Intravitreal bevacizumab versus intravitreal triamcinolone for diabetic macular edema–Systematic review, meta-analysis and meta-regression

Background The most frequent cause of vision loss from diabetic retinopathy is diabetic macular edema (DME). Earlier clinical trials tried to examine the role of intravitreal triamcinolone (IVT) and intravitreal bevacizumab (IVB) in DME; they either qualified IVT over IVB or IVB over IVT or did not exhibit a significant difference. Objective This paper aims to compare the efficacy and safety of IVB versus IVT alone or combined IVB+IVT in the treatment of DME. Methods We systematically searched PubMed, CENTRAL, Scopus, Embase, Science Direct, OVID, and Web of Science for randomized controlled trials of IVB versus IVT alone or combined IVB+IVT and IVT versus the combined IVB+IVT in DME patients. Results A total of 1243 eyes of 17 trials were included in our meta-analysis and regression. Repeated injections of IVB were superior at improving VA comparing with those of IVT at 12, 24, 48-weeks, and IVB+IVT at 12, 24, 48-weeks. Single injections were comparable across the three arms regarding BCVA improvement. CMT reductions were also comparable across the three arms. Meanwhile, the overall safety regarding intraocular pressure and intraocular hypertension significantly favored the IVB group. Improvement in VA was best modified with CMT reduction from 480 um to 320um. This association was significant at 12-weeks in the three arms and persisted till 24-weeks and 48-weeks exclusively in the IVB group. Conclusions and relevance Our analysis reveals that repeated successive injections associate with better BCVA compared to single injection. Current evidence affirms that IVB is superior to IVT and IVB+IVT at improving BCVA, comparable at reducing CMT, and presents a better safety profile in the treatment of DME.


Introduction
Diabetes mellitus (DM) affects over 422 million persons worldwide [1]. About 33% of patients with DM develop some form of diabetic-related eye damage [2]. For instance, the 10-year incidence of diabetic retinopathy in patients with T1DM is nearly 36%, while the 20-year incidence for T2DM is 84% in those taking insulin and 53% in those not taking insulin [3][4][5]. Diabetic retinopathy is a microvascular disorder correlated with the thickening of the peripapillary retinal nerve fiber layer [6]. The most frequent cause of vision loss from diabetic retinopathy is diabetic macular edema (DME), which can develop at any stage of retinopathy and is marked by edema and retinal thickening [7].
In DME chronic hyperglycemia upregulates vascular endothelial growth factor (VEGF), increasing vascular permeability, and angiogenesis [8,9]. Additionally, a decent amount of evidence suggests that inflammatory mediators are partially engaged in the pathophysiology of DME and contribute substantially to the vascular permeability and the development of edema [10][11][12].
The potential treatments for DME involves macular laser photocoagulation (MLP), anti-VEGF agents, ocular corticosteroids, and pars plana vitrectomy [13]. MLP was the primary treatment for DME proving to be effective in limiting vision loss [14]. Over time the intravitreal injections of anti-VEGF agents have rapidly become the standard of care, due to its ability to adjust both visual and anatomical outcomes, besides the avoidance of laser-related complications like subretinal fibrosis and laser scars [15][16][17].
Being an ocular steroid, Triamcinolone performs an anti-inflammatory, anti-angiogenic, and cost-effective role in the treatment of DME; proved to be beneficial through several reports [18,19]. Meanwhile, the humanized monoclonal antibody Bevacizumab presents promising anti-VEGF results working as an off-label therapeutic favorable option -for it is more affordable than most of the anti-VEGF agents. Comparatively, triamcinolone requires fewer injections, and a single intravitreal triamcinolone (IVT) injection might be as effective as three injections of intravitreal bevacizumab (IVB) for the treatment of DME [20]. This implies that IVT may reduce injection-related complications and improve patient compliance. However, the rates of intraocular pressure (IOP) increase and cataract development are expected to be higher in steroids-treated eyes [21,22].
Earlier individual trials either qualified IVT over IVB or IVB over IVT or did not exhibit a significant difference [23][24][25]. Previous cumulative reviews tried to settle this controversy [26,27], but with limited double-arm randomized controlled trials (RCTs) at the time, unreliable statistical methods, and short-term follow-ups; the debate is still unsettled.
In this systematic review, meta-analysis, and meta-regression of multiple-arm (RCTs): we compare the short and long-term safety and efficacy of IVB versus IVT alone or combined with IVB in the treatment DME, regarding visual acuity (VA), central macular thickness (CMT), IOP, intraocular hypertension (IOH), and pathogenesis factors (such as hypertension, diabetes duration, and HbA1C levels).

