The main reasons for exclusion of full-text articles were absence of adverse events reporting or absence of details regarding adverse events.
See the Table footnote for the expanded names of the studies.
Diamonds represent pooled estimates of odds ratios (ORs) with horizontal width representing CIs. The size of the data markers indicates the relative weight of the study. See the Table footnote for the expanded names of the studies.
eTable 1. Search strategy
eTable 2. Summary statistics of anti-VEGF monoclonal antibodies vs control (sham or non-anti-VEGF therapy) comparisons for primary and secondary endpoints, and sub-group analyses
eTable 3. Sensitivity analysis for APTC criteria using different methods, in order to take into account zero cell events
eTable 4. Summary statistics of ranibizumab vs control studies for primary endpoints in length of follow-up subgroup analysis
eTable 5. Summary statistics of bevacizumab vs ranibizumab comparison for primary and secondary endpoints
eTable 6. Summary statistics of ranibizumab studies dose-response studies for primary and secondary endpoints
eTable 7. Sensitivity analysis for nonocular hemorrhage events using different methods, in order to take into account zero cell events
eFigure 1. Risk of bias graph (review authors’ judgments) about each risk of bias item presented as percentages across all included studies
eFigure 2. Funnel plot for the major cardiovascular events (APTC criteria) outcome
eFigure 3. Funnel plot for nonocular hemorrhage events
Thulliez M, Angoulvant D, Le Lez ML, Jonville-Bera A, Pisella P, Gueyffier F, Bejan-Angoulvant T. Cardiovascular Events and Bleeding Risk Associated With Intravitreal Antivascular Endothelial Growth Factor Monoclonal AntibodiesSystematic Review and Meta-analysis. JAMA Ophthalmol. 2014;132(11):1317-1326. doi:10.1001/jamaophthalmol.2014.2333
Few data exist regarding the systemic safety of intravitreal antivascular endothelial growth factor (anti-VEGF) monoclonal antibody (mAb).
To conduct a systematic review and meta-analysis to evaluate the risk of major cardiovascular and nonocular hemorrhagic events in patients with neovascular age-related macular degeneration (AMD), diabetes mellitus–associated macular edema (DME), or retinal vein occlusions (RVOs) who receive intravitreal anti-VEGF mAbs.
The MEDLINE and Cochrane Central databases were searched for potentially eligible studies.
Randomized clinical trials comparing ranibizumab or bevacizumab with no anti-VEGF treatment, as well as those comparing ranibizumab with bevacizumab in patients with AMD, DME, or RVOs.
Data Extraction and Synthesis
We used a fixed-effects model and report the results as odds ratios (ORs) and 95% CIs.
Main Outcomes and Measures
Primary end points were major cardiovascular and nonocular hemorrhagic events. Secondary end points were all-cause mortality, cardiovascular mortality, stroke, myocardial infarction, venous thromboembolic events (VTEs), and hypertension.
Twenty-one trials that evaluated 9557 patients were retrieved. Anti-VEGF mAbs did not significantly increase the risk of major cardiovascular events (OR, 1.18; 95% CI, 0.81-1.71) or nonocular hemorrhagic events (OR, 1.42; 95% CI, 0.95-2.13) in treatment groups compared with control populations. Bevacizumab did not increase the risk of major cardiovascular events (OR, 0.94; 95% CI, 0.59-1.52) or nonocular hemorrhagic events (OR, 2.56; 95% CI, 0.78-8.38) compared with ranibizumab, but significantly increased VTEs (OR, 3.45; 95% CI, 1.25-9.54). Subgroup analysis showed a significant increase of nonocular hemorrhagic events in patients with AMD in ranibizumab vs control trials (OR, 1.57; 95% CI, 1.01-2.44). Anti-VEGF mAbs did not significantly increase overall mortality, cardiovascular mortality, stroke, myocardial infarction, VTEs, or hypertension.
Conclusions and Relevance
We showed that intravitreal anti-VEGF-mAbs were not associated with significant increases in major cardiovascular or nonocular hemorrhagic events, but studies and meta-analyses were not powered enough to correctly assess these risks. Increased risks of VTEs with bevacizumab and nonocular hemorrhagic events in older patients with AMD with ranibizumab should be cautiously interpreted because more safety data are needed.
