Anti‐vascular endothelial growth factor for diabetic macular oedema: a network meta‐analysis

Summary of findings for the main comparison. Antiangiogenic therapy versus control

Antiangiogenic therapy versus control
Patient or population: people with diabetic macular oedema Settings: ophthalmology clinics Interventions: laser photocoagulation, aflibercept, bevacizumab, ranibizumab
Outcomes	Assumed risk*	Corresponding risk and relative risk (95% CI) , mixed evidence**			Certainty of evidence and reason for downgrading
Outcomes	Laser photocoagulation	Aflibercept	Bevacizumab	Ranibizumab	Certainty of evidence and reason for downgrading
Gain 3+ lines of visual acuity at 1 year	100 per 1000	366 per 1000 (279 to 479) RR: 3.66 (2.79 to 4.79)	247 per 1000 (181 to 337) RR: 2.47 (1.81 to 3.37)	276 per 1000 (212 to 359) RR: 2.76 (2.12 to 3.59)	⊕⊕⊕⊕ high
Visual acuity change at 1 year Measured on the logMAR scale, range −0.3 to 1.3. Higher values represent worse visual acuity.	On average visual acuity improved by −0.01 logMAR units in the laser group between the start of treatment and 1 year (effectively no change)	Average change in visual acuity was −0.20 (−0.22 to −0.17) logMAR units better with aflibercept compared with laser photocoagulation	Average change in visual acuity was −0.12 (−0.15 to −0.09) logMAR units better with bevacizumab compared with laser photocoagulation	Average change in visual acuity was −0.12 (−0.14 to −0.10) logMAR units better with ranibizumab compared with laser photocoagulation	⊕⊕⊕⊕ high for aflibercept and ranibizumab ⊕⊕⊕ moderate for bevacizumab (−1 for inconsistency of indirect versus direct evidence)
Central retinal thickness μm (CRT) change at 1 year The aim of treatment is to reduce central retinal thickness so thinner is better.	On average CRT changed by −64 μm in the laser group between the start of treatment and 1 year (became thinner)	Average change in CRT was −114 (−147 to −81) μm more (thinner) with aflibercept compared with laser photocoagulation	Average change in CRT was −46 (−78 to −14) μm more (thinner)with bevacizumab compared with laser photocoagulation	Average change in CRT was −75 (−100 to −50) μm more (thinner) with ranibizumab compared with laser photocoagulation	⊕⊕⊕⊕ high
Quality of life: NEI‐VFQ composite score at 6 to 12 months An improvement by 5 units is clinically significant.	On average the composite score improved by +2 units in the laser group between the start of treatment and 6 to 12 months			Average change in composite score was 5.14 (2.96 to 7.32) with ranibizumab compared with laser photocoagulation	⊕⊕⊕ moderate (−1 for risk of bias)
All serious systemic adverse events at 1 to 2 years	200 per 1000	196 per 1000 (166 to 232) RR: 0.98 (0.83 to 1.16)	186 per 1000 (146 to 238) RR: 0.93 (0.73 to 1.19)	194 per 1000 (160 to 234) RR: 0.97 (0.80 to 1.17)	⊕⊕⊕⊕ high
Arterial thromboembolic events at 1 to 2 years	45 per 1000	38 per 1000 (16 to 94) RR: 0.88 (0.37 to 2.13)	41 per 1000 (15 to 117) RR: 0.94 (0.33 to 2.66)	48 per 1000 (23 to 101) RR: 1.09 (0.52 to 2.29)	⊕⊕ low (−2 for imprecise estimates)
Death at 1 to 2 years	20 per 1000	20 per 1000 (7 to 61) RR: 1.01 (0.34 to 3.03) a	32 per 1000 (9 to 114) RR: 1.61 (0.45 to 5.69)	18 per 1000 (8 to 40) RR: 0.90 (0.40 to 2.01)	⊕⊕ low for bevacizumab and aflibercept (−2 for imprecise estimates) ⊕ very low for aflibercept (additional −1 direct evidence inconsistent, higher risk)
The assumed risk in the laser group was estimated as the row sum of the events divided by the row sum of the participants (eyes) for dichotomous variables, and as the (unweighted) median change of visual acuity or central retina thickness. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). The risk ratio was estimated from mixed (direct and indirect) comparisons. CI: Confidence interval; RR:** Risk ratio.
GRADE Working Group grades of evidence High‐certainty: further research is very unlikely to change our confidence in the estimate of effect. Moderate‐certainty: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low‐certainty: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low‐certainty: we are very uncertain about the estimate.

Summary of findings 2. Ranibizumab versus aflibercept for diabetic macular oedema

Ranibizumab versus aflibercept for diabetic macular oedema
Patient or population: people with diabetic macular oedema Settings: ophthalmology clinics Interventions: aflibercept, ranibizumab
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI), mixed evidence**	Certainty of the evidence (GRADE)	Reason for downgrading certainty of evidence
	Assumed risk	Corresponding risk
	Aflibercept	Ranibizumab
Gain 3+ lines of visual acuity at 1 year	370 per 1000	278 per 1000 (222 to 348)	RR: 0.75 (0.60 to 0.94)	⊕⊕⊕ moderate	−1 for imprecision as confidence intervals include both clinically important and clinically unimportant effects
Visual acuity change at 1 year Measured on the logMAR scale, range −1.3 to 1.3. Higher values represent worse visual acuity.	On average visual acuity improved by−0.23 logMAR units in the aflibercept group between the start of treatment and 1 year	Average change in visual acuity was 0.08 (0.05 to 0.11) logMAR units worse with ranibizumab compared with aflibercept		⊕⊕⊕ moderate	−1 for imprecision as confidence intervals include both clinically important and clinically unimportant effects
Central retinal thickness μm (CRT) change at 1 year The aim of treatment is to reduce central macular thickness so thinner is better.	On average CRT changed by −181 μm in the aflibercept group between the start of treatment and 1 year (became thinner)	Average change in CRT was 39 (2 to 76) μm more (thicker) with ranibizumab compared with aflibercept		⊕⊕ low	−1 for high heterogeneity in two direct comparisons and large predictive intervals −1 for imprecision
Quality of life at 1 year	No data available.
All serious systemic adverse events at 1 to 2 years	345 per 1000	338 per 1000 (283 to 411)	RR 0.98 (0.82 to 1.19)	⊕⊕⊕⊕ high
Arterial thromboembolic events at 1 to 2 years	60 per 1000	74 per 1000 (29 to 191)	RR 1.24 (0.48 to 3.19)	⊕ very low	Inconsistency between direct and indirect evidence (−1), and imprecise estimates (−2)
Death at 1 to 2 years	30 per 1000	35 per 1000 (11 to 108)	RR 1.16 (0.38 to 3.58)	⊕ very low	Inconsistency between direct and indirect evidence (−1), and imprecise estimates (−2)
The assumed risk in the aflibercept group was estimated as the row sum of the events divided by the row sum of the participants (eyes) for dichotomous variables, and as the (unweighted) median change of visual acuity or central retina thickness. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). The risk ratio was estimated from mixed (direct and indirect) comparisons. CI: Confidence interval; RR:** Risk ratio.
GRADE Working Group grades of evidence High‐certainty: further research is very unlikely to change our confidence in the estimate of effect. Moderate‐certainty: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low‐certainty: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low‐certainty: we are very uncertain about the estimate.

See: Characteristics of included studies; Characteristics of excluded studies; Characteristics of ongoing studies and Characteristics of studies awaiting classification.

Summary of findings 3. Ranibizumab versus bevacizumab for diabetic macular oedema

Ranibizumab versus bevacizumab for diabetic macular oedema
Patient or population: people with diabetic macular oedema Settings: ophthalmology clinics Interventions: bevacizumab, ranibizumab
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI), mixed evidence**	Certainty of the evidence (GRADE)	Reason for downgrading certainty of evidence
	Assumed risk	Corresponding risk
	Bevacizumab	Ranibizumab
Gain 3+ lines of visual acuity at 1 year	300 per 1000	333 per 1000 (261 to 429)	RR 1.11 (0.87 to 1.43)	⊕⊕⊕ moderate	Imprecise estimate (−1)
Visual acuity change at 1 year Measured on the logMAR scale, range −1.3 to 1.3. Higher values represent worse visual acuity.	On average visual acuity improved by −0.19 logMAR units in the bevacizumab group between the start of treatment and 1 year	Average change in visual acuity was0.00 (−0.02 to 0.03) logMAR units (same) with ranibizumab compared with bevacizumab		⊕⊕⊕ moderate	Unclear risk of bias (−1)
Central retinal thickness (CRT) change at 1 year The aim of treatment is to reduce central macular thickness so thinner is better.	On average CRT changed by −98 μm in the bevacizumab group between the start of treatment and 1 year (became thinner)	Average change in CRT was −29 (−58 to −1) μm more (thinner) with ranibizumab compared with bevacizumab		⊕⊕ low	Unclear risk of bias (−1) Imprecise estimate (−1)
Quality of life at 1 year	No data available
All serious systemic adverse events at 1 to 2 years	240 per 1000	250 per 1000 (202 to 307)	RR 1.04 (0.84 to 1.28)	⊕⊕⊕ moderate	Unclear risk of bias (−1)
Arterial thromboembolic events at 1 to 2 years	60 per 1000	70 per 1000 (26 to 189)	RR 1.17 (0.43 to 3.13)	⊕ very low	Unclear risk of bias (−1) Imprecise estimate (−2)
Death at 1 to 2 years	40 per 1000	29 per 1000 (9 to 95)	RR 0.73 (0.22 to 2.37)	⊕ very low	High risk of bias (−2) Imprecise estimate (−2)
The assumed risk in the bevacizumab group was estimated as the row sum of the events divided by the row sum of the participants (eyes) for dichotomous variables, and as the (unweighted) median change of visual acuity or central retina thickness. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). The risk ratio was estimated from mixed (direct and indirect) comparisons. CI: Confidence interval; RR:** Risk ratio.
GRADE Working Group grades of evidence High‐certainty: further research is very unlikely to change our confidence in the estimate of effect. Moderate‐certainty: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low‐certainty: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low‐certainty: we are very uncertain about the estimate.

Background

Description of the condition

Diabetic retinopathy (DR) is the most frequent and severe ocular complication of diabetes mellitus (DM) and the leading cause of blindness in the working age population in developed countries (Frank 2004; Klein 1984; Tranos 2004).

Diabetic macular oedema (DMO) is the swelling of the retina resulting from the exudation and accumulation of extracellular fluid and proteins in the macula (Ciulla 2003), due to the breakdown of the blood‐retina barrier with an increase in vascular permeability (Antcliff 1999). About a third of people with diabetes have DR and one in 10 is affected by DMO (Yau 2012). The prevalence of DMO increases with diabetes duration, haemoglobin A1c, and blood pressure levels and is higher in people with type 1 compared with type 2 diabetes (Yau 2012).

Intraretinal fluid accumulation results in significant reduction in visual acuity that may be reversible in the short term, but prolonged oedema can cause irreversible damage resulting in permanent visual loss. Blurred vision represents the most common clinical symptom of DMO. Other symptoms can include metamorphopsia (distortion of visual image), floaters, changes in contrast sensitivity, photophobia (visual intolerance to light), changes in colour vision and scotomas (a localised defect of the visual field).

During the last decades, the clinical gold standard to detect macular oedema has been fundus examination with contact lens, but non‐contact lenses can also be used for this purpose with good sensitivity. Optical coherence tomography (OCT) has progressively been used as an objective and reproducible tool to measure retinal thickness and has been suggested to be the new gold standard for diagnosing DMO (Olson 2013; Ontario HTA 2009). The most severe form of DMO is CSMO, which was defined by the Early Treatment Diabetic Retinopathy Study (ETDRS) as: retinal oedema within 500 µm of the centre of the fovea; hard exudates within 500 µm of the centre of the fovea, if associated with adjacent retinal thickening (which may be outside the 500 µm limit); and one disc area of retinal oedema (1500 µm) or larger, any part of which is within one disc diameter of the centre of the fovea (ETDRS 1985). Since its introduction, OCT was found to be in good agreement with the clinical gold standard (slit‐lamp examination with a contact lens) for detecting the presence of macular oedema and was found to be potentially more sensitive in cases of mild foveal thickening (Brown 2004). A simple OCT‐based classification of DMO is often used as centre‐involving or non‐centre‐involving DMO (Browning 2008).

Description of the intervention

Antiangiogenic therapy has been believed a standard of care for treatment of DMO and has largely replaced laser photocoagulation (Jampol 2014), than which it was proven to be more effective (Virgili 2014). Anti‐vascular endothelial growth factor (anti‐VEGF) treatments inhibit VEGF angiogenic activity, binding to VEGF protein and thus preventing its receptor activation or interaction. These drugs were originally hypothesised as an alternative adjunctive treatment for DMO (Cunningham 2005), following evidence that VEGF‐A plays a key role in the occurrence of increased vascular permeability in ocular diseases such as DMO (Aiello 2005).

