Treatment of myopic choroidal neovascularization: a network meta-analysis and review

Purpose This is, to our knowledge, the first network meta-analysis aiming to compare all treatment modalities for myopic choroidal neovascularization (CNV). Methods After the electronic databases were searched, two independent reviewers screened titles, abstracts, full-texts, and extracted information. Primary endpoints were change in visual outcome and central retinal thickness. We used a network meta-analysis to compare treatment outcomes in the early (≤ 6 months) and late (> 6 months) phase. Results We included 34 studies (2,098 eyes) in our network meta-analysis. In the early phase, the use of anti-VEGF led to a gain of 14.1 letters (95% CI, 10.8–17.4) compared to untreated patients (p < 0.0001), 12.1 letters (95% CI, 8.3–15.8) to photodynamic therapy (PDT) (p < 0.0001), 7.5 (95% CI, 1.2–13.8) letters to intravitreal triamcinolone acetonide (TCA) (p = 0.019), and − 2.9 letters (95% CI, − 6.0–0.2) to the combination of anti-VEGF and PDT (p = 0.065). In the later phase, these results were largely maintained. There were no significant differences in visual outcomes between patients treated with 1 + PRN and 3 + PRN. However, the 1 + PRN group received 1.8 (SD 1.3), while the 3 + PRN group received 3.2 (SD 0.9) injections within 12 months (p < 0.0001). Conclusion This network meta-analysis confirms that anti-VEGF is the most effective treatment for myopic CNV using the 1 + PRN treatment strategy. Supplementary Information The online version contains supplementary material available at 10.1007/s00417-023-06271-2.


Introduction
Pathologic myopia is a major cause of blindness affecting almost 2% of the population worldwide.Although the definition of pathologic myopia has not been standardized yet, it is usually classified as a refractive error of less than − 6.00 diopters and an axial length of ≥ 26.5 mm combined with degenerations of the sclera, choroid, and retina [1].One of the most common complications of pathologic myopia leading to blindness is the development of choroidal neovascularization (CNV).Myopic CNV is associated with a poor prognosis if untreated leading to a decline in visual acuity.More than a third of the patients affected by myopic CNV are at risk of developing myopic CNV in the unaffected eye within 8 years [2].
For a long time, verteporfin photodynamic therapy (PDT) was the only treatment approved for myopic CNV [3].PDT treatment was able to stabilize visual acuity; however, long-term results were discouraging.The development of VEGF inhibitors revolutionized treatment of myopic CNV and soon superseded PDT as the new gold standard treatment [4].
Several studies [5][6][7] have been performed to compare different treatments for myopic CNV; however, no common comparator was used.Therefore, this study is aimed at comparing the efficacy of different treatment options for myopic CNV using a network meta-analysis. 1 3

Data extraction
Two investigators (LP and LG) independently extracted the title, name of authors, year of publication, study design, sample size, treatment, best-corrected visual acuity (BCVA) at baseline and follow-up, central retinal thickness (CRT) at baseline and follow-up, number of treatments, and demographic data.BCVA has been converted to ETDRS letters to enable comparison between the different ways of reporting.Further, descriptive data such as country of origin, definition of myopic CNV, minimal axial length, inclusion and exclusion criteria, and pretreatment were documented.These data were recorded in a Microsoft Excel (Microsoft Cooperation) spreadsheet.

Data analysis
For the analysis, the change in visual outcomes was used, which is given by the mean difference between baseline and follow-up for each treatment group.Most included studies provided means and standard deviations at baseline and for specific follow-up dates.Then, the mean difference can be easily calculated, and for the standard deviation of change, the Cochrane Handbook for Systematic Reviews Studies included in review (n = 64) Studies included in network meta-analysis (n = 34)