Neovascular age-related macular degeneration (AMD) is the leading cause of vision loss among elderly people in developed countries.1- 3 Diabetes mellitus–associated macular edema (DME) is the main cause of vision loss in the working age population, followed by retinal vein occlusion (RVO).4 Treatment of these diseases is of major importance in delaying vision loss in this elderly patient population, and therefore in providing functional benefit. Treatment options for neovascular AMD include laser photocoagulation and verteporfin photodynamic therapy; treatment options for DME and RVO include laser photocoagulation and intravitreal injection of corticosteroids.
The vascular endothelial growth factor (VEGF)–A isoform is a cytokine that promotes angiogenesis and vascular permeability.5 Expression of VEGF is upregulated in pathologic conditions such as hypoxia in regions of the ischemic retina6- 8 or hyperglycemia.9 Several anti-VEGF treatments are available for treatment of macular edema: pegaptanib sodium, aflibercept, and 2 monoclonal antibodies (mAbs): ranibizumab and bevacizumab. Ranibizumab, a humanized mAb fragment, is the only approved mAb for treatment of AMD, DME, and RVO in Europe and the United States.10 Bevacizumab, a full-length humanized antibody, is approved for the treatment of metastatic solid cancers11 but is widely used as an off-label treatment for AMD, DME, and RVO. Its off-label use is worthwhile because of its lower cost compared with other treatments12 and comparable efficacy.13 Anti-VEGF agents administered by intravitreal injection block the action of VEGF-A isoforms, inhibit VEGF-driven neovascularization,14 and have shown efficacy in preserving visual acuity in AMD,15 DME,16 and RVO.17,18
However, the systemic safety of these intravitreal agents is unknown. Systemic use of bevacizumab in colorectal cancer therapy has been associated with serious cardiovascular adverse effects, such as hypertension,19 arterial thromboembolic events,20 hemorrhage,21 and death.22 Because intravitreal antiangiogenic agents have been associated with detectable levels in the systemic circulation,23,24 there is a rationale for the potential occurrence of systemic adverse events. Although intravitreal bevacizumab is administered at a dose of 1.0 to 2.5 mg (150 times less than the systemic dose used in cancer),25 VEGF inhibition may induce systemic adverse effects that could be serious for patients with diabetes or elderly patients who are at increased risk for cardiovascular adverse events.26 Moreover, some clinical trials27 suggested that intravitreal use of ranibizumab was associated with a small increase in nonocular hemorrhage risk.
To address these issues, we performed a systematic review and meta-analysis of clinical trials to investigate the risk of cardiovascular adverse events and nonocular hemorrhage associated with intravitreal use of the anti-VEGF mAbs ranibizumab and bevacizumab in patients with wet AMD, DME, and RVO.
Studies were identified by searching MEDLINE and Cochrane Central Register of Controlled Trials databases from inception until June 30, 2013, without language restrictions. The following key words were used: bevacizumab, ranibizumab, intravitreal, clinical trial, and randomized controlled trial. We also reviewed the reference lists of meta-analyses and selected studies (eTable 1 in the Supplement).
The selection of eligible studies was done by one author (M.T.). Inclusion criteria were parallel randomized clinical trials comparing intravitreal ranibizumab or bevacizumab with no treatment (sham) or a non-antiangiogenic treatment in patients with wet AMD, DME, or RVO. Trials that compared different treatment regimens of ranibizumab or bevacizumab were also included in this systematic review for a dose-response analysis. To address clinically relevant cardiovascular outcomes as well as mortality, we only included studies with a minimum 3-month follow-up period.
Two authors (M.T. and T.B.-A.) assessed the methodologic quality of the selected trials according to the Cochrane risk of bias criteria. We considered the following domains: (1) random sequence generation (selection bias), (2) allocation concealment (selection bias), (3) masking of participants and personnel (performance bias), and (4) masking of outcome assessment (detection bias) for adverse events. We considered the risk of bias to be low if masking of participants, personnel, and outcome assessment was adequate; otherwise, the risk of bias was considered to be unknown or high.