Grid or focal laser photocoagulation could not be used in all patients with DMO; thus, either laser or sham procedures were current practice comparators in initial studies on the efficacy of antiangiogenic drugs for DMO (Macugen 2005; RESOLVE 2010; RESTORE 2011; Soheilian 2007), and only recently have directly comparative RCTs been conducted (DRCRnet 2015).

Safety of intravitreal antiangiogenic therapy is acceptable; endophthalmitis, the major adverse event (< 1/1000 injections) is related to the surgical injection procedure, rather than the drug itself. These drugs were shown not to increase systemic adverse events such as arterial thromboembolic events, but differences between drugs are not well known (Virgili 2014).

Another therapeutic option for DMO treatment is represented by steroids, administered as intravitreal injections or sustained release implants in order to obtain high local concentrations, maximising their anti‐inflammatory, angiostatic and anti‐permeability effects while minimising systemic toxicity (Ciulla 2004; Haller 2010; Kuppermann 2010). However, intravitreal steroids may cause cataract and ocular hypertension and the visual outcome is dependent on the lens status or the need for cataract surgery after about one year (Haller 2010; Campochiaro 2010). Currently, some investigators think intravitreal steroids are preferred in patients with anti‐VEGF resistant and chronic DMO (Hussain 2015), as an alternative to switching between anti‐VEGF drugs. This is also consistent with the EU label of the only approved dexamethasone intravitreal implant in Europe: "Ozurdex is indicated for the treatment of adult patients with visual impairment due to diabetic macular oedema (DME) who are pseudophakic or who are considered insufficiently responsive to, or unsuitable for non‐corticosteroid therapy" (accessed on EMA on 4 December 2016).

For ranibizumab, the EU label prescribes a 0.5 mg dosage, and that "treatment is initiated with one injection per month until maximum visual acuity is achieved and/or there are no signs of disease activity i.e. no change in visual acuity and in other signs and symptoms of the disease under continued treatment. In patients with wet AMD, DME and RVO, initially, three or more consecutive, monthly injections may be needed. Thereafter, monitoring and treatment intervals should be determined by the physician and should be based on disease activity, as assessed by visual acuity and/or anatomical parameters" (accessed on EMA on 4 December 2016). In the USA, ranibizumab "0.3 mg is recommended to be administered by intravitreal injection once a month (approximately 28 days)" (accessed on FDA on 4 December 2016).

Aflibercept has been approved in the USA, as accessed on FDA on 4 December 2016, and "the recommended dose for EYLEA is 2 mg (0.05 mL) administered by intravitreal injection every 4 weeks (monthly) for the first 5 injections followed by 2 mg (0.05 mL) via intravitreal injection once every 8 weeks (2 months)". The EU label is similar (accessed on EMA on 4 December 2016).

Bevacizumab is widely used off‐label although its use has been questioned based on regulatory or safety issues (Banfi 2013), but is still key for treating chorioretinal vascular disease in low‐ and middle‐income countries thanks to its low cost (Stewart 2016).

How the intervention might work

VEGF plays a key role in the occurrence of increased vascular permeability in ocular diseases such as DMO (Aiello 2005). Anti‐VEGF agents inhibit VEGF angiogenic activity, binding to VEGF protein thus preventing its receptor activation and interaction.

Why it is important to do this review

DMO results in a significant burden of low vision and blindness, thus the extent of the existing evidence base for the effectiveness and safety of these agents needs to be assessed and updated. There is a continuing clinical need to establish evidence‐based recommendations regarding anti‐VEGF agents.

Objectives

The 2014 update of this review found high‐quality evidence of benefit with antiangiogenic therapy with anti‐VEGF modalities, compared to laser photocoagulation, for the treatment of DMO. As was concluded in the previous version (Virgili 2014), the objective of this updated review is to compare the effectiveness and safety of the different anti‐VEGF drugs in preserving and improving vision and quality of life using network meta‐analysis methods.

Methods

Criteria for considering studies for this review

Types of studies

We included randomised controlled trials (RCTs).

Types of participants

People with DMO for whom anti‐VEGF treatment is indicated. We expected to include most of the studies also included in Virgili 2014.

Types of interventions

Any antiangiogenic drug with anti‐VEGF modalities compared with another drug with anti‐VEGF modalities, laser treatment, sham treatment or no treatment. The reasons for selecting treatments of direct and indirect treatment have been discussed in the Description of the intervention section. As explained above, we remark that steroids may be compared with anti‐VEGF drugs but this needs a different approach, specifically patient subgroups and timing, and their inclusion could lead to violation of similarity in a review aiming to compare different anti‐VEGF drugs such as this.

Regarding drug dose and monitoring/retreatment regimen, in efficacy analyses we included schemes that are either on‐label or commonly used in clinical practice, such as the PRN regimen, as presented in the Description of the intervention section. Particularly, both 0.3 mg and 0.5 mg ranibizumab dose are included as available in studies. These two ranibizumab doses were merged into one group in our NMA since studies suggest no difference between them when used monthly (Heier 2016). Regarding aflibercept, we selected the bi‐monthly retreatment regimen since this is the approved label in the USA. We used all available data regardless of safety and dose for safety analyses as previously done in Moja 2014.

Types of outcome measures

Primary outcomes

Best‐corrected visual acuity (BCVA) expressed as the proportion of participants with at least 15 ETDRS letters (3 ETDRS lines or 0.3 logMAR) of improvement in BCVA from baseline to 12 months.

Secondary outcomes

Mean change in BCVA from baseline to 12 months, measured using ETDRS charts.
Mean change in central retinal thickness (CRT), from baseline to 12 months, measured using optical coherence tomography (OCT).
Mean change in quality of life from baseline to 12 months, measured using a validated instrument.

Measurements at varying lengths of follow‐up were pooled at annual intervals, plus or minus six months, the primary analysis being that at 12 months. The time point closer to 12 months, or the latest time point in the window frame in the case of symmetry, was chosen where multiple time points were available.

Adverse events

The following adverse events were considered.

All‐cause mortality.
Arterial thromboembolic events (ATC 1994).
Systemic serious adverse events (SSAEs).

Adverse events were analysed at the longest available follow‐up time (Moja 2014).

Search methods for identification of studies

Electronic searches

The Cochrane Eyes and Vision Information Specialist conducted systematic searches in the following databases for randomised controlled trials and controlled clinical trials. There were no language or publication year restrictions. The date of the search was 26 April 2017.

Cochrane Central Register of Controlled Trials (CENTRAL; 2017, Issue 3) (which contains the Cochrane Eyes and Vision Trials Register) in the Cochrane Library (searched 26 April 2017) (Appendix 1);
MEDLINE Ovid (1946 to 26 April 2017) (Appendix 2);
Embase Ovid (1980 to 26 April 2017) (Appendix 3);
LILACS (Latin American and Caribbean Health Science Information database (1982 to 26 April 2017) (Appendix 4);
ISRCTN registry (www.isrctn.com/editAdvancedSearch; searched 26 April 2017) (Appendix 5);
US National Institutes of Health Ongoing Trials Register ClinicalTrials.gov (www.clinicaltrials.gov; searched 26 April 2017) (Appendix 6);
World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP) (www.who.int/ictrp; searched 26 April 2017) (Appendix 7).

Searching other resources

We handsearched the reference lists of the included trials for other possible trials. We accessed the Novartis Clinical Trials database (www.novctrd.com/ctrdWebApp/clinicaltrialrepository/public/main.jsp) on 28 May 2014 and checked all trials indexed under the headings: Ophthalmic Disorders and ranibizumab.

Data collection and analysis

Selection of studies

Two review authors independently selected the studies for inclusion. The titles and abstracts of all reports identified by the electronic searches and handsearching were examined by the review authors. We classified the abstracts as (a) definitely include, (b) unsure and (c) definitely exclude. We obtained and re‐assessed full‐text copies of those classified as either (a) definitely include or (b) unsure. Having reviewed the full‐text copies, we classified the studies as (1) included, (2) awaiting assessment and (3) excluded. Studies identified by both review authors as 'excluded' were excluded and documented in the review. Studies identified as 'included' were included and assessed for methodological quality. The review authors were unmasked to the report authors, institutions and trial results during this assessment. Disagreements between the two review authors were resolved by a third review author.

Data extraction and management

Two review authors independently extracted the data for the primary and secondary outcomes onto paper data extraction forms developed by the Cochrane Eyes and Vision Group. A pilot test of this form was carried out using a small number of studies. We resolved discrepancies by discussion. One review author entered all data into Review Manager 5 (Review Manager 5 2014). The entered data were checked by a second author. In case standard deviations were not available in the publication, and could not be obtained from the authors, these were imputed from standard deviations of other studies with the same comparison.

Assessment of risk of bias in included studies

Two review authors independently assessed the included trials for bias according to the methods described in Chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011b). The following parameters were assessed: sequence generation; allocation concealment; masking (blinding) of participants, personnel and outcome assessors; incomplete outcome data; selective outcome reporting. We evaluated these parameters for each outcome measure or class of outcome measure. We classified each parameter as low risk of bias, high risk of bias or unclear.

If the information available in the published trial reports was inadequate to assess methodological quality, we contacted the trial authors for clarification. We had planned that if they did not respond within six months we would assess the trial based on the available information. However, in the latest update of this review we assessed the trial had the authors not responded within one month.

We followed Salanti 2014 to assess the risk of bias of mixed evidence (mixed evidence not defined previously).

Summary risk of bias for each trial: we considered all domains but gave more importance to allocation concealment and masking of outcome assessor.
Summary risk of bias for the mixed evidence: based on the percentage contribution of each direct comparison to each network estimate using the contribution plot (Chaimani 2013).

We finally integrated the risk of bias of a given comparison with the assessment of transitivity, or similarity of the characteristics of the studies. We expected the transitivity assumption would hold as long as treatment comparisons were not related to:

acute versus chronic DMO, defined using the cut‐off of three or more years of duration;
average severity of DMO using OCT CRT of 400 micrometres as a cut‐off;
treatment regimen, such as monthly versus less than monthly and number of injections in the first year;
drug dose for ranibizumab, since this is commercially available in two doses (0.3 mg in the USA, 0.5 mg otherwise);
whether the trial was industry sponsored.

Measures of treatment effect

Data analysis followed the guidelines set out in Chapter 9 of the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2011). For dichotomous outcomes, we calculated a summary risk ratio (RR). For continuous outcome, we calculated the mean difference (MD). We planned to calculate a standardised mean difference (SMD) had different scales been used to measure the same continuous outcome.

We did not use ranking measures in this review, since our main interest was to compare only three drugs: aflibercept, bevacizumab and ranibizumab.

Unit of analysis issues

The unit of randomisation was the eye of individual participants. We included one cross‐over study comparing ranibizumab and bevacizumab and treated this as a parallel arm study (Wiley 2016), which equals to assume a moderate (0.5) correlation within‐person. However, relative drug safety is impossible to assess with a paired design.

We accepted studies presenting systemic adverse events as the unit of analyses, i.e. when an individual suffers from more than one severe adverse event in the study.

Dealing with missing data

Where data were missing due to dropping out of participants, we conducted a primary analysis based on participants with complete data (available case analysis). Following the guidance available in Chapter 16 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011a), we considered that missing outcome data are missing at random if the reasons for loss to follow‐up are documented and judged to be unrelated to outcome in both study arms.

Assessment of reporting biases

To investigate small‐study bias at the network level we employed the comparison‐adjusted funnel plot, which is an adaptation of the funnel plot. We subtracted from each study‐specific effect size the mean of meta‐analysis of the study‐specific comparison and plotted it against the study’s standard error (Chaimani 2013).

Data synthesis

Methods for direct treatment comparisons

If there was no substantial statistical heterogeneity, and if there was no clinical heterogeneity between the trials, we combined the results in a meta‐analysis using a random‐effects model. A fixed‐effect model was used if the number of trials was three or less. In the case of substantial statistical heterogeneity (that is I² value more than 50%) or clinical heterogeneity, we combined the results in a meta‐analysis using a random‐effects model if the individual trial results were all consistent in the direction of the effect (that is the RR or MD and confidence intervals largely fall on one side of the null line); when the individual trial results were inconsistent in the direction of the effect, we did not combine study results but presented a narrative or tabulated summary of each study.

Methods for indirect and mixed comparisons

We performed network meta‐analysis using the methodology of the multivariate meta‐analysis model where different treatment comparisons are treated as different outcomes (Salanti 2012). For this analysis, we used the 'network' suite of commands available in STATA (StataCorp, 2011; Stata Statistical Software: Release 14. College Station, TX) (White 2015).