Materials and methods
We followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) statement guidelines [28], as well as the standards of the Cochrane Handbook for Systematic Reviews of Intervention [29].

Literature search strategy
The following electronic databases were comprehensively searched: PubMed, CENTRAL, Scopus, Embase, Science Direct, OVID, and Web of Science; using relevant keywords "diabetic macular edema", "DME", "bevacizumab", "Avastin", "triamcinolone", "IVB", "IVT" from inception till 20 August 2020. All published articles were considered with no restrictions in terms of language or publication period. Further, we manually scanned the bibliography of retrieved articles for additional relevant studies.

Eligibility criteria and study selection
We included studies with the following criteria: (a) randomized controlled trials (RCTs) that compared IVB of any dose with or without IVT of any dose; (b) patients of any age and sex, who had any type of diabetes, clinically significant DME and receiving naïve treatment; (c) reported data on any of the following outcomes: best-corrected visual acuity (BCVA) between 0.096 log-MAR (Snellen = 20/25; ETDRS VALS = 80) and 1.3 logMAR (Snellen = 20/400; ETDRS VALS = 20), central macular thickness (CMT) > 300 μm defined by OCT machine, intraocular pressure (IOP), intraocular hypertension (IOH) at various weeks endpoints (4, 6, 12, 24, and 48-weeks); (d) Studies of additional injections or retreatment based on persistence of clinically significant macular edema according to Early Treatment Diabetic Retinopathy Study (ETDRS) criteria; (e) duplicated publications or reports were included once. Articles were excluded if: (I) non-randomized controlled trials or comparative interventional case series; (II) studies with DR without macular edema or studies of macular edema due to causes other than DR; (III) studies that compared IVB or IVT with different intervention, or studies that concerned with non-ocular outcomes or non-DME patients, as well as dissertations/thesis or animal studies; (IV) patients with macular edema related to recent intraocular surgery or other procedures, vitreous traction, history of any treatment for DR at any time or anticipating the need for pan retinal laser photocoagulation, existing/pre-existing glaucoma or ocular hypertension (IOP> 21 mmHg), steroid responders, recent history of arterial thromboembolic event, poorly controlled hypertension, or use of systemic steroids and/or systemic anti-VEGF agents. (II), (III) and (IV) were considered irrelevant during the screening process. Duplicates were removed using EndNote X7.1 software and retrieved references were screened in two step-wise manner: titles/abstracts screening for matching our inclusion criteria, followed by a full-text appraisal of relevant articles for eligibility to meta-analysis. Each step was performed by two independent reviewers.

Data extraction and risk of bias assessment
Each type of dataset was extracted independently by two authors. Discrepancies were settled through discussion and consensus among the reviewers. The extracted data involved the following: (1) study ID (name of the first author and year of publication), location, study design, major inclusion criteria, various intervention groups (arm, dosage, number of injections and the interval in-between), number of eyes, follow up duration and the conclusion of each study; (2) Baseline characteristics for each intervention arm of enrolled patients regarding age, sex, type of DM, insulin users (%), HbA1C level, hypertensive patients (%), and retinopathy severity (%); whether non-proliferative diabetic retinopathy (NPDR), proliferative diabetic retinopathy (PDR), or regressed PDR; (3) Risk of bias (ROB) domains; (4) Treatment outcome measures . The  following outcome measures were extracted at 4, 6, 12, 24 and 48-weeks; to indicate the short,  intermediate and long term efficacy related to the treatment groups: (I) means and standard deviations (SDs) of the different values for BCVA, CMT, and IOP at each endpoint following the  intervention per eye; (II) number of eyes developed IOH of more than 21 mmHg IOP. We adopted the Cochrane risk of bias (ROB) assessment tool, adequately described in chapter 8.5 of the Cochrane handbook [29]. ROB domains included Randomization (selection bias); Allocation concealment (selection bias); Blinding of participants (performance bias); Blinding of outcome assessment (detection bias); Incomplete outcome data (attrition bias), Selective reporting (reporting bias), and other sources of bias including unclear baseline characteristics and trial termination shortly due to data-dependent considerations. We classified RCTs in each domain as low, high, or unclear ROB as defined by Cochrane Handbook. Any discrepancies were resolved through discussion. The assessment of publication bias using the funnel plot and Egger's test was also considered (Fig 3B). We also considered the Grading of Recommendations Assessment Development and Evaluation (GRADE) approach (Table 1).

