Discussion

This systematic review and meta-analysis investigated the test accuracy of OCTA and SD-OCT for the diagnosis of mCNV compared to FA as the reference standard. An extensive literature search identified three eligible OCTA studies and three eligible SD-OCT studies. Although meta-analysis revealed that both OCTA and SD-OCT diagnosed mCNV with high sensitivity, a pooled estimate of specificity for SD-OCT could not be calculated. These findings have implications for the clinical diagnosis of mCNV using either OCTA or SD-OCT.

Using the GRADE criteria^{[11, 12]} to consider uncertainty, a ‘conditional recommendation’ could be made to support the use of OCTA as an initial test when mCNV is clinically suspected. This recommendation is ‘conditional’ because a large proportion of OCTA images were not of acceptable quality to achieve a positive or negative diagnosis. Studies excluded up to 25% of patients from analysis on account of image quality issues^{[25, 26, 28]}. OCTA is known to be prone to motion and projection artefacts as many patients have poor fixation due to co-existing myopic maculopathy^[4]. Future advances in imaging hardware, software tools and image processing may help overcome these image acquisition limitations^[29]. Meanwhile, a weaker ‘suggestion to consider’ SD-OCT was made because pooled specificity could not be estimated. Reliance on SD-OCT alone to diagnose mCNV may result in an excessively high false positive rate leading to unnecessary over-treatment. Nevertheless, we suggest SD-OCT could be considered when an OCTA image of sufficient quality cannot be acquired.

Based on high sensitivity, if either OCTA or SD-OCT return a negative result, clinicians can be reasonably confident in ruling out mCNV. However, if either OCTA or SD-OCT return a positive result, FA should be performed to rule out false positives and confirm the diagnosis. These evidence-based guidance statements are consistent with existing expert opinions which recommend a positive diagnosis of mCNV obtained with OCTA or SD-OCT should be confirmed by FA^{[4, 8, 9]}.

The authors of the included studies reported several reasons to explain why the performance of OCTA and OCT did not match FA. For example, retinal alterations resulting from co-existing myopic maculopathy, such as lacquer cracks, retinal pigment epithelium and chorioretinal atrophy, and retinoschisis, interfered with the ability of OCTA to detect mCNV^{[26, 28]}. The presence of submacular haemorrhage in particular may lead to false-negative detection of mCNV by OCTA compared to FA^[26]. The authors further reported that poor patient fixation and projection artefacts resulted in lower quality images, thus limiting the capability of accurately detecting small or poorly perfused mCNV^[28]. Finally, the small size of many mCNV lesions noted in multiple included studies^[24–26] was suggested to have prohibited proper signal detection of mCNV by OCTA. SD-OCT cannot visualise fine vessels nor provide functional information on the retinal microcirculation^[28], and Leveziel et al.^[30] reported that the exudative features of mCNV were more obvious on FA than on SD-OCT. Milani et al.^[24] reported the absence of retinal fluid, haemorrhage, and hyper-reflective foci resulted in diagnostic difficulties using SD-OCT compared to FA. SD-OCT is also prone to segmentation errors in highly myopic eyes because the thinner retina causes enhanced visualisation of choroidal vessels which may be difficult to differentiate from mCNV^[25].

First described in 1961^[31], FA continues to be considered the reference standard for the diagnosis of mCNV^[4]. In the studies included in this review, active mCNV was diagnosed on FA by well- defined hyperfluorescence in the early phase that increased in leakage size and intensity in the late phase^{[24–26, 28]}. Associated blood or pigmentation showed as blocked fluorescence^[26]. FA findings in inactive mCNV comprised hyperfluorescent staining of a macular scar due to fibrosis in the absence of leakage^{[9, 28]}. Meanwhile on OCTA images, mCNV appeared as an abnormal network of bright interlacing or tangled vessels in the outer retina and choriocapillaris slabs^{[26, 28]}. Bagchi et al.^[26] additionally reported the presence of a tight vascular net and the combination of a perilesional halo with visible core were features that indicate the presence of active mCNV^[26]. On SD-OCT scans, mCNV was accepted to present as a dome-shaped area of homogenous hyperreflectivity either below or above the retinal pigment epithelium^{[24, 26]}. Other features of a positive diagnosis of mCNV on SD-OCT included an overlying fuzzy area, absence of external limiting membrane visibility, disrupted photoreceptor ellipsoid zone, presence of subretinal hyper-reflective material, and subretinal and/or intraretinal fluid^{[26, 28]}.

Methodologically rigorous test accuracy studies are necessary to inform clinical decisions regarding the provision of safe and effective patient care^{[32, 33]}. To assess the methodological quality of the included studies, we employed a well-established, objective grading criteria developed specifically for use in systematic reviews of diagnostic tests^[17]. All studies were found to contain at least one major flaw leading to an overall high risk of bias, with the most concerning potential for bias introduced by the common use (3 of 5 studies) of case-control patient sampling (Fig. 2).