Our main end points were major cardiovascular events using the Antiplatelet Trialists' Collaboration (APTC) criteria28 and nonocular hemorrhage events. The APTC end point is a composite of nonfatal myocardial infarction, nonfatal ischemic or hemorrhagic stroke, or death due to a vascular or unknown cause. Secondary end points included all-cause mortality, cardiovascular mortality, stroke, myocardial infarction, venous thromboembolic events (VTEs), arterial hypertension, and proteinuria.
End point data from eligible trials were extracted by one author (M.T.), with a full review of the data extracted by a second author (T.B.-A.), and differences were adjudicated by both authors. We extracted data from the longest follow-up period whenever possible and if fewer than 10% of patients crossed over from the control to active treatment group. When crossover was above 10% we included only data collected before the crossover.
We extracted aggregate data from published reports. We report the results as odds ratios (ORs) with 95% CIs. We conducted a fixed-effects meta-analysis using the Peto method29 because it is more powerful and less biased in cases of low event rates and no significant imbalance between treatment groups.
Our main comparison was anti-VEGF treatment vs control. In trials that evaluated 2 or more doses of the same mAb we preserved randomization but collapsed the different dose intervention arms (eg, ranibizumab, 0.3 mg and 0.5 mg) into single treatment arms. Secondary comparisons were bevacizumab vs ranibizumab and high-dose vs low-dose regimens. This latter comparison was possible only for studies that evaluated 2 or more doses of the same mAb (ranibizumab-only studies).
Statistical heterogeneity across trials was assessed with χ2 and I2 tests. Heterogeneity was considered significant if the P value was <.1 and considered high if the I2 value was above 50%. We planned subgroup analysis to investigate the effect of different covariates on outcome measures: the type of mAb used (ranibizumab or bevacizumab), type of disease (AMD, DME, or RVO), follow-up duration, and study quality.
For primary outcomes, we conducted sensitivity analysis using the fixed Mantel-Haenszel method with a classical (0.5) and a treatment arm continuity correction as described by Sweeting et al30 and with a logistic method. We performed sensitivity analyses to evaluate the impact of studies in which the control treatment was known to be associated with adverse cardiovascular events (eg, verteporfin).
Publication bias was assessed by examination of the funnel plot asymmetry. The rank correlation test and the weighted linear regression test were used to test for funnel plot asymmetry. Statistical analyses were performed using Revman, version 5.1, and R software, version 2.11.131 (the meta package32).
The number of studies identified at each stage of the systematic review is shown in Figure 1. After removing duplicate references, the searches identified 780 records. According to our selection criteria, 21 randomized clinical trials13,17,18,27,33- 54 were retrieved including 9557 patients (Table).
Twelve studies13,27,33- 46 included patients with AMD (6616 patients; mean age, 78 years) and compared ranibizumab vs control (4 studies), 2 doses of ranibizumab (3 studies), bevacizumab vs control (1 study), and ranibizumab vs bevacizumab (4 studies).
Seven studies47- 52 included patients with DME (2152 patients; mean age, 63 years) and compared ranibizumab vs control (6 studies) or bevacizumab vs control (1 study). Two studies55,56 were excluded because 2 eyes per patient were possibly randomized and data for adverse events were reported by studied eye rather than by patient. We decided to retain data from the Elman et al50 study, even if 2 eyes were possibly randomized, because the authors reported adverse events data by participants. Two studies17,18,53,54 included patients with RVO (n = 789) and evaluated ranibizumab vs sham injections.
Comparison between ranibizumab and control treatment included 12 studies (n = 4346),17,18,27,33- 38,47- 51,53,54 between bevacizumab and control included 2 studies (n = 332),42,52 and between bevacizumab and ranibizumab included 4 studies (n = 2181).13,43- 46 Follow-up for adverse events was 24 months in 5 studies, 12 months in 13 studies, and less than 12 months in 3 studies. Ten studies compared a high dose with a low dose of ranibizumab: either 0.5 mg or 0.3 mg on a monthly basis, or the same dose in a monthly vs quarterly regimen. For this comparison longer follow-up was possible.
