Different data marker sizes indicate weight. See Table 1 for definition of the abbreviations.
eMethods. Parameters Used for Online Literature Search
eTable. Risk for of Bias Assessment
eFigure 1. Anti-VEGF Treatment and Vascular Deaths
eFigure 2. Anti-VEGF Treatment and Cerebrovascular Accidents (CVAs)
eFigure 3. Anti-VEGF Treatment and Myocardial Infarction (MI)
eFigure 4. Anti-VEGF Treatment and Arteriothrombotic Events (ATEs)
eFigure 5. Anti-VEGF Treatment and Death Excluding a RISE Protocol Violation Patient
eFigure 6. Anti-VEGF Treatment and Cerebrovascular Accidents (CVAs) Excluding a RISE Protocol Violation Patient
eFigure 7. Anti-VEGF Treatment and Vascular Death Excluding a RISE Protocol Violation Patient
eFigure 8. Anti-VEGF Treatment and Vascular Death When All 3 Deaths Are Vascular in VISTA Sham Arm
Avery RL, Gordon GM. Systemic Safety of Prolonged Monthly Anti–Vascular Endothelial Growth Factor Therapy for Diabetic Macular EdemaA Systematic Review and Meta-analysis. JAMA Ophthalmol. 2016;134(1):21-29. doi:10.1001/jamaophthalmol.2015.4070
Anti–vascular endothelial growth factor (VEGF) therapy is commonly used to treat numerous retinal conditions and appears safe, yet controversy remains regarding systemic safety.
To evaluate the systemic safety of intravitreous anti-VEGF injections in high-risk patients with diabetic macular edema (DME) and to investigate separately the subgroup of these patients with the highest level of exposure to anti-VEGF monthly treatment for 2 years.
A search of MEDLINE, Cochrane Central Register of Controlled Trials, clincaltrials.gov, and ophthalmology congress abstracts January 1, 1947, to May 19, 2015.
Randomized clinical trials were selected that evaluated monthly anti-VEGF injections for DME for 2 years and reported the outcome measures of cerebrovascular accidents, myocardial infarctions, arteriothrombotic events, and mortality.
Data Extraction and Synthesis
Two reviewers collected data independently from each study for the meta-analysis. Data were pooled using a fixed-effects model and analyzed from November 6, 2014, to June 28, 2015. Peto odds ratios with 95% CIs were calculated.
Main Outcomes and Measures
Primary end points included cerebrovascular accidents and all-cause mortality in the highest-dose arms. Secondary outcomes included myocardial infarctions, arteriothrombotic events, and vascular-related death.
Of 1126 articles reviewed, 598 were removed as duplicate studies and 524, for lack of monthly treatment data for 2 years, leaving 4 studies for the meta-analysis that met the search criteria: 2 trials using monthly aflibercept and 2 using monthly ranibizumab, representing 1328 patients. The primary evaluation (1078 patients) combined the monthly aflibercept and the 0.5-mg ranibizumab arms and yielded an increased risk for death compared with sham and laser treatments (odds ratio [OR], 2.98; 95% CI, 1.44-6.14; P = .003). Analysis including monthly aflibercept and 0.5-mg ranibizumab yielded an increased risk for cerebrovascular accidents (OR, 2.33; 95% CI, 1.04-5.22; P = .04) and vascular death (OR, 2.51; 95% CI, 1.08-5.82; P = .03). No definitive increased risk for myocardial infarctions and arteriothrombotic events was seen with all dose combinations.
Conclusions and Relevance
In this meta-analysis of anti-VEGF agents for patients with DME, assessment of the highest-level exposure group (those high-risk patients with DME who received 2 years of monthly treatment) revealed a possible increased risk for death and potentially for cerebrovascular accidents. Consideration of total exposure to anti-VEGF agents when treating those at high risk for vascular disease may be important.