Data analysis
Statistical analysis was performed using Open Meta[Analyst] package from Brown University -School of Public Health. Applying the random-effects model with Der-Simonian Liard method: continuous data of means and standard deviations were pooled as weighted mean differences (MD), dichotomous data of event-total were calculated as relative risks (RR). Subsequently, the MD (VA, CMT, IOP) and RR (IOH) among the three arms (IVB vs. IVT or IVB vs. IVT+IVB or IVT vs. IVT+IVB) were analyzed and provided a 95% confidence interval (CI). Missing SD of mean change from baseline was calculated from the standard error or 95% CI. To test for statistical heterogeneity between trials Chi-square and I2 tests were performed; values of 0%-40%, 30%-60%, 50%-90%, and 75%-100% represented low, moderate, substantial, and considerable heterogeneity, respectively according to Cochrane Handbook of Systemic Review and Meta-analysis. P<0.1 was set as a level of significant heterogeneity. When significant heterogeneity was detected, we conducted a sensitivity analysis to find the source of heterogeneity by excluding one study at a time. Subgroup analysis according to study arms and repeated injections was also performed. Additionally, a meta-regression was employed to examine whether injections, sex, age, hemoglobin A1C level, diabetes duration, insulin usages, hypertension, or degree of retinopathy may predict alterations in VA and CMT.

PLOS ONE
Bevacizumab versus triamcinolone in DME; a meta-analysis

The potential source of bias
According to the Cochrane ROB tool, the quality of the included studies was from moderate to high. The main concern was incomplete outcome data (loss of follow-up), which was

Meta-regression models
Results from multiple regression models showed that the rates of BCVA, CMT and IOP were significantly modified by sex, DM duration, insulin use, HbA1C levels, hypertension (HTN); this combinations yielded R 2 100% (Coefficients 0.