Twelve studies (57%) were considered to be at low risk regarding consideration of both performance and detection bias (Table, Figure 2, and eFigure 1 in the Supplement). Selection bias was judged at low risk in 13 studies (62%) and unknown (information missing) in 8 studies (38%).
Anti-VEGF mAb treatment did not significantly increase the risk of major cardiovascular events (APTC criteria) compared with control treatment, with no significant heterogeneity (OR, 1.18; 95% CI, 0.81-1.71; P = .38; I2 = 0%) (Figure 3 and eTable 2 in the Supplement). No asymmetry was observed in the funnel plot (eFigure 2 in the Supplement). The results did not change in sensitivity analysis when different methods to pool the data were used (eTable 3 in the Supplement) or when trials with active verteporfin treatment were excluded (OR, 1.12; 95% CI, 0.76-1.67; P = .56). We found no significant effect of follow-up duration in ranibizumab studies (P = .97 for interaction) (eTable 4 in in the Supplement). The type of disease, type of mAb used, or quality of the studies did not significantly influence treatment effect (P =.98, P =.40, and P = .38 for interaction, respectively) (eTable 2 in the Supplement). No significant difference was observed regarding the risk of major cardiovascular events in the 3 trials directly comparing bevacizumab with ranibizumab (OR, 0.94; 95% CI, 0.59-1.52; P = .81; I2 = 43%) (eTable 5 in the Supplement). Low-dose ranibizumab was not associated with a lower risk compared with a high dose of the drug (OR, 0.86; 95% CI, 0.62-1.21; P = .40; I2 = 0%) (eTable 6 in the Supplement).
Anti-VEGF mAb treatment did not significantly increase the risk of nonocular hemorrhage events when compared to control, with no significant heterogeneity (OR, 1.42; 95% CI, 0.95-2.13; P = .09; I2 = 0%) (Figure 4, eTable 2 in the Supplement). No asymmetry was observed in the funnel plot (eFigure 3 in the Supplement). The results obtained with sensitivity analysis did not change when different methods were used to pool the data (eTable 7 in the Supplement). We found no significant impact of follow-up duration in ranibizumab studies (P = .48) (eTable 4 in the Supplement). The type of disease or study quality did not influence the treatment effect (P = .16 and P = .75 for interaction, respectively) (eTable 2 in the Supplement). We observed a significantly increased risk of nonocular hemorrhage events in patients with AMD (OR, 1.57; 95% CI, 1.01-2.44; P = .04; I2 = 0%) but not in those with DME (OR, 0.54; 95% CI, 0.17-1.74; P = .31; I2 = 0%) or RVO (OR, 4.50; 95% CI, 0.40-50.07; P = .22). A nonsignificant increase of nonocular hemorrhage was observed with bevacizumab in the only trial13 comparing bevacizumab with ranibizumab and reported events in patients with AMD (OR, 2.56; 95% CI, 0.78-8.38; P = .10) (eTable 5 in the Supplement). Low-dose ranibizumab was not associated with a lower risk compared with high-dose ranibizumab (OR, 0.92; 95% CI, 0.67-1.26; P = .61; I2 = 0%) (eTable 6 in the Supplement).
Anti-VEGF treatment did not significantly increase the risks of overall mortality (OR, 1.53; 95% CI, 0.92-2.56; P = .10; I2 = 0%), cardiovascular mortality (OR, 1.29; 95% CI, 0.70-2.37; P = .42; I2 = 0%), stroke (OR, 1.61; 95% CI, 0.85-3.05; P = .14; I2 = 0%), myocardial infarction (OR, 0.92; 95% CI, 0.54-1.59; P = .77; I2 = 2%), hypertension (OR, 0.97; 95% CI, 0.71-1.32; P = .84; I2 = 6%), or VTEs (OR, 1.39; 95% CI, 0.17-11.38; P = .76; I2 = 0%) (eTable 2 in the Supplement). Proteinuria was rarely reported and only in ranibizumab trials. In trials comparing bevacizumab vs ranibizumab VTEs were significantly increased with bevacizumab (OR, 3.45; 95% CI, 1.25-9.54; P = .02; I2 = 0%) (eTable 5 in the Supplement). Low-dose ranibizumab was associated with a nonsignificantly lower risk of stroke compared with high-dose ranibizumab (OR, 0.59; 95% CI, 0.34-1.04; P = .07; I2 = 10%) (eTable 6 in the Supplement).