We presented mixed effects as RRs or MDs against laser photocoagulation as a single comparison. We prepared league tables presenting mixed comparisons in the inferior‐left part and direct comparisons in the superior‐right part of the table in order to allow for the inspection of both types of evidence. The same information was presented graphically. We also presented the contribution of direct and indirect evidence to mixed evidence using contribution plots (Chaimani 2013).

Assessment of statistical heterogeneity

In standard pairwise meta‐analyses, we estimated heterogeneity variances for each direct comparison. We assessed statistically the presence of heterogeneity within each pairwise comparison using the I² statistic (Higgins 2011b). The I² statistic measures the percentage of variability that cannot be attributed to random error. In network meta‐analysis, we assumed a common estimate for the heterogeneity variance across the different comparisons. The assessment of statistical heterogeneity in the entire network was based on the magnitude of the heterogeneity variance parameter (τ²) estimated from the network meta‐analysis models.

Assessment of statistical inconsistency

Local approaches for evaluating inconsistency

To evaluate the presence of inconsistency locally, we used the node‐splitting approach (Dias 2010). We assumed a common heterogeneity estimate within each loop.

Global approaches for evaluating inconsistency

To check the assumption of consistency in the entire network, we used the 'design‐by‐treatment' model using the 'network' command in STATA (White 2015). This method accounts for different sources of inconsistency that can occur when studies with different designs (two‐arm trials versus three‐arm trials) give different results as well as disagreement between direct and indirect evidence. Using this approach, we judged the presence of inconsistency from any source in the entire network based on a Chi² test.

'Summary of findings' table and GRADE assessment

We prepared one 'Summary of findings' table for each relevant comparison, including all seven outcomes in a table (GRADEpro 2014). As originally intended, the primary analysis was conducted at 12 months. Relevant comparisons were identified to answer the question of which antiangiogenic drug is most effective among on‐label (aflibercept, ranibizumab) and off‐label (bevacizumab) drugs that are currently available. Because most of the available evidence is around ranibizumab, we reported on the comparison of aflibercept and bevacizumab with ranibizumab. Analyses conducted at 24 months were presented textually because a network meta‐analysis was not feasible.

We graded the certainty of the evidence for mixed estimates as explained above. We started from the premise that RCTs provide high‐certainty evidence and downgraded for each GRADE parameter to get an overall certainty for each outcome as high, moderate, low or very low (Higgins 2014; Salanti 2014; Schünemann 2011). We estimated the absolute risk in the control group from the data in the included studies as the raw proportion with event for dichotomous outcomes and the median value for continuous outcomes.

Sensitivity and subgroup analyses

We had not planned sensitivity analyses but we decided post‐hoc to conduct one excluding studies which were assessed as being at overall high or unclear risk of bias. Moreover, we acknowledge that DRCRnet 2015 emphasised that the differences in absolute benefit between aflibercept, bevacizumab and ranibizumab at one year were dependent on baseline visual acuity, but when we considered this post hoc subgroup analysis we did not find enough study data to conduct such meta‐regression analyses.

Results

Description of studies

The previous version of this review included 18 trials. Update searches run in April 2017 yielded a further 1166 records (Figure 1). After 397 duplicates were removed, the Cochrane Information Specialist screened the remaining 769 records and removed 577 references that were not relevant to the scope of the review. We screened the remaining 192 references and obtained 19 full‐text reports for further assessment. We identified eight reports of six new trials for inclusion in the review (DRCRnet 2015, Ishibashi 2014, Lopez‐Galvez 2014, REVEAL 2015, Turkoglu 2015, Wiley 2016). A further four trials were deemed eligible but did not provide sufficient data for analysis (Chen 2016; Huang 2016; Jovanovic 2015;Fouda 2017. We have contacted these authors to ask for further information and will assess these studies if we receive additional data. We excluded one study (NCT02985619 (BEVATAAC)) and have identified six new ongoing studies and will assess these for inclusion in the review when data becomes available (NCT02194634;NCT02259088; NCT02348918; NCT02645734; NCT02699450; NCT02712008).

Figure 1

Study flow diagram.

Included studies

We included a total of 24 studies in this updated systematic review and network meta‐analysis. BOLT 2010, DA VINCI 2011, Ishibashi 2014, Korobelnik 2014, Macugen 2005, Macugen 2011, READ2 2009, RELATION 2012, RESOLVE 2010, RESPOND 2013, RESTORE 2011, and RISE‐RIDE were industry‐sponsored, multicentre RCTs conducted in the USA or Europe, whereas REVEAL 2015 was industry‐sponsored but conducted in Asia. Ahmadieh 2008, Azad 2012, Ekinci 2014, LUCIDATE 2014, Nepomuceno 2013, Soheilian 2007, and Turkoglu 2015 were independent studies conducted in Brazil, India, Iran, Turkey, and the UK, five of which included bevacizumab. DRCRnet 2010, DRCRnet 2015, Wiley 2016 were publicly‐sponsored studies, mainly by the US National Eye Institute, and conducted in the USA or UK. DRCRnet 2015 was the only large parallel‐arm study that compared all commercially available drugs (aflibercept, bevacizumab, ranibizumab) and was a large publicly‐funded trial comparing aflibercept, bevacizumab and ranibizumab with monthly monitoring and treatment as needed (PRN). Wiley 2016 was a cross‐over trial comparing the same three drugs. Lopez‐Galvez 2014 was an open‐label trial comparing ranibizumab with laser; it was conducted in Spain and results were available only in abstract form.

Only six trials maintained the randomisation scheme at two years' follow‐up (BOLT 2010; DRCRnet 2010; DRCRnet 2015; Macugen 2011; READ2 2009; RISE‐RIDE). Two industry‐sponsored trials used randomisation up to two years (Macugen 2011; RISE‐RIDE), while three others obtained follow‐up data but allowed anti‐VEGF treatment in the control arm after one year (Korobelnik 2014; RESOLVE 2010; RESTORE 2011).

We did not extract data on comparisons of antiangiogenic therapy with triamcinolone and other intravitreal steroids, which were study arms in Ahmadieh 2008, Azad 2012, DRCRnet 2010 and Soheilian 2007, for reasons presented above and also because this comparison is the subject of another Cochrane Review (Grover 2008). Standard deviations of change in CRT were imputed from other studies in REVEAL 2015.

Types of participants

Trials included participants with DMO diagnosed clinically, and often these trials used OCT for confirming macular centre involvement. Baseline visual acuity of participants was generally between 20/200 and 20/40. Most trials required a three‐ to six‐month interval from previous central or peripheral laser, and a few small studies required that participants had not received previous antiangiogenic treatment.

Types of interventions

Eleven studies assessed ranibizumab (DRCRnet 2010; Lopez‐Galvez 2014; LUCIDATE 2014; READ2 2009; RELATION 2012; RESOLVE 2010; RESPOND 2013; RESTORE 2011; REVEAL 2015; RISE‐RIDE; Turkoglu 2015), six investigated bevacizumab (Ahmadieh 2008; Azad 2012; BOLT 2010; Ekinci 2014; Nepomuceno 2013; Soheilian 2007), two pegaptanib (Macugen 2005; Macugen 2011), and three aflibercept (DA VINCI 2011; and two studies conducted in the USA and Europe using the same protocol, which we will refer to as a single study (Korobelnik 2014). DRCRnet 2015 and Wiley 2016 were the only studies comparing ranibizumab, bevacizumab or aflibercept directly. The drug dose was the same in most studies (0.5 mg ranibizumab, 1.25 mg bevacizumab, 0.3 mg pegaptanib, 2 mg aflibercept) except for RESOLVE 2010 where dose adjustment was allowed for ranibizumab, and also RISE‐RIDE, DRCRnet 2015 and Wiley 2016 where 0.3 mg ranibizumab was also delivered.

Anti‐VEGF treatment regimens were monthly in RISE‐RIDE, in one arm of Korobelnik 2014 and in Wiley 2016. Monthly, bimonthly and 'as needed' or pro re nata (PRN) regimens were adopted in four arms of DA VINCI 2011, and we selected PRN for efficacy data extraction because this is current practice with other anti‐VEGF drugs. Ahmadieh 2008 was a short‐term study which delivered only the first three injections. Most other studies adopted three initial injections followed by various maintenance regimens. Two studies on aflibercept, reported in Korobelnik 2014 (VISTA and VIVID), compared laser photocoagulation with both monthly injections (2q4) and a regimen of five initial monthly injections followed by bimonthly injections (2q8) followed by a 'treat‐and‐extend' regimen in year two.

PRN retreatment criteria were based on: visual acuity only in Nepomuceno 2013 and REVEAL 2015; OCT only in BOLT 2010, Macugen 2011 and READ2 2009; OCT and visual acuity in Azad 2012, DRCRnet 2010, DRCRnet 2015, Ekinci 2014, RESOLVE 2010 and in the PRN arm of DA VINCI 2011; inclusion of clinical examination or at the examiners' discretion in Macugen 2005, RESTORE 2011 and Soheilian 2007. They were unclear in Lopez‐Galvez 2014, RELATION 2012 and RESPOND 2013.

Types of outcomes

The data structure of our efficacy and safety outcomes can be seen in Table 1 where the sum of cases for each outcome is shown.

See: Summary of findings for the main comparison Antiangiogenic therapy versus control; Summary of findings 2 Ranibizumab versus aflibercept for diabetic macular oedema; Summary of findings 3 Ranibizumab versus bevacizumab for diabetic macular oedema

Table 1. Reporting of all outcomes across studies

Study	Gain 3+ VA lines		Mean VA change		Mean CMT change		QOL		SSAE		ATC		Death
Follow‐up year	1	2	1	2	1	2	1	2	1	2	1	2	1	2
Ahmadieh 2008			78		78
Azad 2012	40								40		40		40
BOLT 2010	80	65	80	65	80	65				80		80	80
DA VINCI 2011	89		89		87				89		89		89
DRCRnet 2010	668	486	668	486	617	483				240		505		505
DRCRnet 2015	620	577	620	577	620	577				660		660		660
Ekinci 2014			100		100				100		100		100
Ishibashi 2014	233		233				233		233		233		233
Korobelnik 2014	573		573		573					865		865		865
LUCIDATE 2014	33		33		33				33		33		33
Macugen 2005	86		86						86		86		86
Macugen 2011	260	207	260				236			286		286		286
Nepomuceno 2013	60		60		60				60
READ2 2009	115		115								117		117
RELATION 2012			128						128		128		128
RESOLVE 2010	151		151		151				151		151		151
RESPOND 2013	203		203		202				237				237
RESTORE 2011	343		343		343		299		345		345		345
REVEAL 2015	390		390		268
RISE‐RIDE		509						504		500				500
Soheilian 2007	87		85		85				96
Turkoglu 2015			70		70		70
Wiley 2016			124		124
Total studies	17	5	21	3	16	3	4	1	12	6	10	5	12	5
Total participants	4031	1844	4489	1128	3491	1125	838	504	1598	2631	1322	2396	1639	2816

Numbers in the table are the total number of eyes for each study,as available by follow‐up year (1 or 2) and outcome measure.

Studies awaiting assessment

Several trials were included as ongoing in the previous version of this review. We checked the completion status on the study trial register and tried to contact the authors, but were not able to obtain additional information (NCT00387582; NCT00997191 (IBeTA); NCT01445899 (MATISSE); NCT01565148 (IDEAL)).

Two Chinese trials (Chen 2016, 72 participants; Huang 2016, 78 participants) compared ranibizumab plus laser or, respectively, ranibizumab to grid laser. These trials provided baseline and final CRT data at six months as well as the proportion with visual improvement, but the improvement cut‐off was unclear, as was the measurement tool.

Jovanovic 2015 included 72 participants (120 eyes) randomised to either bevacizumab (one or more injections with or without macular laser photocoagulation depending on results after four to six weeks) or macular laser alone to treat DMO. However, results were not provided at desired fixed follow‐up times by each randomisation group.

Fouda 2017 included 42 participants (70 eyes) randomised to aflibercept or ranibizumab and treated with three initial injections and then PRN. The authors did not find any significant difference between the two drugs in terms of BCVA, but used decimal rather than logMAR visual acuity and we could not use these data in analyses (authors contacted). The authors reported no difference regarding CRT and a smaller but statistically significant number of injections with aflibercept versus ranibizumab.

Excluded studies

See 'Characteristics of excluded studies' table for the list of exclusions with reasons.

Risk of bias in included studies

See 'Risk of bias in included studies'; Figure 2.

Figure 2

Methodological quality summary: review authors' judgements about each methodological quality item for each included study.

Allocation

Sequence generation was judged at low risk of bias in 12 studies and was unclear in nine (Azad 2012; Ekinci 2014; Lopez‐Galvez 2014; Ishibashi 2014; Korobelnik 2014; READ2 2009; RELATION 2012; RESPOND 2013; Turkoglu 2015). Method for allocation concealment was also unclear in these studies, as they were in Nepomuceno 2013. Allocation concealment was judged at high risk of bias in Ishibashi 2014.