Included
of Interventions [12] was followed, assuming a correlation of 0.6.This value was chosen due to a Methods Research Report [13] that refers to a median of 0.59 for correlation of change from baseline.Furthermore, one of the included studies [14] reported a correlation of 0.646.As a sensitivity analysis, no correlation was assumed.If not directly specified, further measures were taken into account to calculate the change and the corresponding standard deviation in visual outcomes.This included the use of p values, confidence intervals, and as a final option, if the standard deviation for the baseline was given but the standard deviation for the follow-up date was missing, the baseline value was used as a surrogate.The network meta-analysis was based on a random effects model, and correlation in multi-arm studies was considered [15].The common heterogeneity variance 2 used in the random effects model was estimated by a generalized DerSimonian-Laird estimator [16].To assess inconsistency, the between-designs Q statistic was calculated based on a full design-by-treatment interaction random effects model [17].The fitted models were used to compare the efficacy of different treatments, for two distinct time points and two separate outcomes.The follow-up dates were grouped into two phases, the first describing treatments in the earlier phase, one to six months.If more than one follow-up date was specified, priority was given to 3 months, then 6 months, and 1 month as the last option.The second time point was considered the later phase, where 24 months, then 12 months, and as the final option, any follow-up dates beyond 24 months were prioritized.
The outcomes determining the efficacy of the treatment referred to the visual improvement measured by the BCVA in letters on the one hand and to the anatomical recovery measured by the CRT in micrometers on the other hand.
Furthermore, subgroup analysis for the different anti VEGFs was performed for the same time points and outcomes.In the primary analysis, we did not distinguish between one initial injection followed by a pro re nata approach (1 + PRN) and three initial injections followed by a PRN approach (3 + PRN); furthermore, we performed a separate pairwise meta-analysis to evaluate possible differences between the two treatment regimens.To compare the number of treatments, we used a two-sample t-test with Welch-Satterthwaite correction on pooled standard deviations and means.
A p value of less than 0.05 was considered statistically significant.All analyses were performed in R, Version 4.1.3[18].

Results
Our literature search yielded 1,156 articles (see Fig. 1).166 full text articles of these were screened for eligibility.We included 64 studies for our qualitative and 34 studies for our quantitative analysis (see Tables 1 and 2).

Study characteristics
In the quantitative analysis, we included 34 studies comprising 2,098 eyes from 2,059 patients.29 studies had two arms and 5 three arms.In the qualitative analysis comprising 64 studies and 4,641 eyes, 52 were two-arm studies, 9 threearm studies, one was a four-arm study, and 2 were five-arm studies.

Outcome in the earlier phase (≤ 6 months)
The evidence network for BCVA in the early phase included 10 studies, representing 5 treatments and no treatment (see Fig. 2).
The other treatment modalities showed less favorable results in the early phase (≤ 6 months).Patients treated with TCA had gained in the mean 6.6 letters (95% CI, − 0.5-13.7)more compared to untreated patients (p = 0.068).The PDT treatment group had no significant change in visual acuity compared to the untreated group (MD − 2.01 letters; 95% CI, − 7.0 − 3.0; p = 0.430).There was no evidence of inconsistency within the network (p = 0.204).
For central retinal thickness (CRT) in the early phase, only 2 studies were included (one two arm and one three arm study).The resulting network structure is therefore very simple (see Fig. 4).Even though the number of comparisons is small, the fitted network meta-analysis shows similar results compared to the analysis of BCVA.We can observe a significant decrease in CRT in patients treated with anti-VEGF compared to untreated patients (66.8 μm; 95% CI, 40.2 − 93.4; p < 0.0001) and patients treated with PDT (27.7 μm; 95% CI, 16.1-39.3;p < 0.0001).The combination treatment of PDT and anti VEGF therapy had a significant larger decrease in CRT than patients treated solely with anti-VEGF (12.0 μm; 95% CI, 21.4-2.6;p = 0.013) (see Fig. 5).Due to the small number of included studies, it is not reasonable to assess inconsistency.

Outcome in the later phase (> 6 months)
Concerning the long-term results of BCVA, the evidence network consists of 16 studies, comparing five different treatments as well as no treatment (see Fig. 6).In the anti-VEGF treatment group, the early outcome could be maintained in the long-term analysis with a mean estimated gain of 28.4 letters (95% CI, 22.7-34.1)when compared to untreated patients (p < 0.0001).Patients treated with anti-VEGF gained 13.1 letters (95% CI, 9.7-16.5)more than patients treated with PDT (p < 0.0001) and 7.5 letters (95% CI, − 1.0-16.0)more than patients treated with TCA, although this was not significant (p = 0.084).There was no significant difference between the anti-VEGF group and the combination (PDT and anti VEGF) group (− 0.02; 95% CI, − 3.9-3.8;p = 0.991).Also, the gain of 9.91 letters (95% CI, − 11.27-31.08) in the surgical group compared to anti-VEGF treatment stayed not significant (see Fig. 7).We did not observe inconsistency in the network (p = 0.328).Central retinal thickness in the later phase was compared using 5 studies with three treatments.Therefore, the network structure shows a triangle shape (see Fig. 8).The network meta-analysis showed no significant difference in the anti-VEGF group compared to the PDT group (10.4 μm; 95% CI, − 37.1-57.8)and no difference to the combination (PDT and anti-VEGF) group (25.3 μm; 95% CI, − 56.7-107.2) (see Fig. 9).Again, this network did not show signs of inconsistency (p = 0.447).

