas (RE) and diffuse peripheral sensitivity loss (RE and LE), but with preserved central function. The multifocal electroretinogram (mERG) was the most sensitive approach, showing a diffuse decrease in P1-wave amplitudes, more evident in the peripheral isoptera of both eyes (blue hexagons), and more pronounced in the LE. LE, left eye; RE, right eye.

Definitions of hydroxychloroquine-induced retinal toxicity vary due to the use of different diagnostic techniques such as OCT and FAF, and the lack of objective assessment tools in previous studies. Modern imaging allows for more objective classification of retinopathy by severity and pattern, but variations in the interpretation of these tests still contribute to inconsistencies in toxicity definitions.34 Allahdina et al proposed a new staging system based on OCT findings: stage 1 includes minor changes limited to the parafoveal region; stage 2 includes clear localised changes in the parafoveal region; stage 3 includes extensive parafoveal changes; and stage 4 includes foveal involvement. While this system is primarily applicable to eyes with parafoveal retinopathy, these stages have been functionally related to visual acuity, visual field test results or mfERG and are significantly associated with the degree of further progression.35 In our study, all patients were diagnosed at stage 1, leading to improvements in CV and reversals in mfERG findings during follow-up.

In line with our findings, a recent study compared the use of en face OCT and mfERG for screening hydroxychloroquine retinopathy. The study shows that although there is some agreement between the two methods, they cannot replace each other as they assess different retinal changes. The results suggest that en face OCT could serve as an additional tool in screening, but requires confirmation by OCT B-scan for accurate diagnosis.29

The current study has several limitations, including a small number of patients with hydroxychloroquine toxicity and limited power owing to the low incidence of retinal toxicity. We may have overestimated true adherence to hydroxychloroquine given the absence of a reliable technique to measure blood levels in patients with SLE. In addition, the lack of a control group did not ensure that some abnormalities observed could also occur in healthy individuals. One of the main strengths of this study was the inclusion of only incidental cases diagnosed within the 15-month study period, which allows the comparison of different diagnostic procedures, in contrast to other reports that included prevalent cases. Finally, despite the modest size of the study, the en face OCT and en face OCTA techniques allowed the acquisition of the largest unicentric collection of structural and microvascular retinal data in patients with SLE to date, as we have recently published.9

In conclusion, this study highlights the importance of using a combination of functional and structural retinal assessment tools to detect the subclinical features of hydroxychloroquine-induced maculopathy. In our study, mfERG was more sensitive and specific than VF in detecting subclinical toxicity, and in conjunction with en face OCT and en face OCTA, which showed a high concordance with mfERG, are also useful in detecting subclinical structural abnormalities in patients with SLE. This study provides valuable insights into the multimodal imaging of hydroxychloroquine-induced retinal toxicity and may contribute to the development of improved screening methods. Although further studies are needed to directly compare their utility in subclinical hydroxychloroquine-induced retinopathy, new retinal imaging techniques such as en face OCT and en face OCTA offer a promising way to detect the extent of retinopathy from a single image. As a result of the study, an algorithm for monitoring retinal impact in patients treated with hydroxychloroquine is proposed (figure 4).

Figure 4
Figure 4

Suggested algorithm for monitoring retinal impact of hydroxychloroquine. eGRF, estimated glomerular rate filtrate; FAF, fundus autofluorescence; HCQ, hydroxychloroquine; mfERG, multifocal electroretinography; OCT, optical coherence tomography; OCTA, optical coherence tomography angiography; VF, visual field.