Blinding

Masking of participants and outcome assessors was obtained in 14 and 12 trials respectively, and was unclear in seven and nine trials respectively. LUCIDATE 2014, READ2 2009 and RESPOND 2013 were unmasked.

Incomplete outcome data

Eleven trials were judged at low risk of attrition bias (Azad 2012; BOLT 2010; DA VINCI 2011; DRCRnet 2010; DRCRnet 2015; Korobelnik 2014; LUCIDATE 2014; Macugen 2005; Nepomuceno 2013; RESOLVE 2010; RESTORE 2011); and eight trials were judged at unclear risk of bias in which some participants were missing but reasons for missingness were not fully reported (Ahmadieh 2008; Ishibashi 2014; Macugen 2011; READ2 2009; RISE‐RIDE; Soheilian 2007; Turkoglu 2015; Wiley 2016). Five trials were judged at high risk of attrition bias: Ekinci 2014 excluded 15 participants after randomisation due to ocular and systemic complications; Lopez‐Galvez 2014 lost about 20% of participants in each arm and did not report the reasons; RELATION 2012, RESPOND 2013 and REVEAL 2015 lost many more participants in the laser arm than in the ranibizumab arms.

Selective reporting

Table 1 shows the reporting of all outcomes across 24 trials. Reporting was almost complete for mean VA change at one year (21 studies, 4489 complete cases). Mean CRT change was available in 16 studies (3491 cases). Gain of 3 or more VA lines was reported at one year in 17 studies (4031 cases). SSAEs at one or two years were reported from 18 studies (4229 cases). ATC thromboembolic events were reported in 15 (3718 cases) and death in 17 (4455 cases).

Other potential sources of bias

The baseline visual acuity was not balanced in Soheilian 2007; the visual acuity was around 20/100 in the bevacizumab and bevacizumab‐triamcinolone arms and 20/70 in the laser arm, suggesting that milder CSMO was included in the laser arm. The trial investigators adjusted for baseline values in the analyses, which also took into account the within‐participant correlation (150 eyes of 129 participants, 16% of participants with both eyes in the analyses). However, we could not take within‐participant correlation into account when analysing dichotomous visual acuity.

Three studies included both eyes of some participants in analyses: Ahmadieh 2008 14 out of 101 participants; Nepomuceno 2013 15 out of 48 participants; Wiley 2016 6 out of 56 participants.

RELATION 2012 was terminated early when ranibizumab was approved for DMO in Germany. Early termination was unlikely to be associated with treatment effect.

Effects of interventions

Antiangiogenic drugs versus laser photocoagulation or control: efficacy and safety

summary of findings Table for the main comparison presents the evidence on the comparison of each drug with laser photocoagulation (efficacy at one year) or control (laser photocoagulation or sham at the longest available follow‐up of one or two years).

Efficacy at one year

As found in the previous version of this review based on direct meta‐analyses (Virgili 2014), there was high‐certainty of evidence of benefit for aflibercept, bevacizumab and ranibizumab compared to laser photocoagulation at one year. Specifically, aflibercept, bevacizumab and ranibizumab were all more effective than laser for improving vision by 3 or more lines after one year, since about one in 10 people improve vision with laser, and about three in 10 people improve with anti‐VEGF treatment: risk ratio (RR) versus laser was 3.66 (95% CI 2.79 to 4.79) for aflibercept; 2.47 (95% CI 1.81 to 3.37) for bevacizumab; and 2.76 (95% CI 2.12 to 3.59) for ranibizumab. Regarding change of mean BCVA, on average there was no change with laser after one year, compared with a gain of 1 or 2 lines with anti‐VEGF treatment: laser versus aflibercept mean difference (MD) −0.20 (95% CI −0.22 to −0.17) logMAR; versus bevacizumab −0.12 (95% CI −0.15 to −0.09) logMAR; versus ranibizumab −0.12 (95% CI −0.14 to −0.10) logMAR (negative logMAR in favour of anti‐VEGF group).

The certainty of evidence was moderate for bevacizumab versus laser regarding mean BCVA change due to inconsistency of direct and indirect evidence.

Safety at the longest available follow‐up

This network meta‐analysis confirms that aflibercept, bevacizumab and ranibizumab do not increase the risk of all SSAEs compared to laser photocoagulation or sham at one year. We considered this evidence of high‐certainty. (summary of findings Table for the main comparison), Of notice, SSAEs are a generic indicator of harm, mostly including hospitalisation or death for any cause and unrelated to antiangiogenic effect.

Regarding 'Antiplatelet Trialists Collaboration arterial thromboembolic events' and all‐cause death, no statistically significant difference was found between any anti‐VEGF drug and control, but the certainty of the evidence was generally low due to imprecision (large 95% CIs).

Quality of life

Only RESTORE 2011, RESPOND 2013 and Turkoglu 2015 presented quality of life data for ranibizumab versus laser photocoagulation at six to 12 months (3 studies, 412 participants). Ranibizumab improved NEI‐VFQ composite score by 5.14 units (95% CI 2.96 to 7.32) compared to laser (summary of findings Table for the main comparison). The certainty of the evidence was moderate due to risk of bias issues (RESPOND 2013 was unmasked and Turkoglu 2015 was unclear for most items).

Macugen 2011 obtained QOL data at two years and we did not include these data since pegaptanib was not of direct interest and sham, rather than laser, was the comparator. RISE‐RIDE obtained QOL data at two years and was not included since sham, rather than laser, was the control group.

Ranibizumab versus aflibercept and bevacizumab

Efficacy at one year

Table 2 presents the number of studies (participants/eyes) in all treatment arms of the network for the efficacy outcomes at one year. Figure 3 presents the corresponding networks' structure. As seen, more data was available for ranibizumab, alone or combined with laser, with respect to aflibercept and bevacizumab. Figure 4, Figure 5 and Figure 6 present forest plots with effects for each study, estimates from direct pairwise meta‐analysis and mixed estimate from the network meta‐analysis. summary of findings Table 2 and summary of findings Table 3 present comparisons of ranibizumab versus aflibercept and bevacizumab.

Figure 3

Network structure for efficacy outcomes at 1 year

Figure 4

All direct and mixed comparisons: gain of 3 or more lines of visual acuity at 1 year

Figure 5

All direct and mixed comparisons: mean change in visual acuity at 1 year

Figure 6

All direct and mixed comparisons: mean change in central retinal thickness at 1 year (micron)

Table 2. Network structure: efficacy at 12 months

	Laser	Aflibercept	Bevacizumab	Pegaptanib	Ranibizumab	Ranibizumab deferredlaser	Ranibizumab prompt laser	Sham	Overall
Gain 3+ lines	12	4	5	3	8	1	5	3	17
	1074	539	344	410	713	188	545	218	4031
Mean VA change	13	4	7	3	11	1	6	4	21
	1131	539	476	410	861	188	629	255	4489
Mean CRT change	11	4	7		10	1	4	2	16
	986	538	476		779	175	451	86	3491
QOL				4					4
				838					838

For each efficacy outcome, numbers in the table are the total number of studies (upper line for each outcome) and the total number of eyes (lower line for each outcome), as available by treatment and measured at one year.

Comparing the available drugs as monotherapy, all efficacy outcomes significantly favoured aflibercept over ranibizumab and bevacizumab (Table 3; Table 4; Table 5). Compared with ranibizumab and bevacizumab, aflibercept increased the chances of gaining 3 or more lines (17 studies, 4031 eyes) by about 30%, since the RR for gain was 0.75 (95% CI 0.60 to 0.94) and 0.68 (95% CI 0.53 to 0.86) versus ranibizumab and bevacizumab, respectively. The corresponding figures for mean BCVA change (21 studies, 2689 eyes) were a difference of 0.08 (95% CI 0.05 to 0.11) logMAR and 0.08 (95% CI 0.05 to 0.11) logMAR and were 38.90 (95% CI 2.27 to 75.52) micron and 68.32 (95% CI 28.69 to 107.96) micron for CRT change (16 studies, 3491 eyes), all favouring aflibercept.

Table 3. Gain of 3 or more lines of visual acuity at 12 months: direct (upper‐right triangle) and mixed (lower‐left triangle) estimates

LASER	3.82 (2.61 to 5.58)	2.74 (1.34 to 5.61)		2.82 (1.82 to 4.38)	1.88 (1.31 to 2.70)	2.30 (1.74 to 3.03)
3.66 (2.79 to 4.79)	AFLI	0.68 (0.52 to 0.90)		0.77 (0.59 to 0.99)
2.47 (1.81 to 3.37)	0.68 (0.53 to 0.86)	BEVA		1.14 (0.88 to 1.48)
1.70 (0.58 to 4.94)	0.46 (0.16 to 1.34)	0.69 (0.24 to 1.89)	PEGA				0.51 (0.30 to 0.89)
2.76 (2.12 to 3.59)	0.75 (0.60 to 0.94)	1.11 (0.87 to 1.43)	1.62 (0.58 to 4.57)	RANI		0.90 (0.67 to 1.21)	0.31 (0.13 to 0.76)
2.02 (1.46 to 2.81)	0.55 (0.37 to 0.82)	0.82 (0.54 to 1.24)	1.19 (0.40 to 3.58)	0.73 (0.51 to 1.06)	RANI‐DL	1.10 (0.80 to 1.51)
2.33 (1.81 to 3.00)	0.64 (0.47 to 0.86)	0.94 (0.68 to 1.31)	1.37 (0.47 to 3.99)	0.85 (0.65 to 1.09)	1.15 (0.85 to 1.56)	RANI‐PL
0.87 (0.35 to 2.17)	0.24 (0.10 to 0.59)	0.35 (0.14 to 0.87)	0.51 (0.30 to 0.89)	0.32 (0.13 to 0.76)	0.43 (0.17 to 1.11)	0.37 (0.15 to 0.93)	SHAM

P value for overall inconsistency = 0.883 in the network meta‐analysis.

Values in the table are risk ratios and 95% confidence intervals. Values in bold are ones where the 95% confidence intervals does not include 1 (null effect).

Table 4. Mean visual acuity change at 12 months: direct (upper‐right triangle) and mixed (lower‐left triangle) estimates

LASER	−0.20 (−0.24 to −0.17)	−0.20 (−0.28 to −0.12)^a		−0.11 (−0.13 to −0.08)	−0.12 (−0.16 to −0.075)	−0.10 (−0.13 to −0.08)
−0.20 (−0.22 to −0.17)	AFLI	0.07 (0.03 to 0.11)		0.04 (0.00 to 0.08)
−0.12 (−0.15 to −0.09)^a	0.08 (0.05 to 0.11)	BEVA		−0.02 (−0.05 to 0.01)
0.01 (−0.09 to 0.07)	0.19 (0.11 to 0.27)	0.11 (0.04 to 0.19)	PEGA				0.08 (0.03 to 0.13)
−0.12 (−0.14 to −0.10)	0.08 (0.05 to 0.11)	0.00 (−0.02 to 0.03)	−0.11 (−0.19 to −0.04)	RANI		0.01 (−0.02 to 0.03)	0.23 (0.15 to 0.32)
−0.11 (−0.13 to −0.09)	0.08 (0.04 to 0.13)	0.01 (−0.04 to 0.06)	−0.11 (−0.19 to −0.02)	0.00 (−0.04 to 0.05)	RANI‐DL	0.00 (−0.05 to 0.05)
−0.11 (−0.14 to −0.08)	0.09 (0.06 to 0.12)	0.01 (−0.02 to 0.05)	−0.10 (−0.18 to −0.02)	0.01 (−0.01 to 0.03)	0.01 (−0.04 to 0.05)	RANI‐PL
0.08 (0.01 to 0.15)	0.28 (0.21 to 0.35)	0.20 (0.13 to 0.27)	0.09 (0.06 to 0.12)	0.20 (0.13 to 0.27)	0.20 (0.11 to 0.28)	0.19 (0.12 to 0.26)	SHAM

^a P value for differences between direct and indirect estimates = 0.031 in the network meta‐analysis.

P value for overall inconsistency = 0.665.

Table 5. Mean central retinal thickness change at 12 months: direct (upper‐right triangle) and mixed (lower‐left triangle) estimates

LASER	−119 (−143 to −95)	−44 (−82 to −5)	−71 (−120 to −22)^{^}	−35 (−62 to −8)	−64 (−103 to −25)^b*
−114 (−147 to −81)	AFLI	68 (43 to 94)	22 (−4 to 48)
−46 (−78 to −14)	68 (29 to 108)	BEVA	−38 (−56 to −20)			132 (72 to 187)
−75 (−100 to −50)	39 (2 to 76)	−29 (−58 to −1)	RANI		−19 (−39 to 2)^a	1470 (95 to 196)
−57 (−111 to −2)	57 (−6 to 120)	−11 (−73 to 51)	18 (−40 to 76)	RANI‐DL	6 (−22 to 34)
−72.90 (−103 to −42)^b	41 (−2 to 84)	−27 (−68 to 13)	2 (−31 to 35)^a	−16 (−71 to 38)	RANI‐PL
77 (18 to 137)	191 (127 to 256)	123 (67 to 179)	153 (97 to 208)	134 (55 to 213)	150 (87 to 214)	SHAM

^a P value for differences between direct and indirect estimates = 0.003.