To our knowledge, this systematic review and meta-analysis is the first specifically investigating systemic cardiovascular and hemorrhagic adverse events associated with intravitreal administration of anti-VEGF mAbs in a large population of patients included in randomized clinical trials. We considered studies that included patients with AMD, DME, or RVO to increase the power to detect safety signals and because these diseases are the only approved indications for intravitreal anti-VEGF treatment. Although cardiovascular risks may differ among these populations, randomization allows group comparability and relative risk estimation. These conditions are also associated with a high cardiovascular risk (age, diabetes, and associated cardiovascular risk factors). Anti-VEGF treatment adverse vascular events are therefore more likely to be detected in this population at high risk for cardiovascular events.
Our results suggest that intravitreal administration of the anti-VEGF mAbs ranibizumab or bevacizumab was not associated with an increased composite APTC end point compared with control treatments (sham, laser, and other non–anti-VEGF interventions). The effect on each component of the composite end point was not homogenous. We observed nonsignificant increases in stroke and cardiovascular death risks, but no effect on myocardial infarction. No increased risk of hypertension was apparent, but this end point was heterogeneously reported in clinical trials. The nonsignificant increase in stroke risk observed in our meta-analysis is consistent with previous findings in a pooled analysis of 5 studies in patients with AMD.57 Controversial results were published regarding the risk of myocardial infarction and stroke in patients treated with intravitreal anti-VEGF mAbs.58- 62 All of these studies were observational (case-control or retrospective cohorts) and therefore subject to biases even if adjustment for confounding factors was performed in some of them.
We did not observe any significant differences in APTC risk or in its components between bevacizumab and ranibizumab despite a rationale for a potential risk increase with bevacizumab. Both ranibizumab and bevacizumab undergo systemic passage after intravitreal injection, but only bevacizumab was associated with a persistent decrease in plasma levels of VEGF in patients with AMD and DME.24,63 This is consistent with bevacizumab’s pharmacologic profile as a full mAb with a half-life longer than that of ranibizumab. Furthermore, experiments64 in animal models suggested that bevacizumab may increase vascular inflammation and platelet activation and therefore the development of thrombosis.
Nonocular hemorrhagic events were not significantly increased with ranibizumab compared with control groups. No hemorrhagic events were reported in bevacizumab vs control studies. The increase in nonocular hemorrhagic risk was significant in patients with AMD who received ranibizumab, consistent with the MARINA27 study results and a recent meta-analysis by Schmucker et al.65 This finding could be explained by the confounding effect of age, a factor known to increase bleeding risk in medically ill patients.66 No significant hemorrhagic risk was apparent in patients with DME or RVO, but the number of reported events was low.
To our knowledge, this is the first meta-analysis to report VTE risk with intravitreal anti-VEGF treatments compared with control treatments. Only 2 studies49,51 reported 4 VTE events with ranibizumab in patients with DME, showing a nonsignificant increase with a very wide CI. When combined, 2 studies44,45 showed a significant increase in VTE risk with bevacizumab when directly compared with ranibizumab. An increased VTE risk associated with systemic bevacizumab in patients with cancer has been reported.67
A nonsignificant increase in total mortality was apparent with intravitreal anti-VEGF mAbs compared with control treatment, consistent with both ranibizumab and bevacizumab, but this finding should be interpreted with caution, given the limited statistical power of the included studies. A previous meta-analysis22 showed a significant increase in bevacizumab-related mortality in patients with cancer mainly because of hemorrhagic events, but also because of VTE and stroke; however, doses of bevacizumab were much higher and were administered by the systemic route.