Differences in anti-VEGF drugs
We compared the change in BCVA of different anti-VEGF drugs in the early phase including 8 studies and in the later phase including 13 studies.There was no significant difference in letters gained in patients receiving bevacizumab compared to aflibercept (p = 0.222), ranibizumab (p = 0.124), and conbercept (p = 0.572) in the early phase, the same was seen in the later phase (p = 0.250, p = 0.265, respectively, p = 0.382).
For CRT, we investigated 5 studies for both time points.In the early phase, CRT decreased significantly in patients  4 The structure of the network comparing different treatments regarding BCVA in the early phase (< 6 month).The numbers represent the numbers of direct comparisons, while the thickness of the lines is proportional to the inverse standard error of the estimates.PDT, photodynamic treatment; TCA, intravitreal triamcinolone acetonide; VEGF, vascular endothelial growth factors receiving aflibercept compared to bevacizumab (12.1 μm; 95% CI, 3.0-21.2;p = 0.009).There was no significant difference in the change of CRT between bevacizumab, ranibizumab (7.6 μm; 95% CI, − 13.3-28.5),and conbercept (− 5.4 μm; 95% CI, − 41.5-30.8).Moreover, there was also no significant difference observed comparing long-term results of the different anti-VEGF factors.
Also, the number of injections in patients receiving PDT + anti-VEGF versus solely anti-VEGF was compared.Patients receiving combination treatment required 2.2 (SD 1.5) injections, and patients receiving only anti-VEGF treatment required 2.6 (SD 1.3).This difference was not significant (p = 0.155).

Other treatments
Other treatment options for myopic CNV had too few comparators for our quantitative analysis.A summary statement for each option is given in our supplementary table.