^b P value for differences between direct and indirect estimates = 0.044.

^* P value for heterogeneity = 0.002; I² = 80% in the direct meta‐analysis.

^{^} P value for heterogeneity = 0.000; I² = 91% in the direct meta‐analysis.

P value for overall inconsistency = 0.209 in the network meta‐analysis.

Ranibizumab and bevacizumab did not differ in term of functional outcomes: RR of gain 1.11 (95% CI 0.87 to 1.43) and difference in mean VA change 0.00 (95% CI −0.02 to 0.03) logMAR. However, CRT reduction favoured ranibizumab by −29.4 (95% CI −58.2 to −0.70) micron.

There was no evidence of overall statistical inconsistency in our efficacy analyses (Table 3; Table 4; Table 5). We found evidence of statistical inconsistency in one comparison (bevacizumab versus laser) for mean BCVA change and in the loop connecting ranibizumab, ranibizumab plus prompt laser and laser for mean CRT change, where two direct meta‐analyses also showed high heterogeneity in the same loop.

Mean risk of bias was low for mixed and direct comparisons among aflibercept, bevacizumab and ranibizumab for all efficacy outcomes, except for the comparison between bevacizumab and ranibizumab regarding mean BCVA change and mean CRT change, which were judged at unclear risk of bias (Figure 7).

Figure 7

Contribution plot of mean overall study risk of bias to pairwise network estimates. Legends show the risk of bias of each direct comparison.

We had not pre‐planned any subgroup analyses and were unable to obtain data to carry out post hoc subgroup analyses by baseline BCVA. DRCRnet 2015, the only large study comparing the three drugs, found that aflibercept was superior to bevacizumab and ranibizumab for participants with lower vision (69 ETDRS letter or less or approximately 20/50 or 0.4 logMAR or worse), whereas differences between the three drugs were unimportant for participants with better vision.

Efficacy at two years

Three publicly funded studies (BOLT 2010; DRCRnet 2010; DRCRnet 2015) and two industry‐sponsored studies (Macugen 2011; RISE‐RIDE) provided data at two years. There was only one study for each comparison, making data unsuitable for a network meta‐analysis.

Only DRCRnet 2015 (complete cases: aflibercept n = 201, bevacizumab n = 185, ranibizumab n = 191) compared different antiangiogenic drugs, and found no VA differences between ranibizumab 0.3 mg and aflibercept (gain 3+ VA lines, RR 0.94, 95% CI 0.73 to 1.22; difference in mean VA change 0.01, 95% CI −0.04 to 0.06). Ranibizumab and bevacizumab did not differ in terms of gain of 3 or more VA lines (RR 0.94, 95% CI 0.72 to 1.24) but the difference in mean VA change favoured ranibizumab (mean difference −0.05, 95% CI −0.09 to 0.00), although it was not precisely estimated. Although effects on CRT favoured aflibercept over ranibizumab and ranibizumab over bevacizumab, none was statistically significant: mean difference −22 micron (95% CI −50 to 6 micron) and −23 micron (95% CI −52 to 6 micron) respectively.

We were unable to obtain data allowing subgroup analyses by baseline BCVA. DRCRnet 2015 found that such subgroup differences were attenuated at two years.

Safety at the longest available follow‐up

Table 6 presents the number of studies (participants/eyes) in the network for safety outcomes at the longest available follow‐up, and Figure 8 presents the corresponding network structure. Figure 9, Figure 10 and Figure 11 present forest plots for each study as well as their direct meta‐analysis and mixed estimates from the network meta‐analysis.

Figure 8

Network structure for safety outcomes at 1 year

Figure 9

All direct and mixed comparisons: serious systemic adverse events at the longest available follow‐up (1 or 2 years)

Figure 10

All direct and mixed comparisons: arterial thromboembolic events at the longest available follow‐up (1 or 2 years)

Figure 11

All direct and mixed comparisons: all‐cause death at the longest available follow‐up (1 or 2 years)

Table 6. Network structure: safety at the longest available follow‐up

	Laser	Aflibercept	Bevacizumab	Pegaptanib	Ranibizumab	Sham	Overall
SSAE	9	3	6	2	10	5	18
	1013	556	410	186	1303	528	4229
ATC*	10	3	4	2	8	2	15
	824	846	330	188	1113	184	3718
Death	11	3	4	2	10	3	17
	903	846	333	188	1521	434	4455

For each safety outcome, numbers in the table are the total number of studies (upper line for each outcome) and the total number of eyes (lower line for each outcome), as available by treatment and measured at the longest available follow‐up.

(*) combined incidence of (1) cardiovascular, hemorrhagic, and unknown death; (2) nonfatal MI; and (3) nonfatal stroke.

Two‐year data were available and reported in only four RCTs in this review. Most industry‐sponsored studies were open‐label after one year. Differently from efficacy analyses at one year, our safety analyses included data from RISE‐RIDE on ranibizumab with monthly treatments up to two years, as well as data from the monthly treatment arm (2q4) of Korobelnik 2014, which became PRN in the second year.

Though no analysis suggested a difference among drugs for any safety outcome, only estimates for SSAEs (18 studies, 4229 eyes) reached sufficient precision to exclude very large differences among drugs. Overall, no difference was detected in mixed evidence estimates for any drug compared to laser or sham. Moreover, RR 95% CI width excluded differences of 20% to 30% or more between aflibercept, bevacizumab and ranibizumab, while estimates for pegaptanib were less precise (Table 7). No overall (P = 0.86) or loop‐specific inconsistency was detected.

Table 7. All serious systemic adverse events (longest available follow‐up)

CONTROL	0.95 (0.75 to 1.20)	1.29 (0.43 to 3.84)	1.02 (0.67 to 1.53)	0.98 (0.76 to 1.25)
0.98 (0.83 to 1.16)	AFLI	0.95 (0.75 to 1.20)		1.04 (0.83 to 1.32)
0.93 (0.73 to 1.19)	0.95 (0.76 to 1.18)	BEVA		0.96 (0.77 to 1.20)
1.02 (0.64 to 1.64)	1.04 (0.63 to 1.72)	1.09 (0.64 to 1.86)	PEGA
0.97 (0.80 to 1.17)	0.98 (0.82 to 1.19)	1.04 (0.84 to 1.28)	0.95 (0.57 to 1.58)	RANI

P value for overall inconsistency = 0.859.

Fifteen studies (3718 eyes) contributed to this analysis on 'Antiplatelet Trialists Collaboration arterial thromboembolic events' (Table 8). No difference was detected in mixed evidence estimates for any drug compared to laser or sham or between drugs, but estimates were very imprecise. No overall inconsistency was detected (P = 0.19), but direct evidence from DRCRnet 2015 showed increased risk for ranibizumab compared to aflibercept (RR 2.26, 95% CI 1.15 to 4.23) which was larger and inconsistent with indirect evidence (P = 0.002), resulting in mixed evidence showing no difference (RR 1.25, 95% CI 0.50 to 3.05).

Table 8. Antiplatelet Trialists Collaboration arterial thromboembolic events at the longest available follow‐up

CONTROL	1.50 (0.81 to 2.79)	0.92 (0.17 to 5.12)	0.78 (0.31 to 1.97)	0.64 (0.38 to 1.07)
0.88 (0.37 to 2.13)	AFLI	1.46 (0.71 to 2.98)		2.26 (1.15 to 4.23)^a
0.94 (0.33 to 2.66)	1.06 (0.36 to 3.11)	BEVA		1.51 (0.85 to 2.69)
0.79 (0.20 to 3.02)	0.89 (0.18 to 4.43)	0.83 (0.15 to 4.61)	PEGA
1.09 (0.52 to 2.29)	1.24 (0.48 to 3.19)^a	1.17 (0.43 to 3.13)	1.17 (0.43 to 3.16)	RANI

^a P value for differences between direct and indirect estimates = 0.002.

P value for overall inconsistency = 0.274 in the network meta‐analysis.

Seventeen studies (4455 eyes) contributed to the analysis of 'all‐cause mortality' (Table 9). No difference was detected for direct, indirect and mixed evidence estimates for any drug compared to laser or sham or between drugs, but estimates were imprecise.

Table 9. All‐cause mortality at the longest available follow‐up

CONTROL	1.69 (0.30 to 9.42)^a	0.95 (0.06 to 14.85)	0.82 (0.25 to 2.65)	0.64 (0.32 to 1.25)^d
1.01 (0.34 to 3.03)^a	AFLI	2.67 (0.97 to 7.37)^b		2.26 (0.80 to 6.40)^c
1.61 (0.45 to 5.69)	1.59 (0.43 to 5.94)^b	BEVA		0.85 (0.40 to 1.83)
0.81 (0.16 to 4.03)	0.81 (0.12 to 5.62)	0.51(0.07 to 3.90)	PEGA
0.90 (0.40 to 2.01)	1.16 (0.38 to 3.58)^c	0.73 (0.22 to 2.37)	1.44 (0.24 to 8.48)	RANI

^a P value for differences between direct and indirect estimates = 0.011.

^b P value for differences between direct and indirect estimates = 0.030.

^c P value for differences between direct and indirect estimates = 0.015.

^d P value for differences between direct and indirect estimates = 0.022.

P value for overall inconsistency = 0.087 in the network meta‐analysis.

Mean risk of bias was low for mixed and direct comparisons between aflibercept and ranibizumab and unclear for bevacizumab versus ranibizumab for SSAEs. Regarding ATC arterial thromboembolic events and all‐cause death, risk of bias was low for aflibercept versus ranibizumab but it was unclear or high for bevacizumab versus ranibizumab (Figure 7).

Quality of the evidence

See above for the discussion of risk of bias of mixed evidence in pairwise comparisons of interest.

Statistical heterogeneity between studies

When a direct meta‐analysis was possible, no heterogeneity of effects was found for the following outcomes: gain of 3 or more BCVA lines, mean BCVA change, SSAEs, ATC arterial thrombembolic events, death. As reported above, there was high heterogeneity in the meta‐analysis of change in CRT for the comparisons of ranibizumab with laser (I² = 91%) and ranibizumab plus deferred laser versus laser (I² = 80%), but not in other two meta‐analyses in this network.

Estimates of between‐study standard deviation τ in the network meta‐analyses suggested little heterogeneity for dichotomous outcomes, except for ATC arterial thrombembolic events when it was moderate (τ = 0.51) . Values for BCVA change and CRT change were 8^‐10 logMAR and 27 micron, respectively. These values mean that heterogeneity was negligible for VA change, but was compatible with a predictive intervals width increased by at least 100 micron for the CRT change.

Similarity between studies

Table 10 shows baseline characteristics (BCVA, CRT) and the number of injections across study and treatment arms. Overall, most studies included participants with mean BCVA about 20/60 and CRT between 400 and 500 micron, which we believe sufficiently homogeneous. The number of injections was high compared with current practice (seven to 10 in year one), except for few small studies delivering a low number of injections. No heterogeneity was suspected between studies using 0.3 versus 0.5 mg ranibizumab. In the safety analyses, we included two studies with monthly injections — one arm of Korobelnik 2014 for aflibercept and RISE‐RIDE for ranibizumab — and, again, no heterogeneity seemed to arise from lower intensity regimens in other studies. Regarding sponsorship, there were fewer industry‐sponsored studies on bevacizumab, but these studies were also smaller than other studies, and the impact of such differences cannot be assessed. Finally, we did not consider the OCT model used in each study as a source of heterogeneity in CRT change since this was balanced between the arms of each study.