Case-control studies are prone to selection bias when the cases or controls are obtained in ways such that either cases or controls are not truly representative of the broad spectrum of patients to whom the diagnostic test will be applied in practice^[34]. Case-control studies are accepted to overestimate diagnostic accuracy^{[17, 32]}. Specific to our systematic review, selection of control patients was not broad enough to allow stable estimates of false positive or true negative event rates, thus a pooled estimate of specificity for SD-OCT could not be calculated (Table 2). One remedy for this issue would be to employ population-based random or consecutive sampling with a focus on obtaining representative populations of patients who are tested but return false positive results.

Our systematic review found only 2 of 5 included studies (40%) recruited consecutive patients in order to obtain representative populations. Johnson et al.^[35] evaluated the quality of diagnostic accuracy studies using OCT to diagnose glaucoma and similarly noted only 8 of 30 publications (26.7%) reported using consecutive or random selection. With both our systematic review and Johnson et al.’s systematic review identifying a need for improvements in patient selection and avoidance of case-control studies, we strongly recommend future investigators become aware of the QUADAS-2^[17] and GRADE^{[11, 12]} risk of bias assessment and reporting criteria before starting their projects. Familiarity with these guidelines will also help to avoid other types of major methodological flaws.

Strengths and limitations

It is unlikely our systematic review missed any relevant studies. The comprehensive literature search was performed across two major databases (MEDLINE and EMBASE)^[36] using search strategies and specific terms optimised to each database (See Supplement eTable 1)^{[14, 15, 37]}. Language restrictions were not applied to the electronic search and reference lists of retrieved articles were hand-searched for additional eligible studies. Whilst two non-English studies were identified by the search strategy^{[38, 39]}, their English-language abstracts revealed they were not on-topic and therefore would not have qualified for inclusion. Furthermore, we did not specify any test device or model restrictions in the search, which additionally increased our ability to identify meaningful studies.

Unfortunately, the comprehensive search identified relatively few small studies that were on-topic. As such, inferences could not be drawn on potential sources of heterogeneity, e.g. the patient population or specific imaging device used. Nevertheless, by using an objective risk of bias assessment tool and following established methodological guideline development processes^{[11, 12]}, we were able to communicate the uncertainty arising from lack of high-quality evidence by choosing conservative wording to frame our clinical guidance statements.

One key limitation of this study may be associated with the use of the term “choroidal neovascularisation”. Despite recommendations for standardisation towards replacing “choroidal neovascularisation” with the term “macular neovascularisation”^[40], the National Library of Medicine’s MEDLINE database still uses “choroidal neovasculariza- tion” as a MeSH category to index articles on this topic. Furthermore, searching PubMed with the term “macular neovascularisation” does not map to the established MeSH category “choroidal neovascular- ization”. Therefore, to aid readers in finding our systematic review when conducting an electronic literature search, we used the term “choroidal neovascularisation” throughout our manuscript^[16].

Future research directions

Future studies evaluating diagnostic test performance should ensure their study population is representative of the broad spectrum of patients expected to undergo the test in practice. This is best achieved by avoiding case-series and case-control designs and enrolling consecutive or randomly selected patients^{[17, 41]}. Furthermore, patients clinically suspected of mCNV may have co- existing diseases such as age-related macular degeneration or media opacities, thus avoiding inappropriate exclusions based on co-existing disease allows for a more representative study popula- tion and a more pragmatic estimation of test accuracy^{[17, 41]}.

Evaluations of novel diagnostic tests should be prospective and ensure the index test and reference test are applied to all enrolled patients, independent of their respective test results^{[17, 41]}. Future projects should also aim to link clinical decisions guided by diagnostic test results to downstream patient consequences measured using validated health outcomes, such as quality of life, that adequately capture the impact of inappropriate treatment decisions^[21]. Another avenue for investigation is the diagnostic accuracy of swept-source OCTA or swept-source OCT for mCNV as this newer technology allows improved visualisation of deeper retinal structures, including the choriocapillaris and choroid, due to the penetration of the longer wavelength light source and increased scanning speeds compared to SD devices^[42]. Furthermore, manual segmentation of OCTA volumes has been found to increase the sensitivity for detection of choroidal neovascularisation due to various pathologies including mCNV compared to automatic segmentation, emphasising its importance in future research and test procedures^[43]. Whilst this current systematic review focused on the diagnosis of mCNV, future investigators may address questions relating to the utility of OCTA/ OCT for monitoring mCNV progression and resolution, and the diagnostic performance of specific imaging markers of mCNV.

We acknowledge these recommendations will require the conduct of larger studies that may need significant funding. However, appropriately conducted research will pave the way towards stronger clinical recommendations that ultimately result in practice change and improved patient outcomes.