Discussion
This network meta-analysis showed that the intravitreal injection of anti-VEGF using the regimen of 1 + PRN is an effective treatment for myopic CNV, with both short-and long-term beneficial results.
Intravitreal injection of anti-VEGF is considered the gold standard treatment for myopic CNV, which is confirmed in this network meta-analysis.In diabetic macular edema, aflibercept is proposed to lead to a greater improvement in visual acuity compared to other VEGF inhibitors in patients with low baseline BCVA (< 69 letters) [24].Therefore, we compared the different VEGF inhibitors, i.e., bevacizumab, ranibizumab, aflibercept, and conbercept.However, we found no difference between the VEGF inhibitors.Aflibercept led to a larger decrease in CRT, but this had no impact on visual acuity.Due to small sample sizes, we did not differentiate between low and high baseline BCVA.Future research should investigate this.
We then compared the different treatment strategies for VEGF inhibitors.There was no significant difference in letters gained whether three injections were administered consecutively as loading dose or only one.However, patients treated with 1 + PRN required significantly less injections than patients with 3 + PRN.This outcome might indicate that the 3 + PRN treatment strategy leads to an overtreatment.Future research should investigate in subgroup analysis, whether this is true for different VEGF inhibitors.
Combining anti-VEGF treatment with PDT showed a slightly greater decrease in CRT in the early phase, although the absolute difference of 12 μm may be clinically insignificant.There was a tendency to gain more estimated letters, but this was not significant.In the long-term results (> 6 months), change in BCVA and CRT was the same for anti-VEGF treatment and the combination of PDT and anti-VEGF.There was no difference between these two groups in the number of injections required within 12 months.Considering the absence of randomized controlled trials and the lack of differing results, anti-VEGF monotherapy seems the more reasonable first line treatment.
Intravitreal TCA was inferior to anti-VEGF in terms of letters gained in the short-term analysis, but no statistical difference was seen in long-term analysis.Intravitreal TCA is known to cause an IOP increase in nearly one-third of all patients and has a high prevalence of cataract formation and progression over time.In regard of these known side effects, anti-VEGF appears to be the more favorable choice.
When comparing the previous gold standard PDT for myopic CNV to anti-VEGF, patients with PDT gained significantly less letters over all time periods.This strengthens the use of anti-VEGF over PDT.
In our systematic review, it seems unlikely that other treatment options for myopic CNV show similar visual improvement compared to intravitreal VEGF inhibitors, although patient numbers were too small to prove this in our quantitative network meta-analysis (see supplementary table 2).
The numbers of complications were too small to calculate the risk of complications.In Table 2, we reported complications rates, which were low in general.Surgical interventions had the highest complication rates.Intravitreal steroids, as known, showed an increase of intraocular pressure and cataract progression.In patients with intravitreal VEGF inhibitors, some patients showed corneal erosions and dry eye symptoms after injection.Not all studies reported on these relatively common adverse events, which is the reason why no numbers can be given.The same applies to IOP elevation, as most studies did not measure IOP after injection.There were three (0.001%) reports of retinal detachment after intravitreal injection and one (0.0004%) case of sterile vitritis in the studies reporting on complications.
This network meta-analysis has several limitations.The included studies showed a high degree of heterogeneity of patients' characteristics, most likely attributable to differences in inclusion and exclusion criteria (see Table 1).Some studies included pretreated patients, while other studies included only treatment-naïve patients.Furthermore, there exists no clear definition of pathologic myopia, and so the studies included slightly different patient populations.Some studies did not report on the definition of myopic CNV used in their study, making a comparison even more difficult.Another very relevant exclusion criteria for intravitreal treatment is the history of vitreous surgery.Again, some studies excluded these patients explicitly, while others included them.As the search was limited to publications in English, we might have missed some studies.However, based on visual inspection of funnel plots and analytical methods, we did not observe signs of publication bias.Further, databases were searched for specific keywords, which did not include all treatment options (for example, laser photocoagulation).
Databases were searched for the following keywords: "myopic choroidal neovascularization".
Another limitation of this study was the different reporting times of the studies e.g., some studies reported on results after one month, three months, or six months.As our sample size would have been too small to compare the exact time points, we had to pool the different follow-up data under the assumption that the different time points were effectively the same.To make the results more comparable, we gave priority to certain time points, i.e., 3 months, then 6 months and 1 month in the early phase, and 24 months, then 12 months, and as a last option, all follow-up time points after 24 months in the late phase.However, the classification of follow-up dates might bias our results.Further, not all studies used the EDTRS charts for visual acuity testing, and we had to calculate the letter score from other scales.Different OCT devices were used for measuring the central retinal thickness in the studies, making comparison difficult.Additionally, few studies reported CRT as an outcome, which weakens the validity of our results.
Another major limitation of this network meta-analysis is the inclusion of non-randomized trials, which could lead to potential bias within each study.In addition, the inclusion of RCTs and observational studies could result in study designs and data collection which are not comparable.

Conclusion
This network meta-analysis shows that intravitreal VEGF inhibitors are the most effective treatment of myopic CNV with few adverse events and a preferred treatment regimen of 1 + PRN.

Fig. 2 Fig. 3
Fig.2The structure of the network comparing different treatments regarding BCVA in the early phase (< 6 month).The numbers represent the numbers of direct comparisons, while the thickness of the lines is proportional to the inverse standard error of the estimates.BCVA, best-corrected visual acuity; PDT, photodynamic treatment; TCA, intravitreal triamcinolone acetonide; VEGF, vascular endothelial growth factors

Fig. 5 Fig. 6 Fig. 7
Fig. 5 Forrest plot comparing change in central retinal thickness before six months in the anti-VEGF treatment group compared to the other treatment groups.CI, confidence interval; MD, mean difference; PDT, photodynamic treatment; VEGF, vascular endothelial growth factors

Fig. 8 Fig. 9
Fig.8The structure of the network comparing different treatments regarding BCVA in the early phase (< 6 month).The numbers represent the numbers of direct comparisons, while the thickness of the lines is proportional to the inverse standard error of the estimates.PDT, photodynamic treatment; VEGF, vascular endothelial growth factors

Table 2
Table of study characteristic and inclusion criteria of all 64 studies included for qualitative analysis Table of complications rates for all 64 studies included for qualitative analysis