Table 10. Similarity among studies. baseline values and number of injections

Study	Participants	Interventions	Mean n. injections	Visual acuity (logMAR)	Retinal thickness (µm)	Study sponsor
Ahmadieh 2008	78	Bevacizumab Sham				None reported
BOLT 2010	80	Bevacizumab Laser	9* 3*	0.59 0.61	507 482	Public
DA VINCI 2011	89	Aflibercept Laser	3.6 to 5.5 1.7	0.55	441	Industry
DRCRnet 2010	668	Laser Ranibizumab‐DL Ranibizumab‐PL	3* 9* 8*	0.38 0.39 0.38		Public
DRCRnet 2015	620	Aflibercept Bevacizumab Ranibizumab	9.2 9.4 9.7	0.40	412 414 407	Public
Ekinci 2014	100	Bevacizumab Ranibizumab	5.1 6.5	0.22 0.24	484 490	Public
Ishibashi 2014	233	Pegaptanib Sham	4	0.56		Industry
Korobelnik 2014	268	Aflibercept Laser	8.5 2.4	0.50		Industry
Lopez‐Galvez 2014	83	Ranibizumab Laser	5.3 2.1			Industry
LUCIDATE 2014	33	Laser Ranibizumab	2.6 9	0.42 0.30	489 455	No details
Macugen 2005	86	Pegaptanib Sham	5 4.5	0.56 0.58	476 423	Industry
Macugen 2011	260	Pegaptanib Sham	8.3 8.4	0.56 0.58	442 465	Industry
RELATION 2012	128	Laser Ranibizumab‐PL				Industry
Nepomuceno 2013	60	Bevacizumab Ranibizumab	9.8 7.7	0.60 0.63	451 421	Public
READ2 2009	115	Laser Ranibizumab Ranibizumab‐PL	4.4 5.3 2.9	0.60 0.54 0.60	228 190 263	Industry
RESOLVE 2010	151	Ranibizumab Sham	10.2 8.9	0.50 0.48	455 449	Industry
RESPOND 2013	203	Laser Ranibizumab Ranibizumab‐PL	9.2 8.8	0.46 0.44 0.40	458 448 422	Industry
RESTORE 2011	343	Laser Ranibizumab Ranibizumab‐PL	7.3 7 6.8	0.46 0.40 0.42	412 427 416	Industry
REVEAL 2015	390	Laser Ranibizumab Ranibizumab‐PL	1.9 7.8 7	0.54 0.52 0.52	395 419 430	Industry
Soheilian 2007	85	Bevacizumab Laser		0.71 0.55	352 319	Public
Turkoglu 2015	70	Laser Ranibizumab		0.84 0.80	460 488	None reported
Wiley 2016	124	Bevacizumab Ranibizumab	3 3	0.42	477	Public

DL: plus deferred laser
PL: plus prompt laser

(*): median, not mean, available and reported

Selective reporting

Comparison adjusted funnel plots showed some asymmetry for the outcome 'gain of 3 or more VA lines', but no specific direct comparison seemed to be affected (Figure 12). Instead, asymmetric observations to the left of the non‐significance area were seen for the outcome 'mean CRT change at one year' regarding the comparison 'ranibizumab versus laser', which also suffered high heterogeneity.

Figure 12

Comparison‐adjusted funnel plot for all outcome measures

Overall quality of evidence for the main comparisons between drugs

summary of findings Table 2 and summary of findings Table 3 show summary data as well as the overall quality of evidence for the comparisons of interest in this review, based on the data reported above.

Sensitivity analyses on studies at low risk of bias

We conducted analyses of efficacy outcomes after excluding 10 studies at unclear or high risk of bias. These analyses confirmed the findings with all studies.

Gain of 3 or more BCVA lines: ranibizumab versus aflibercept RR 0.74 (95% CI 0.59 to 0.93); bevacizumab versus ranibizumab RR 0.89 (95% CI 0.69 to 1.15); no overall or comparison‐specific inconsistency.
Mean change in BCVA: ranibizumab versus aflibercept 0.08 (95% CI 0.05 to 0.12) logMAR; bevacizumab versus ranibizumab 0.0 (95% CI 0.2 to −0.03); no overall inconsistency (P = 0.130), some inconsistency for the comparisons of bevacizumab versus laser (P = 0.046) and ranibizumab versus laser (P = 0.043), same direction of effects.
Mean change in CRT: ranibizumab versus aflibercept 84 (95% CI 43 to 125) micron; bevacizumab versus ranibizumab 16 (95% CI −35 to 67) micron; overall inconsistency was detected (P = 0.012), which was due to the loop ranibizumab plus prompt laser versus ranibizumab plus deferred laser versus laser (P < 0.001 for all comparisons).

Therefore, sensitivity analyses confirmed the results of main analyses.

Discussion

Summary of main results

This review and network meta‐analysis confirms the findings of the previous version which found high‐certainty evidence that antiangiogenic therapy provides benefit over laser treatment in people with DMO at one year and concluded that further studies should compare different drugs. This update found four studies with direct comparisons between drugs and augmented this evidence with indirect comparisons in a network meta‐analysis based on 24 included studies.

At one year, all efficacy outcomes significantly favoured aflibercept over ranibizumab and bevacizumab. Aflibercept increased the chances of gaining 3 or more BCVA lines by about 30%, and conferred an advantage of between half and 1 BCVA line over the other drugs (moderate‐certainty of evidence). Ranibizumab and bevacizumab did not differ in terms of functional outcomes (moderate‐certainty of evidence), but ranibizumab was more effective in terms of CRT reduction (low‐certainty of evidence). There was no evidence of statistical inconsistency in our analyses, except for the comparison between ranibizumab and bevacizumab for mean BCVA change, but the differences between direct and indirect evidence were clinically irrelevant.

The BCVA difference between aflibercept versus ranibizumab or bevacizumab was largely below the threshold of 1 ETDRS line (five letters or 0.1 logMAR) that was used for non‐inferiority in trials on DMO (OZDRY 2015, PLACID 2013) and AMD (CATT 2011), suggesting this difference was not clinically relevant.

The previous version of this review found moderate‐certainty evidence of good safety of antiangiogenic drugs, including aflibercept, bevacizumab, ranibizumab and pegaptanib, versus control. This update found high‐certainty evidence that aflibercept, ranibizumab and bevacizumab do not differ regarding SAEs, excluding RR differences between drugs by more than 25%. However, estimates were imprecise on well‐defined hard events such as death or arterial thromboembolic events (very low‐certainty evidence).

Two‐year data on direct comparisons were available only for one large multicentre publicly funded study showing smaller differences between aflibercept, bevacizumab and ranibizumab as compared to one year (DRCRnet 2015).

Overall completeness and applicability of evidence

This review confirms and enhances the findings of DRCRnet 2015: that aflibercept confers some advantage over bevacizumab and ranibizumab at one year. Two‐year data were available and reported in only four RCTs in this review. Most industry sponsored studies were open‐label after one year. Thus, long‐term outcomes have to be inferred from observational trials.

We could not investigate subgroup effects in this review due to lack of subgroup data. DRCRnet 2015 found the relative benefit with aflibercept versus ranibizumab and bevacizumab is larger when visual acuity is lower than about 20/50, and modest above this level. Moreover, we could not investigate the potential effect of differences in dose and regimen. Nonetheless, our review includes studies with a broad range of characteristics, but without major differences in populations. Particularly, 0.3 mg ranibizumab was used in the direct comparison with aflibercept and bevacizumab in DRCRnet 2015; yet indirect evidence, mostly based on 0.5 mg ranibizumab, was consistent. Regarding treatment frequency: the only study on monthly 0.3 mg ranibizumab was RISE‐RIDE for indirect evidence, but this could not be included since one‐year data were not available. We excluded the monthly aflibercept treatment arms of Korobelnik 2014 for efficacy outcomes at one year since this is not the registered label.

We would like to remark that this evidence is obtained in clinical trials with high treatment and monitoring standards. A pragmatic RCT would be needed to assess the real‐world effectiveness of anti‐VEGF treatment for DMO, since it could be dependent on the adequacy of monitoring treatment response, which is also sensitive to resource constraints, as found for AMD (Pagliarini 2014). Moreover, evidence on safety from non‐randomised, real‐world data was not included in our review. As found for AMD, real‐world studies suggest that people with DMO may differ from those in RCTs (Ziemssen 2017). However there is more compelling real‐world evidence that patients are under‐treated and have less favourable outcomes than in RCTs for AMD (Chong 2016) compared to DMO (Jiang 2015; Patrao 2016).

Quality of the evidence

The quality of evidence has been presented above with reasons for downgrading shown in summary of findings Table 2 and summary of findings Table 3. Overall, inconsistency was not an issue in our network meta‐analyses. We also think that transitivity and generalisablilty, or indirectness according to GRADE (Schünemann 2011), were not a problem since studies included a broad range of people with DMO that resembles those in clinical practice. Minor funnel plot asymmetry was detected only for CRT change and did not involve the treatments of direct interest in this review.

The tight monitoring of participants in RCTs differs from clinical practice, where a lower number of intravitreal injections and under‐treatment are common. As for age‐related macular degeneration, this may overestimate benefit with anti‐VEGF treatment. However, effect differences among drugs may be less biased if similar regimens are compared in RCTs, although this remains presumptive.

Potential biases in the review process

Bevacizumab is an off‐label drug for treating DMO in most countries. Because small RCTs using bevacizumab may have been conducted but not published because no difference was found, we could have missed small unpublished studies.

Agreements and disagreements with other studies or reviews

Although we did not systematically search for other reviews on anti‐VEGF treatments for DMO, the previous version of this review, which focused on anti‐VEGF drugs effects compared to control, reported on other network meta‐analyses which were inconclusive (Ford 2012; MEDCAC 2012; Regnier 2014); and included a much smaller set of studies, as did Korobelnik 2015 more recently. Zhang 2016 conducted a network meta‐analysis of 21 studies of anti‐VEGF drugs versus any control, including studies on intravitreal steroids, such as dexamethasone and triamcinolone, and one retrospective comparative, non‐randomised study (Arevalo 2013). The authors concluded that aflibercept was superior to other drugs at 12 months: they found a difference of about 0.04 logMAR (2 ETDRS letters) between aflibercept and ranibizumab as well as between ranibizumab and bevacizumab, but these did not reach statistical significance, as did differences in retinal thickness. Differently from Zhang 2016, we included a larger number of studies on anti‐VEGF drugs, but not intravitreal steroids since their benefit profile, as well as their local and systemic harm profile, only partly overlap with that of anti‐VEGF drugs and similarity of studies regarding study design and target population would be less likely achieved. Currently, some investigators think intravitreal steroids are preferred in people with anti‐VEGF resistant and chronic DMO (Hussain 2015), as an alternative to switching between anti‐VEGF drugs. We suggest that a network meta‐analysis including both anti‐VEGF drugs and steroids is of interest, but a different approach should be used, specifically regarding heterogeneity of effects by time horizon and participants' subgroups.

The European Society of Retina Specialists have recently published guidelines on the management of DMO, which cover a broad spectrum of clinical questions ranging from imaging interpretation to diabetes management (EURETINA 2017). These guidelines rely on individual study results, particularly those of DRCRnet 2015 which compared aflibercept, bevacizumab and ranibizumab directly regarding anti‐VEGF drug choice. They concluded that "aflibercept is the drug of choice in DME eyes with baseline BCVA below 69 letters, as it shows superiority to bevacizumab over 2 years and over ranibizumab in the first year of treatment" and that "all three medications are equivalent in improving vision in eyes with a baseline BCVA letter score of 69 or more". Our review does not provide additional evidence regarding the effect of baseline vision on visual outcome, since few studies maintained randomisation beyond one year and subgroup data were not available to conduct a network meta‐analysis. Regarding the difference between aflibercept and ranibizumab at one year, EURETINA 2017 stated that "it remains unclear to which extent the slower effect of ranibizumab seen in Protocol T [DRCRnet 2015] compared to aflibercept can be accounted to the lower dose (0.3 mg) of ranibizumab used in this study". Our review found that the difference between aflibercept and ranibizumab was consistent with indirect evidence based on studies that mostly used ranibizumab 0.5 mg, suggesting no dose effect as previously suggested (Heier 2016).

Figure 1

Study flow diagram.

Figure 2

Methodological quality summary: review authors' judgements about each methodological quality item for each included study.

Figure 3

Network structure for efficacy outcomes at 1 year

Figure 4

All direct and mixed comparisons: gain of 3 or more lines of visual acuity at 1 year

Figure 5

All direct and mixed comparisons: mean change in visual acuity at 1 year

Figure 6

All direct and mixed comparisons: mean change in central retinal thickness at 1 year (micron)

Figure 7

Contribution plot of mean overall study risk of bias to pairwise network estimates. Legends show the risk of bias of each direct comparison.

Figure 8

Network structure for safety outcomes at 1 year

Figure 9

All direct and mixed comparisons: serious systemic adverse events at the longest available follow‐up (1 or 2 years)

Figure 10

All direct and mixed comparisons: arterial thromboembolic events at the longest available follow‐up (1 or 2 years)

Figure 11

All direct and mixed comparisons: all‐cause death at the longest available follow‐up (1 or 2 years)

Figure 12

Comparison‐adjusted funnel plot for all outcome measures

Analysis 1.1

Comparison 1 Ranibizumab versus laser photocoagulation at 6 to 12 months, Outcome 1 Quality of life: NEI‐VFQ composite score at 6 to 12 months.

