Interpretation of SD-OCT imaging data in real-life conditions versus standardized reading centre analysis in eyes with diabetic macular oedema or macular oedema secondary to retinal vein occlusion: 24-month follow-up of the ORCA study

In both diabetic retinopathy (DR) and retinal vein occlusion (RVO), macular oedema (ME) may disrupt the retinal morphology and impair vision [1]. Intravitreal anti-vascular endothelial growth factor (anti-VEGF) therapeutics improve retinal morphology and function [2‐4] and are a first-line treatment option for both conditions [5, 6]. Retreatment criteria rely on a combination of changes in visual acuity (VA) and monitoring of the extent, localisation, and temporal duration of retinal fluid accumulation visible on optical coherence tomography (OCT). Regular control visits and adequate interpretation of OCT scans are therefore essential to determine when retreatment is required [7].

In real-world clinical settings, treatment outcomes frequently do not match those reported in clinical trials, which may be at least partly due to misinterpretation of OCT imaging findings in addition to multiple other clinician- and patient-dependent factors [8‐10]. Accurate interpretation of OCT findings is crucial for optimal treatment decisions.

The ORCA module of the 24-month OCEAN study compared the interpretation of spectral domain OCT (SD-OCT) scans by physicians and RCs in cohorts of patients undergoing anti-VEGF (ranibizumab) treatment for neovascular age-related macular degeneration (nAMD) [8], diabetic macular oedema (DME) or ME secondary to RVO [11‐13]. The comparative analysis between the assessment of the initial diagnosis at baseline by treating physicians and Reading Centres (RCs) has already been reported [14].

In the current analysis, longitudinal data from ORCA patients diagnosed with either DME or ME secondary to RVO were assessed to determine the accuracy of OCT interpretations by treating physicians and the adequacy of physicians’ treatment decisions over 24 months of follow-up in real-world clinical practice.

The cohorts analysed were comprised of patients from the ORCA module as part of the prospective, multi-centre, non-interventional Phase IV OCEAN study¹ [8, 11, 13]. Physicians at clinical sites performing SD-OCT examinations as part of their routine clinical care could participate in the ORCA module [8]. To allow adequate evaluation of SD-OCT volume scans, the RCs developed a consensus for minimum quality requirements for colour fundus photography (FP), fluorescein angiography (FA) and SD-OCT imaging that were aligned with criteria issued by professional associations [15, 16]. A handbook for image acquisition and image data transmission to the RCs was provided to the participating sites.

Patients who had not received anti-VEGF injections within three months prior to study entry or any previous treatment with intravitreal steroids with RC-confirmed diagnosis of DME or ME secondary to RVO were included. Subjects were followed over 24 months at 36 ophthalmic centres in Germany. Information on best corrected visual acuity (BCVA) at all visits, number and type of anti-VEGF injections, and any additional treatments were documented.

Collected imaging data included all routinely performed SD-OCT scans (Optopol Copernicus, Zeiss, Topcon and Heidelberg Engineering) and available FP and FA. Images were evaluated by treating physicians through a standardised OCT questionnaire and by three independent RCs.² All RCs were blinded to the results of the physicians [14].

Retinal thickness parameters, calculated from the internal limiting membrane to Bruch’s membrane, included central subfield retinal thickness (CSRT, i.e., the average thickness in the central 1 mm² area of the fovea) and foveal centre point thickness (FCPT, i.e., the thickness at the fovea as measured on the most central B-scan). CSRT values were provided by the automated viewing software after checking for correct segmentations and a correctly centred positioning of the ETDRS grid. In case of segmentation errors, manual correction of segmentation lines was performed if the required time to perform the correction was less than 2 min.

OCT signs of disease activity were defined as presence of intraretinal fluid (IRF) and/or subretinal fluid (SRF) with foveal involvement and/or diffuse foveal thickening (Online Resource 1). In addition to the comparison of the OCT evaluation, the adequacy of the physicians’ treatment indications was assessed by the RCs based on OCT imaging findings, also considering the course of oedema and the course of VA. Possible reasons for suspected undertreatment or overtreatment were evaluated. Grading options for presence of IRF/SRF included ‘yes’, ‘no’, ‘questionable’ and ‘cannot grade’. For statistical analysis, ‘questionable’ and ‘cannot grade’ were counted as ‘no’, as no definite presence was confirmed. The RCs also evaluated whether an OCT parameter was strongly or discreetly visible. The statistical analyses were descriptive.