The intravitreous use of anti–vascular endothelial growth factor (VEGF) agents has transformed the treatment of many retinal diseases.1- 5 Despite systemic safety concerns of these agents when used intravenously for cancer treatment, the much smaller intravitreous dose has been found to be safe in numerous registered trials.6- 12 However, these trials are not powered to evaluate the safety risks of uncommon events, and many trials10,12- 14 have excluded patients at highest risk of systemic adverse effects of these agents (ie, those with recent cerebrovascular accident [CVA] or myocardial infarction [MI]) from enrollment. Meta-analyses have been performed to evaluate larger populations of patients and have demonstrated safety when evaluating the entire study population; however, there may be increased risk in certain high-risk subgroups.1,2,4,15- 20 Data demonstrate a reduction in circulating VEGF levels after intravitreous injection of these agents, providing biological plausibility for possible systemic effects.21- 26
Epidemiologists often look for factors associated with an outcome in high-risk groups by studying the groups with the highest levels of exposure. Because systemic administration of these agents for cancer is associated with increased risks for arteriothrombotic events (ATEs), we first chose persons with diabetes mellitus as a high-risk group given their propensity for ATEs. This choice was also supported by a recent meta-analysis of ranibizumab treatment for age-related macular degeneration (AMD), retinal vascular occlusion, and diabetic macular edema (DME)18 that found imbalances in the risks for death, CVA, and impaired wound healing, but only in DME. Second, several meta-analyses of ranibizumab treatment for AMD17,27 have identified an increased risk for CVA but only in the higher-dose (0.5 mg) arm or only with monthly instead of as-needed dosing. A recent Cochrane meta-analysis of anti-VEGF for DME2 analyzed 15 trials and showed no increased risk for ATEs or death. The intent of the present analysis was to evaluate the subpopulation of patients with DME who had the highest levels of exposure to anti-VEGF agents, specifically, monthly injections for 2 years.
Because of controversy regarding the systemic safety of anti–vascular endothelial growth factor (VEGF) commonly used to treat diabetic macular edema (DME), this study evaluated a pooled analysis on systemic safety of intensive intravitreous anti-VEGF injections for DME for 2 years.
Although several meta-analyses evaluating the entire population of patients in anti-VEGF trials with different treatment regimens have not shown safety signals, patients receiving maximum monthly doses had an increased risk for death compared with those receiving sham and laser treatments.
In an analysis of pooled patients receiving monthly aflibercept and different doses of monthly ranibizumab (0.5 mg, 0.3 mg, or both), the Peto odds ratios (95% CI) were 2.98 (1.44-6.14; P = .003), 2.56 (1.17-5.57; P = .02), and 2.57 (1.31-5.05; P = .006), respectively, for an increased risk for death.
These data suggest an increased risk for death in a select, intensively treated population receiving monthly anti-VEGF treatment for 2 years.
We conducted a systematic search without language restrictions of MEDLINE, the Cochrane Central Register of Controlled Trials, and clinicaltrials.gov from January 1, 1947 (start date for MEDLINE) through May 19, 2015, for randomized clinical trials evaluating anti-VEGF agents for DME. In addition, we reviewed ophthalmology congress abstracts (Association for Research in Vision and Ophthalmology, American Academy of Ophthalmology, and European Society of Retina Specialists) and contacted representatives from Genentech, Novartis, and Regeneron to determine whether unpublished studies met selection criteria. Search details of the predefined protocol are available in the eMethods in the Supplement and further described in the PRISMA diagram in Figure 1. Randomized clinical trials were included if they met the eligibility criteria of 2 years of monthly anti-VEGF therapy for DME and reported the safety outcomes of CVAs, MIs, ATEs, and mortality.
Data regarding study quality, drug dosing, incidence of CVAs, MIs, ATEs, and all-cause and vascular-related deaths were collected from each study independently by 2 reviewers (R.L.A. and G.M.G.). When the causes of death were provided, we applied the definition of vascular death used by the Antiplatelet Trialists’ Collaboration (APTC): “definitely or possibly vascular, which includes all deaths attributed to cardiac, cerebral, hemorrhagic, embolic, other vascular, or unknown causes.”28(p83) Most disparities were resolved by communication with trial sponsors; alternative calculations for potential interpretations were performed to address unresolved disparities. The assessment of risk for bias followed the recommendations of the Cochrane Collaboration.29
Data were analyzed from November 6, 2014, to June 28, 2015. The clinical and methodological characteristics of the included trials were evaluated following Cochrane Collaboration guidelines, and statistical analysis was performed using RevMan 5.3 software.29 We used a fixed-effects method with the Peto odds ratio (OR) and 95% CI, which has been reported to have superior statistical properties to analyze uncommon events.29 The ranibizumab trials used 2 different monthly doses: 0.3 mg, approved in the United States and 0.5 mg, approved outside the United States. The prespecified primary analysis included the 0.5-mg arms (high dose) with monthly aflibercept; however, given the importance of both doses, secondary analyses combined monthly aflibercept treatment with the 0.3-mg ranibizumab arms (low dose) or with the 0.3- and 0.5-mg ranibizumab arms (combined dose). Both drugs are presented in the same forest plot to test for subgroup differences. Because several prior studies17,18,20,27 had raised concern about an increased risk for CVA and death, these constituted the primary outcome measures. We used the GRADE (Grading of Recommendations Assessment, Development and Evaluation) system to assess the quality of the evidence.30 Heterogeneity was assessed by calculating the I2 statistic and χ2 tests.31
Evaluation of recent meta-analyses of the safety of anti-VEGF use in DME2,4,15,20 has not shown a safety signal in the entire population. The most recent Cochrane analysis of 15 trials2 showed no increased risk for ATEs or death.