Summary of findings for the main comparison. Antiangiogenic therapy versus control

Antiangiogenic therapy versus control
Patient or population: people with diabetic macular oedema Settings: ophthalmology clinics Interventions: laser photocoagulation, aflibercept, bevacizumab, ranibizumab
Outcomes	Assumed risk*	Corresponding risk and relative risk (95% CI) , mixed evidence**			Certainty of evidence and reason for downgrading
Outcomes	Laser photocoagulation	Aflibercept	Bevacizumab	Ranibizumab	Certainty of evidence and reason for downgrading
Gain 3+ lines of visual acuity at 1 year	100 per 1000	366 per 1000 (279 to 479) RR: 3.66 (2.79 to 4.79)	247 per 1000 (181 to 337) RR: 2.47 (1.81 to 3.37)	276 per 1000 (212 to 359) RR: 2.76 (2.12 to 3.59)	⊕⊕⊕⊕ high
Visual acuity change at 1 year Measured on the logMAR scale, range −0.3 to 1.3. Higher values represent worse visual acuity.	On average visual acuity improved by −0.01 logMAR units in the laser group between the start of treatment and 1 year (effectively no change)	Average change in visual acuity was −0.20 (−0.22 to −0.17) logMAR units better with aflibercept compared with laser photocoagulation	Average change in visual acuity was −0.12 (−0.15 to −0.09) logMAR units better with bevacizumab compared with laser photocoagulation	Average change in visual acuity was −0.12 (−0.14 to −0.10) logMAR units better with ranibizumab compared with laser photocoagulation	⊕⊕⊕⊕ high for aflibercept and ranibizumab ⊕⊕⊕ moderate for bevacizumab (−1 for inconsistency of indirect versus direct evidence)
Central retinal thickness μm (CRT) change at 1 year The aim of treatment is to reduce central retinal thickness so thinner is better.	On average CRT changed by −64 μm in the laser group between the start of treatment and 1 year (became thinner)	Average change in CRT was −114 (−147 to −81) μm more (thinner) with aflibercept compared with laser photocoagulation	Average change in CRT was −46 (−78 to −14) μm more (thinner)with bevacizumab compared with laser photocoagulation	Average change in CRT was −75 (−100 to −50) μm more (thinner) with ranibizumab compared with laser photocoagulation	⊕⊕⊕⊕ high
Quality of life: NEI‐VFQ composite score at 6 to 12 months An improvement by 5 units is clinically significant.	On average the composite score improved by +2 units in the laser group between the start of treatment and 6 to 12 months			Average change in composite score was 5.14 (2.96 to 7.32) with ranibizumab compared with laser photocoagulation	⊕⊕⊕ moderate (−1 for risk of bias)
All serious systemic adverse events at 1 to 2 years	200 per 1000	196 per 1000 (166 to 232) RR: 0.98 (0.83 to 1.16)	186 per 1000 (146 to 238) RR: 0.93 (0.73 to 1.19)	194 per 1000 (160 to 234) RR: 0.97 (0.80 to 1.17)	⊕⊕⊕⊕ high
Arterial thromboembolic events at 1 to 2 years	45 per 1000	38 per 1000 (16 to 94) RR: 0.88 (0.37 to 2.13)	41 per 1000 (15 to 117) RR: 0.94 (0.33 to 2.66)	48 per 1000 (23 to 101) RR: 1.09 (0.52 to 2.29)	⊕⊕ low (−2 for imprecise estimates)
Death at 1 to 2 years	20 per 1000	20 per 1000 (7 to 61) RR: 1.01 (0.34 to 3.03) a	32 per 1000 (9 to 114) RR: 1.61 (0.45 to 5.69)	18 per 1000 (8 to 40) RR: 0.90 (0.40 to 2.01)	⊕⊕ low for bevacizumab and aflibercept (−2 for imprecise estimates) ⊕ very low for aflibercept (additional −1 direct evidence inconsistent, higher risk)
The assumed risk in the laser group was estimated as the row sum of the events divided by the row sum of the participants (eyes) for dichotomous variables, and as the (unweighted) median change of visual acuity or central retina thickness. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). The risk ratio was estimated from mixed (direct and indirect) comparisons. CI: Confidence interval; RR:** Risk ratio.
GRADE Working Group grades of evidence High‐certainty: further research is very unlikely to change our confidence in the estimate of effect. Moderate‐certainty: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low‐certainty: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low‐certainty: we are very uncertain about the estimate.

Summary of findings for the main comparison. Antiangiogenic therapy versus control

Summary of findings 2. Ranibizumab versus aflibercept for diabetic macular oedema

Ranibizumab versus aflibercept for diabetic macular oedema
Patient or population: people with diabetic macular oedema Settings: ophthalmology clinics Interventions: aflibercept, ranibizumab
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI), mixed evidence**	Certainty of the evidence (GRADE)	Reason for downgrading certainty of evidence
	Assumed risk	Corresponding risk
	Aflibercept	Ranibizumab
Gain 3+ lines of visual acuity at 1 year	370 per 1000	278 per 1000 (222 to 348)	RR: 0.75 (0.60 to 0.94)	⊕⊕⊕ moderate	−1 for imprecision as confidence intervals include both clinically important and clinically unimportant effects
Visual acuity change at 1 year Measured on the logMAR scale, range −1.3 to 1.3. Higher values represent worse visual acuity.	On average visual acuity improved by−0.23 logMAR units in the aflibercept group between the start of treatment and 1 year	Average change in visual acuity was 0.08 (0.05 to 0.11) logMAR units worse with ranibizumab compared with aflibercept		⊕⊕⊕ moderate	−1 for imprecision as confidence intervals include both clinically important and clinically unimportant effects
Central retinal thickness μm (CRT) change at 1 year The aim of treatment is to reduce central macular thickness so thinner is better.	On average CRT changed by −181 μm in the aflibercept group between the start of treatment and 1 year (became thinner)	Average change in CRT was 39 (2 to 76) μm more (thicker) with ranibizumab compared with aflibercept		⊕⊕ low	−1 for high heterogeneity in two direct comparisons and large predictive intervals −1 for imprecision
Quality of life at 1 year	No data available.
All serious systemic adverse events at 1 to 2 years	345 per 1000	338 per 1000 (283 to 411)	RR 0.98 (0.82 to 1.19)	⊕⊕⊕⊕ high
Arterial thromboembolic events at 1 to 2 years	60 per 1000	74 per 1000 (29 to 191)	RR 1.24 (0.48 to 3.19)	⊕ very low	Inconsistency between direct and indirect evidence (−1), and imprecise estimates (−2)
Death at 1 to 2 years	30 per 1000	35 per 1000 (11 to 108)	RR 1.16 (0.38 to 3.58)	⊕ very low	Inconsistency between direct and indirect evidence (−1), and imprecise estimates (−2)
The assumed risk in the aflibercept group was estimated as the row sum of the events divided by the row sum of the participants (eyes) for dichotomous variables, and as the (unweighted) median change of visual acuity or central retina thickness. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). The risk ratio was estimated from mixed (direct and indirect) comparisons. CI: Confidence interval; RR:** Risk ratio.
GRADE Working Group grades of evidence High‐certainty: further research is very unlikely to change our confidence in the estimate of effect. Moderate‐certainty: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low‐certainty: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low‐certainty: we are very uncertain about the estimate.

Summary of findings 2. Ranibizumab versus aflibercept for diabetic macular oedema

Summary of findings 3. Ranibizumab versus bevacizumab for diabetic macular oedema

Ranibizumab versus bevacizumab for diabetic macular oedema
Patient or population: people with diabetic macular oedema Settings: ophthalmology clinics Interventions: bevacizumab, ranibizumab
Outcomes	*Illustrative comparative risks (95% CI)**		Relative effect (95% CI), mixed evidence**	Certainty of the evidence (GRADE)	Reason for downgrading certainty of evidence
	Assumed risk	Corresponding risk
	Bevacizumab	Ranibizumab
Gain 3+ lines of visual acuity at 1 year	300 per 1000	333 per 1000 (261 to 429)	RR 1.11 (0.87 to 1.43)	⊕⊕⊕ moderate	Imprecise estimate (−1)
Visual acuity change at 1 year Measured on the logMAR scale, range −1.3 to 1.3. Higher values represent worse visual acuity.	On average visual acuity improved by −0.19 logMAR units in the bevacizumab group between the start of treatment and 1 year	Average change in visual acuity was0.00 (−0.02 to 0.03) logMAR units (same) with ranibizumab compared with bevacizumab		⊕⊕⊕ moderate	Unclear risk of bias (−1)
Central retinal thickness (CRT) change at 1 year The aim of treatment is to reduce central macular thickness so thinner is better.	On average CRT changed by −98 μm in the bevacizumab group between the start of treatment and 1 year (became thinner)	Average change in CRT was −29 (−58 to −1) μm more (thinner) with ranibizumab compared with bevacizumab		⊕⊕ low	Unclear risk of bias (−1) Imprecise estimate (−1)
Quality of life at 1 year	No data available
All serious systemic adverse events at 1 to 2 years	240 per 1000	250 per 1000 (202 to 307)	RR 1.04 (0.84 to 1.28)	⊕⊕⊕ moderate	Unclear risk of bias (−1)
Arterial thromboembolic events at 1 to 2 years	60 per 1000	70 per 1000 (26 to 189)	RR 1.17 (0.43 to 3.13)	⊕ very low	Unclear risk of bias (−1) Imprecise estimate (−2)
Death at 1 to 2 years	40 per 1000	29 per 1000 (9 to 95)	RR 0.73 (0.22 to 2.37)	⊕ very low	High risk of bias (−2) Imprecise estimate (−2)
The assumed risk in the bevacizumab group was estimated as the row sum of the events divided by the row sum of the participants (eyes) for dichotomous variables, and as the (unweighted) median change of visual acuity or central retina thickness. The corresponding risk (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). The risk ratio was estimated from mixed (direct and indirect) comparisons. CI: Confidence interval; RR:** Risk ratio.
GRADE Working Group grades of evidence High‐certainty: further research is very unlikely to change our confidence in the estimate of effect. Moderate‐certainty: further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Low‐certainty: further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate. Very low‐certainty: we are very uncertain about the estimate.

Summary of findings 3. Ranibizumab versus bevacizumab for diabetic macular oedema

Table 1. Reporting of all outcomes across studies

Study	Gain 3+ VA lines		Mean VA change		Mean CMT change		QOL		SSAE		ATC		Death
Follow‐up year	1	2	1	2	1	2	1	2	1	2	1	2	1	2
Ahmadieh 2008			78		78
Azad 2012	40								40		40		40
BOLT 2010	80	65	80	65	80	65				80		80	80
DA VINCI 2011	89		89		87				89		89		89
DRCRnet 2010	668	486	668	486	617	483				240		505		505
DRCRnet 2015	620	577	620	577	620	577				660		660		660
Ekinci 2014			100		100				100		100		100
Ishibashi 2014	233		233				233		233		233		233
Korobelnik 2014	573		573		573					865		865		865
LUCIDATE 2014	33		33		33				33		33		33
Macugen 2005	86		86						86		86		86
Macugen 2011	260	207	260				236			286		286		286
Nepomuceno 2013	60		60		60				60
READ2 2009	115		115								117		117
RELATION 2012			128						128		128		128
RESOLVE 2010	151		151		151				151		151		151
RESPOND 2013	203		203		202				237				237
RESTORE 2011	343		343		343		299		345		345		345
REVEAL 2015	390		390		268
RISE‐RIDE		509						504		500				500
Soheilian 2007	87		85		85				96
Turkoglu 2015			70		70		70
Wiley 2016			124		124
Total studies	17	5	21	3	16	3	4	1	12	6	10	5	12	5
Total participants	4031	1844	4489	1128	3491	1125	838	504	1598	2631	1322	2396	1639	2816
Numbers in the table are the total number of eyes for each study,as available by follow‐up year (1 or 2) and outcome measure.

Table 1. Reporting of all outcomes across studies

Table 2. Network structure: efficacy at 12 months

	Laser	Aflibercept	Bevacizumab	Pegaptanib	Ranibizumab	Ranibizumab deferredlaser	Ranibizumab prompt laser	Sham	Overall
Gain 3+ lines	12	4	5	3	8	1	5	3	17
	1074	539	344	410	713	188	545	218	4031
Mean VA change	13	4	7	3	11	1	6	4	21
	1131	539	476	410	861	188	629	255	4489
Mean CRT change	11	4	7		10	1	4	2	16
	986	538	476		779	175	451	86	3491
QOL				4					4
				838					838
For each efficacy outcome, numbers in the table are the total number of studies (upper line for each outcome) and the total number of eyes (lower line for each outcome), as available by treatment and measured at one year.