OCT-guided anti-VEGF therapy is an established first line treatment option for both DME and ME secondary to RVO [5, 6, 17, 18]. Discrepancies between clinical trials and real-world settings have been noted with regards to both injection frequency and VA outcomes [19‐24].

This study shows an overall high rate of agreement but also identifies relevant differences in parts of the interpretation of OCT images by physicians and RCs. Misinterpretation of OCT images may be related to potential undertreatment or possible overtreatment. These results support the hypothesis that OCT scan misinterpretation contributes to unsatisfactory outcomes as previously reported for real-life situations [25, 26].

In contrast to data for nAMD in the ORCA module, the automatic OCT-segmentation was deemed correct for most cases of ME secondary to RVO or DME. Even if the data showed significant differences between the physicians and RCs, the localisation of the fovea and the correct positioning of the ETDRS grid seemed to be less challenging in DME and in ME secondary to RVO than in nAMD [14]. Identification of foveal centre and a correct grid position is mandatory for a correct measurement of CSRT and FCPT.

The levels of agreement on presence or absence of IRF and/or SRF with foveal involvement between physicians and RCs were higher for scans where physicians reported presence of IRF/SRF than for scans where physicians reported absence of IRF/SRF [27, 28].

Physicians reported false positive IRF/SRF approximately twice as frequently as false negative among both the DME and RVO cohort patients. Although false positive OCT interpretations may lead to overtreatment, it is unlikely to have a negative impact on VA outcomes but may incur avoidable costs to the healthcare system and increase the risk of treatment-related adverse events [8].

We assume that underestimation of IRF/SRF by physicians has contributed to the fact that in two thirds of visits with foveal oedema no injection was given. Although false-negative evaluation of IRF and/or SRF by physicians was low, this could be related to approximately one quarter of OCT scans that did not lead to anti-VEGF injection despite fovea involving oedema.

Other possible reasons for not injecting include stability of VA and/or stability of edema over three consecutive visits. However, in about 30% of all cases with foveal oedema no treatment was given and no comprehensive reason for a possible ‘watch and wait’ procedure could be discovered.

The number of no-treatment decisions in cases where foveal IRF/SRF was detected by RCs suggests that the interpretation of OCT scans may constitute an opportunity for improvement in real-life clinical practice decision making. Even if the negative effect of SRF and IRF on retinal function and integrity is much more pronounced and more rapid in nAMD compared to ME caused by retinal vascular diseases, a BCVA deterioration was seen in the following visit in a quarter of cases (DME and RVO 27.3% and 24.7%, respectively) with presumed undertreatment. That finding suggests that a more rigorous treatment regimen may have been beneficial for optimal preservation of vision over time in these patients. Real-life studies demonstrated that undertreatment often leads to unsatisfying results in treatment of DME and in ME secondary to RVO. To achieve an effective treatment frequency in an as needed anti-VEGF-therapy an adequate number and a correct interpretation of OCT-examinations apparently seems to be crucial.

Study limitations include the non-interventional design of the ORCA module, small cohort sizes; heterogeneity of German OCT systems, and proprietary data formats and regulatory requirements [29]. Additionally, since standard physician OCT image interpretation skills were required for study participation, more experienced physicians may be even over-represented, which may have led to underestimation of the true real-world extent of deviating assessments. Also, individual reasons for treatment discontinuation despite correctly detected signs of IRF/SRF were not regularly captured, which limits treatment discontinuation analysis.

In summary, the overall approximately 70% consensus rate for determining presence or absence of IRF/SRF between independent RCs and treating physicians indicates that IRF/SRF is often correctly identified in real-life clinical practice conditions. In contrast to nAMD, segmentation problems of the OCT algorithm are rare in retinal vascular diseases and the retinal thickness measurements of the CSRT and the FCPT seem to be robust parameters, likely due to the clear identification of retinal borders and the foveal centre. Misinterpretation of OCT images and omitting injections despite presence of foveal fluid may have contributed to potential undertreatment in up to two thirds (62.3%) of ORCA-DME patients and half of (51.8%) ORCA-RVO patients, which may negatively impact VA outcomes. Misinterpretation of OCT scans could also result in overtreatment in a subset of patients, thus increasing the risk of treatment-related side effects and costs to the healthcare system. In conclusion, improving treating physicians’ ability of adequate SD-OCT interpretation by education programs along with additional help in daily OCT evaluation by future artificial intelligence assistance, may lead to improved clinical outcomes and more precise anti-VEGF treatment for patients diagnosed with DME or ME secondary to RVO.