Discussion
In this systematic review and meta-analysis of 17 RCTs and 1243 eyes: six of our included studies reported superiority of IVT or combined IVB+IVT compared with IVB alone in the treatment of DME [23,24,31,38,43]. However, eight other trials demonstrated that IVB was more efficient in reducing DME relative to IVT or IVB+IVT [32,33,35,37,[39][40][41]44]. To complicate this even further: three other trials reported that the two drugs did not differ markedly in terms of their effects in improving VA and reducing CMT [34,36,42]. In a previous meta-analysis of 6 RCTs by Zhang et al.: IVT was superior in improving short-term VA and reducing long-term CMT [26]. Nonetheless, the relatively small sample size, short term follow-ups, absence of repeated-dose consideration, fixed-effect model reliance and substantial heterogeneity left the question unanswered. Which treatment is more efficient remains a valid debate. Thus, we performed this meta-analysis to compare the efficacy and safety of IVB with IVT alone or combined IVB+IVT in DME patients. We considered the long-term follow-ups, the effect of multiple injections, and the possible associations between the underlying pathogenesis and the drug's mechanisms of action. It's important to note that we could not include the work of Shimura et al. in our final analysis [45]. Though it was included as an RCT in the previous metaanalysis, we found no characteristics of an RCT design in the original manuscript. Attempts to contact the authors for clarification received no response, so we excluded the study.
In our analysis: we found that the group who received repeated injections of IVB had a statistically significant improvement in BCVA over the relative IVT and IVT+IVB groups at 12, 24, and 48-weeks follow-up. Still, the three groups were comparable regarding CMT reduction as the difference was not significant during the early, intermediate, and late follow-ups (up to 48 weeks). Even though a slight superiority was present for IVB at 4-weeks, IVB+IVT at 12-weeks, and IVT at 24-weeks; the very wide CI of these findings exclude it from clinical significance. Although IVT presented a slight increase at 12 and 24-weeks, the wide confidence interval (CI) yield it clinically insignificant. CMT reductions were also comparable across the three arms. These findings reveal that there is no independent correlation between anatomical change (CMT) and functional change (BCVA). Our meta-regression for injections showed no favorable overlap for both BCVA and CMT, which could indicate that no single regimen can guarantee both increase in VA and decrease in CMT at the same time. But this should not be the case at certain specific ranges and injections. The double interaction regression between the two outcomes VA & CMT with subgroup consideration revealed that improvement in VA was best associated with CMT reduction from 480um to 320um. This association was significant at 12-weeks in the three arms and persisted till 24-weeks and 48-weeks exclusively in the IVB group. This further solidifies the multifactorial idea that age, hemoglobin A1C level, diabetes duration, insulin usages, and degree of retinopathy proliferation are all responsible for the change in VA and CMT [25,46]. Moreover, the different degrees of macular ischemia could explain why some patients have no significant improvement in vision despite the reduction of thickness.
Attempting to analyze this multivariate pathogenesis, we considered performing additional meta-regression analysis. The duration of DM, insulin usages, levels of HbA1C, and HTN were all inversely associated with visual outcomes. According to this regression, the type of patient who responds best on treatment is a diabetic female with mild or no HTN with a short history of DM-II <10 years, HbA1C <8%, and low or no insulin intake. The interaction regression of this combination yields an R 2 of 100%. Other factors like age, and degree of retinopathy proliferation do not seem to affect the outcomes as much.
Our findings exhibit a favorable response to IVB compared with IVT or IVB+IVT in improving VA up to 48-weeks. The reason why this difference did not persist with the single injection can be attributed to the limited effective duration of these injections. At first glance, the statistical insignificance between the two drugs regarding CMT reduction may indicate an equivalent share within VEGF angiogenesis and inflammatory transduction proposed mechanisms. However, the correlation between the two mechanisms appears to be non-linear, as the combined IVB+IVT also presented no statistical difference. This finding could either indicate that antagonizing multiple mechanisms simultaneously may lead to more resistance and less improvements, or that another unclear balancing factor could be compromised by this combination. Either way, this critical relation needs further investigation in the future pathological and pharmacological studies. Still, the 12-weeks improvement of VA surpasses the traditional pharmacological data that estimated a single IVB injection as effective only for 6 weeks [47,48]. Also, it defies the prevailing assumption that an IVT injection better improves VA in the first 12 weeks of follow-up; and the presumption that a single IVT injection can be comparable to three IVB injections [20,49,50]. Some reports extend this even further, in a 96 weeks follow-up: Soheilian et al. 2012 reported a significant superiority of the IVB over the combined IVB+IVT up to 24 weeks [31].