We acknowledge several limitations of our meta-analysis. First, cardiovascular and hemorrhagic events were secondary safety outcomes, and therefore inherently subject to potential detection or reporting bias. These biases were difficult to evaluate because included studies contained limited information on how harms were reported. Four studies only mentioned arteriothrombotic or thromboembolic events as being specifically assessed. Several studies reported zero events, which could be problematic, but our results were consistent even when using different methods to pool the data. The present review focused on published clinical trial data; publication bias resulting from unpublished trials cannot be excluded even if all tests for funnel plot asymmetry were nonsignificant. We included data from the Elman et al50 study, despite reporting of adverse events according to study participants rather than eyes randomized. This resulted in a received-treatment and not intention-to-treat analysis. We believed that the sample size of this study justified its inclusion, even if it could generate potential bias. However, excluding this study did not change our final results (data not shown).
Finally, our results should be interpreted as safety signals that need to be confirmed. Indeed, included studies were of small sample size and therefore not powered enough to show an increase in adverse events risk. Furthermore, the multiplicity of comparisons in this meta-analysis could have led to spurious findings. By using the Framingham risk score we estimated that the baseline risk score of patients with AMD would be approximately 3.5% annually for major cardiovascular events. In this hypothesis, more than 20 000 patients would be necessary to have 80% power to show a 19% increase in APTC risk by anti-VEGF treatment; this population is far more than the 4162 patients included in the APTC evaluation in the present meta-analysis. The lack of statistical significance of our results may be the result of a lack of effect of these treatments on cardiovascular events, or, as mentioned above, a lack of power of the analysis. Furthermore, the long-term effect of these treatments (>2 years) needs to be evaluated.
Our meta-analysis suggests that intravitreal administration of anti-VEGF mAbs is not associated with significant increases in risks of systemic cardiovascular and hemorrhagic events or in overall mortality, cardiovascular mortality, or stroke in elderly patients. However, some safety signals, such as nonocular hemorrhagic risk in older patients with AMD observed with ranibizumab and VTE risk with bevacizumab, warrant continued monitoring in sufficiently powered studies. Studies of these safety risks are needed to establish the relative safety of off-label use of bevacizumab compared with ranibizumab and of both drugs compared with placebo.
Submitted for Publication: December 2, 2013; final revision received April 1, 2014; accepted April 24, 2014.
Corresponding Author: Theodora Bejan-Angoulvant, MD, PhD, Pharmacologie Clinique, Centre Hospitalier Regional Universitaire de Tours, Hôpital Bretonneau, 2 Bd Tonnellé, Tours 37044 CEDEX 9, France (email@example.com).
Published Online: July 24, 2014. doi:10.1001/jamaophthalmol.2014.2333.
Author Contributions: Dr Bejan-Angoulvant had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Thulliez, Bejan-Angoulvant.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Thulliez, Angoulvant, Bejan-Angoulvant.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Thulliez, Bejan-Angoulvant.
Administrative, technical, or material support: Jonville-Bera.
Study supervision: Angoulvant, Pisella, Gueyffier, Bejan-Angoulvant.
Conflict of Interest Disclosures: Dr Angoulvant receives personal fees from AstraZeneca, Novartis, Lilly, MSD, Servier Laboratories, Amgen, and Bayer; grants from Lilly; and nonfinancial support from MSD; no funds were received for the present study. Dr Gueyffier receives nonfinancial support from Servier Laboratories; grants from Teva, Bristol-Myers Squibb, Lilly, Janssen-Cilag, UCB Pharma, Novartis, Urgo Pharmaceutical, Schering-Plough, Novo Nordisk, Trophos, and Teikoku Pharma; and has Novadiscovery shares; no funds were received for the present study. No other disclosures are reported.
Additional Contributions: Clémence Bourgeois, BSc (Pharmacology Department, Centre Hospitalier Regional Universitaire de Tours) assisted with review of some of the selected publications. Gilles Paintaud, MD, PhD, and Hervé Watier, MD, PhD, (MabImprove Labex) provided support and advice. There was no financial compensation for the services.