Table 2. Network structure: efficacy at 12 months

Table 3. Gain of 3 or more lines of visual acuity at 12 months: direct (upper‐right triangle) and mixed (lower‐left triangle) estimates

LASER	3.82 (2.61 to 5.58)	2.74 (1.34 to 5.61)		2.82 (1.82 to 4.38)	1.88 (1.31 to 2.70)	2.30 (1.74 to 3.03)
3.66 (2.79 to 4.79)	AFLI	0.68 (0.52 to 0.90)		0.77 (0.59 to 0.99)
2.47 (1.81 to 3.37)	0.68 (0.53 to 0.86)	BEVA		1.14 (0.88 to 1.48)
1.70 (0.58 to 4.94)	0.46 (0.16 to 1.34)	0.69 (0.24 to 1.89)	PEGA				0.51 (0.30 to 0.89)
2.76 (2.12 to 3.59)	0.75 (0.60 to 0.94)	1.11 (0.87 to 1.43)	1.62 (0.58 to 4.57)	RANI		0.90 (0.67 to 1.21)	0.31 (0.13 to 0.76)
2.02 (1.46 to 2.81)	0.55 (0.37 to 0.82)	0.82 (0.54 to 1.24)	1.19 (0.40 to 3.58)	0.73 (0.51 to 1.06)	RANI‐DL	1.10 (0.80 to 1.51)
2.33 (1.81 to 3.00)	0.64 (0.47 to 0.86)	0.94 (0.68 to 1.31)	1.37 (0.47 to 3.99)	0.85 (0.65 to 1.09)	1.15 (0.85 to 1.56)	RANI‐PL
0.87 (0.35 to 2.17)	0.24 (0.10 to 0.59)	0.35 (0.14 to 0.87)	0.51 (0.30 to 0.89)	0.32 (0.13 to 0.76)	0.43 (0.17 to 1.11)	0.37 (0.15 to 0.93)	SHAM
P value for overall inconsistency = 0.883 in the network meta‐analysis. Values in the table are risk ratios and 95% confidence intervals. Values in bold are ones where the 95% confidence intervals does not include 1 (null effect).

Table 3. Gain of 3 or more lines of visual acuity at 12 months: direct (upper‐right triangle) and mixed (lower‐left triangle) estimates

Table 4. Mean visual acuity change at 12 months: direct (upper‐right triangle) and mixed (lower‐left triangle) estimates

LASER	−0.20 (−0.24 to −0.17)	−0.20 (−0.28 to −0.12)^a		−0.11 (−0.13 to −0.08)	−0.12 (−0.16 to −0.075)	−0.10 (−0.13 to −0.08)
−0.20 (−0.22 to −0.17)	AFLI	0.07 (0.03 to 0.11)		0.04 (0.00 to 0.08)
−0.12 (−0.15 to −0.09)^a	0.08 (0.05 to 0.11)	BEVA		−0.02 (−0.05 to 0.01)
0.01 (−0.09 to 0.07)	0.19 (0.11 to 0.27)	0.11 (0.04 to 0.19)	PEGA				0.08 (0.03 to 0.13)
−0.12 (−0.14 to −0.10)	0.08 (0.05 to 0.11)	0.00 (−0.02 to 0.03)	−0.11 (−0.19 to −0.04)	RANI		0.01 (−0.02 to 0.03)	0.23 (0.15 to 0.32)
−0.11 (−0.13 to −0.09)	0.08 (0.04 to 0.13)	0.01 (−0.04 to 0.06)	−0.11 (−0.19 to −0.02)	0.00 (−0.04 to 0.05)	RANI‐DL	0.00 (−0.05 to 0.05)
−0.11 (−0.14 to −0.08)	0.09 (0.06 to 0.12)	0.01 (−0.02 to 0.05)	−0.10 (−0.18 to −0.02)	0.01 (−0.01 to 0.03)	0.01 (−0.04 to 0.05)	RANI‐PL
0.08 (0.01 to 0.15)	0.28 (0.21 to 0.35)	0.20 (0.13 to 0.27)	0.09 (0.06 to 0.12)	0.20 (0.13 to 0.27)	0.20 (0.11 to 0.28)	0.19 (0.12 to 0.26)	SHAM
^a P value for differences between direct and indirect estimates = 0.031 in the network meta‐analysis. P value for overall inconsistency = 0.665.

Table 4. Mean visual acuity change at 12 months: direct (upper‐right triangle) and mixed (lower‐left triangle) estimates

Table 5. Mean central retinal thickness change at 12 months: direct (upper‐right triangle) and mixed (lower‐left triangle) estimates

LASER	−119 (−143 to −95)	−44 (−82 to −5)	−71 (−120 to −22)^{^}	−35 (−62 to −8)	−64 (−103 to −25)^b*
−114 (−147 to −81)	AFLI	68 (43 to 94)	22 (−4 to 48)
−46 (−78 to −14)	68 (29 to 108)	BEVA	−38 (−56 to −20)			132 (72 to 187)
−75 (−100 to −50)	39 (2 to 76)	−29 (−58 to −1)	RANI		−19 (−39 to 2)^a	1470 (95 to 196)
−57 (−111 to −2)	57 (−6 to 120)	−11 (−73 to 51)	18 (−40 to 76)	RANI‐DL	6 (−22 to 34)
−72.90 (−103 to −42)^b	41 (−2 to 84)	−27 (−68 to 13)	2 (−31 to 35)^a	−16 (−71 to 38)	RANI‐PL
77 (18 to 137)	191 (127 to 256)	123 (67 to 179)	153 (97 to 208)	134 (55 to 213)	150 (87 to 214)	SHAM
^a P value for differences between direct and indirect estimates = 0.003. ^b P value for differences between direct and indirect estimates = 0.044. ^* P value for heterogeneity = 0.002; I² = 80% in the direct meta‐analysis. ^{^} P value for heterogeneity = 0.000; I² = 91% in the direct meta‐analysis. P value for overall inconsistency = 0.209 in the network meta‐analysis.

Table 5. Mean central retinal thickness change at 12 months: direct (upper‐right triangle) and mixed (lower‐left triangle) estimates

Table 6. Network structure: safety at the longest available follow‐up

	Laser	Aflibercept	Bevacizumab	Pegaptanib	Ranibizumab	Sham	Overall
SSAE	9	3	6	2	10	5	18
	1013	556	410	186	1303	528	4229
ATC*	10	3	4	2	8	2	15
	824	846	330	188	1113	184	3718
Death	11	3	4	2	10	3	17
	903	846	333	188	1521	434	4455
For each safety outcome, numbers in the table are the total number of studies (upper line for each outcome) and the total number of eyes (lower line for each outcome), as available by treatment and measured at the longest available follow‐up. (*) combined incidence of (1) cardiovascular, hemorrhagic, and unknown death; (2) nonfatal MI; and (3) nonfatal stroke.

Table 6. Network structure: safety at the longest available follow‐up

Table 7. All serious systemic adverse events (longest available follow‐up)

CONTROL	0.95 (0.75 to 1.20)	1.29 (0.43 to 3.84)	1.02 (0.67 to 1.53)	0.98 (0.76 to 1.25)
0.98 (0.83 to 1.16)	AFLI	0.95 (0.75 to 1.20)		1.04 (0.83 to 1.32)
0.93 (0.73 to 1.19)	0.95 (0.76 to 1.18)	BEVA		0.96 (0.77 to 1.20)
1.02 (0.64 to 1.64)	1.04 (0.63 to 1.72)	1.09 (0.64 to 1.86)	PEGA
0.97 (0.80 to 1.17)	0.98 (0.82 to 1.19)	1.04 (0.84 to 1.28)	0.95 (0.57 to 1.58)	RANI
P value for overall inconsistency = 0.859.

Table 7. All serious systemic adverse events (longest available follow‐up)

Table 8. Antiplatelet Trialists Collaboration arterial thromboembolic events at the longest available follow‐up

CONTROL	1.50 (0.81 to 2.79)	0.92 (0.17 to 5.12)	0.78 (0.31 to 1.97)	0.64 (0.38 to 1.07)
0.88 (0.37 to 2.13)	AFLI	1.46 (0.71 to 2.98)		2.26 (1.15 to 4.23)^a
0.94 (0.33 to 2.66)	1.06 (0.36 to 3.11)	BEVA		1.51 (0.85 to 2.69)
0.79 (0.20 to 3.02)	0.89 (0.18 to 4.43)	0.83 (0.15 to 4.61)	PEGA
1.09 (0.52 to 2.29)	1.24 (0.48 to 3.19)^a	1.17 (0.43 to 3.13)	1.17 (0.43 to 3.16)	RANI
^a P value for differences between direct and indirect estimates = 0.002. P value for overall inconsistency = 0.274 in the network meta‐analysis.

Table 8. Antiplatelet Trialists Collaboration arterial thromboembolic events at the longest available follow‐up

Table 9. All‐cause mortality at the longest available follow‐up

CONTROL	1.69 (0.30 to 9.42)^a	0.95 (0.06 to 14.85)	0.82 (0.25 to 2.65)	0.64 (0.32 to 1.25)^d
1.01 (0.34 to 3.03)^a	AFLI	2.67 (0.97 to 7.37)^b		2.26 (0.80 to 6.40)^c
1.61 (0.45 to 5.69)	1.59 (0.43 to 5.94)^b	BEVA		0.85 (0.40 to 1.83)
0.81 (0.16 to 4.03)	0.81 (0.12 to 5.62)	0.51(0.07 to 3.90)	PEGA
0.90 (0.40 to 2.01)	1.16 (0.38 to 3.58)^c	0.73 (0.22 to 2.37)	1.44 (0.24 to 8.48)	RANI
^a P value for differences between direct and indirect estimates = 0.011. ^b P value for differences between direct and indirect estimates = 0.030. ^c P value for differences between direct and indirect estimates = 0.015. ^d P value for differences between direct and indirect estimates = 0.022. P value for overall inconsistency = 0.087 in the network meta‐analysis.

Table 9. All‐cause mortality at the longest available follow‐up

Table 10. Similarity among studies. baseline values and number of injections

Study	Participants	Interventions	Mean n. injections	Visual acuity (logMAR)	Retinal thickness (µm)	Study sponsor
Ahmadieh 2008	78	Bevacizumab Sham				None reported
BOLT 2010	80	Bevacizumab Laser	9* 3*	0.59 0.61	507 482	Public
DA VINCI 2011	89	Aflibercept Laser	3.6 to 5.5 1.7	0.55	441	Industry
DRCRnet 2010	668	Laser Ranibizumab‐DL Ranibizumab‐PL	3* 9* 8*	0.38 0.39 0.38		Public
DRCRnet 2015	620	Aflibercept Bevacizumab Ranibizumab	9.2 9.4 9.7	0.40	412 414 407	Public
Ekinci 2014	100	Bevacizumab Ranibizumab	5.1 6.5	0.22 0.24	484 490	Public
Ishibashi 2014	233	Pegaptanib Sham	4	0.56		Industry
Korobelnik 2014	268	Aflibercept Laser	8.5 2.4	0.50		Industry
Lopez‐Galvez 2014	83	Ranibizumab Laser	5.3 2.1			Industry
LUCIDATE 2014	33	Laser Ranibizumab	2.6 9	0.42 0.30	489 455	No details
Macugen 2005	86	Pegaptanib Sham	5 4.5	0.56 0.58	476 423	Industry
Macugen 2011	260	Pegaptanib Sham	8.3 8.4	0.56 0.58	442 465	Industry
RELATION 2012	128	Laser Ranibizumab‐PL				Industry
Nepomuceno 2013	60	Bevacizumab Ranibizumab	9.8 7.7	0.60 0.63	451 421	Public
READ2 2009	115	Laser Ranibizumab Ranibizumab‐PL	4.4 5.3 2.9	0.60 0.54 0.60	228 190 263	Industry
RESOLVE 2010	151	Ranibizumab Sham	10.2 8.9	0.50 0.48	455 449	Industry
RESPOND 2013	203	Laser Ranibizumab Ranibizumab‐PL	9.2 8.8	0.46 0.44 0.40	458 448 422	Industry
RESTORE 2011	343	Laser Ranibizumab Ranibizumab‐PL	7.3 7 6.8	0.46 0.40 0.42	412 427 416	Industry
REVEAL 2015	390	Laser Ranibizumab Ranibizumab‐PL	1.9 7.8 7	0.54 0.52 0.52	395 419 430	Industry
Soheilian 2007	85	Bevacizumab Laser		0.71 0.55	352 319	Public
Turkoglu 2015	70	Laser Ranibizumab		0.84 0.80	460 488	None reported
Wiley 2016	124	Bevacizumab Ranibizumab	3 3	0.42	477	Public
DL: plus deferred laser PL: plus prompt laser (*): median, not mean, available and reported

Table 10. Similarity among studies. baseline values and number of injections

Comparison 1. Ranibizumab versus laser photocoagulation at 6 to 12 months

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Quality of life: NEI‐VFQ composite score at 6 to 12 months Show forest plot	3	412	Mean Difference (Fixed, 95% CI)	5.14 [2.96, 7.32]

Comparison 1. Ranibizumab versus laser photocoagulation at 6 to 12 months