The majority of our included studies used a standard dose of 1.25mg/0.05ml for IVB, 4mg/ 0.05ml for IVT, 1.25+2mg/0.05ml for IVB+IVT. Nine studies considered repeated injections and retreatments at different intervals, ranging from 4-12 weeks. Meta-regression revealed that repeated successive injections associate with better VA, and the range of 1.5-4.8 injection predicts more promising improvement in VA and reduction in CMT. This further supports the idea of dose-response proportional efficacy; but still promotes the idea of fewer injections as possible, to guarantee a lower incidence of injection-related complications such as endophthalmitis, high IOP, and weak patient compliance [20,51]. A possible explanation for why the efficacy is not better in over 4.8 injections could be due to the fact that some eyes show low-response to treatment with regard to VA gain and CMT reduction, as pointed out by Menke et al.; 30% of patients showed low response to ranibizumab after 4 weeks [52].
A higher level of intraocular VEGF is considered a major pathogenic factor in DME, and a contributor to the increased IOP [53,54]. Thus, measuring the long-term change in the IOP is critical in the assessment of drugs' safety and efficacy. In our findings: the group who received IVB had a statistically significant lower IOP than the group who received IVT at 36 and 48 weeks follow-up. Additionally, the incidence of IOH was significantly lower in the IVB group in comparison with IVT or IVB+IVT. This effect seems to persist even with different IVB and IVT injections. For instance, Rodrigues et al. considered 10-13 repeated injection of 1.25mg/ 0.05ml IVB and 7-9 repeated injection of 1.20mg/0.03ml IVT; Lim et al. considered repeated injection of IVB only with no repetition of IVT -even in the combination arm, and Kriechbaum et al. considered 3 injections of 2.5mg IVB and 8mg IVT [24,34,45]. Their results were consistent with our findings, together with the results of the previous meta-analysis of Zhang et al.
Meanwhile, the effect of combined laser with IVB or IVT is still unclear. Pappas et al. concluded that IVT+Laser is better than IVB alone in reducing CMT, which went against the findings of Lam et al. who indicated that IVT+Laser has no better CMT reduction at 24 weeks than IVT alone; and laser alone was significantly worse than the 2 aforementioned groups [55,56]. Surprisingly, Preti et al. concluded that IVB+Laser may significantly increase the thickness compared with laser alone at a 4-week follow-up, however, they investigated macular choroidal thickness and not CMT [57]. It was proposed that IVT+laser might be superior due to its coupled anti-inflammatory and anti-angiogenic effects [13]. However, this could only affect the CMT reduction, but it does not guarantee an improvement in VA. Also, no cumulative analysis of these trials can yield statistical significance. They suffer from a critical degree of bias and heterogeneity: no description of allocation concealment, blinding of participants, or blinding of outcome assessment were reported in the work of Pappas et al. Comparatively, a high risk of blinding of participants was found in the report of Lam et al; a relatively small sample size, very short follow-up and unjustified inclusion of patients with and without DME were detected in the work of Preti et al. Moreover, clinicians should reconsider the IVB and IVT combination with laser for a further investigation.
The quality of a systematic review and meta-analysis rests upon the qualities of its included studies. Our included studies exhibit relatively high quality. Our findings settle a group of assumptions and provide a reliable reference for future clinical decisions. To our knowledge, this is the first systematic review and meta-analysis that analyzes the long-term outcomes of IVB and IVT over 48 weeks and provide a meta-regression for injections, pathogenesis, and interaction between VA & CMT. Even so, there were some limitations to our work. The results in CMT at 6, 12, 24, 48 weeks were limited by the heterogeneity of the included studies. Likewise, only two studies investigated IVT vs. IVB/IVT, making it difficult to conclude anything definitive on this aspect. The variations in the clinical definitions and subtypes of DME may contribute to the clinical heterogeneity. Additionally, most of the included trials had relatively small sample sizes.
Our analysis reveals that repeated IVB injections associate with better VA. Overall, the current evidence indicates that IVB is superior to IVT and IVB+IVT in improving VA for DME patients up to 48 weeks. The combined IVB+IVT does not seems to be more promising or beneficial. CMT reduction appears to be comparable across the three arms (IVB vs. IVT, IVB vs. IVB+IVT, and IVT vs. IVB+IVT) in short and long-terms. It is in favor of IVB when it comes to the significant reduction of IOP and the avoidance of IOH compared to IVT and IVB+IVT. Further multi-center, large sample RCTs are needed to investigate the efficacy of laser photocoagulation combined with IVB versus laser combined with IVT. Future studies should consider accurate reporting of pathogenic markers across groups and subgroups to allow for a better understanding of DME's underlying pathogenesis. The overall meta-regression mean difference of the interaction between each pathogenic factor on x-axis and CMT on y-axis. The diamond indicates significant prediction. B. The overall meta-regression mean difference of the interaction between each pathogenic factor on x-axis and BCVA on y-axis. The diamond indicates significant prediction. C. The overall meta-regression mean difference of the interaction between each pathogenic factor on x-axis and IOP on y-axis. The diamond indicates significant prediction. D. The overall meta-regression mean difference of the interaction between number of injections on x-axis and each outcome on y-axis. The diamond indicates significant prediction.