The database query concerning persons with DME at the highest level of exposure to anti-VEGF revealed 1126 titles, 598 of which were duplicates. After we reviewed 528 titles and abstracts, we evaluated 20 full-length articles and 1 presentation, which yielded 4 reports of 4 studies that met the inclusion criteria (Figure 1). A review of abstracts allowed exclusion of the majority of studies and was followed by a full-text review of the remaining titles.10- 13,32- 48 Two studies had been published with 1-year results only, but the 2-year follow-up had been presented, and a detailed summary of 2-year results is available from Regeneron (Intravitreal aflibercept injection for diabetic macular edema: primary results [unpublished data; December 4, 2014]).12,13 The excluded studies, including all studies of pegaptanib sodium and bevacizumab, did not have monthly treatment for 2 years.
Table 1 shows the characteristics of the 4 included trials, representing 1328 patients.10- 13 They were all randomized, registered trials sponsored by pharmaceutical companies with 2-year completion rates ranging from 105 of 135 (77.8%) to 110 of 127 (86.6%) for individual study arms. Two essentially identical studies, RISE (Study of Ranibizumab Injection in Subjects With Clinically Significant Macular Edema [ME] With Center Involvement Secondary to Diabetes Mellitus)10,11 and RIDE (Study of Ranibizumab Injection in Subjects With Clinically Significant Macular Edema [ME] With Center Involvement Secondary to Diabetes Mellitus),10,11 evaluated 2 monthly doses of ranibizumab vs a sham injection or laser treatment (sham/laser treatment), and 2 essentially identical studies, VISTA (Study of Intravitreal Aflibercept Injection in Patients With DME)12,13 and VIVID (Intravitreal Aflibercept Injection in Vision Impairment Due to DME),12,13 evaluated several dosing regimens of aflibercept vs sham/laser treatment. The entry criteria were similar except that the ranibizumab trials excluded patients with MI or CVA within 3 months of entry; the aflibercept trials excluded patients if the event was within 6 months of entry. No individual trial reported a significantly increased risk for ATE, MI, CVA, or death.
All trials were deemed of low risk with respect to allocation, blinding, selective outcome reporting, and other potential sources of bias. Attrition was deemed unclear in the aflibercept trials because the second year is unpublished, and details of incomplete outcome data remain unclear.13 However, the trial design is well documented in the 1-year publication12 (eTable in the Supplement).
Table 2 summarizes the findings for the main comparison, which included the monthly aflibercept and 0.5-mg ranibizumab doses. Figure 2 shows all-cause mortality in the 3 analyses combining monthly aflibercept and 0.5-mg, 0.3-mg, and combined monthly ranibizumab doses. The risk for death (OR; 95% CI) in aflibercept-ranibizumab treatment compared with sham/laser treatment increased in the 0.5-mg (2.98; 1.44-6.14; P = .003), 0.3-mg (2.56; 1.17-5.57; P = .02), and combined (2.57; 1.31-5.05; P = .006) ranibizumab injection groups. The subgroups receiving monthly 0.5-mg doses of ranibizumab with aflibercept individually demonstrated increased mortality (OR, 3.24; 95% CI, 1.12-9.36; P = .03) when compared with those groups receiving sham/laser treatment (OR, 2.76; 95% CI, 1.02-7.46; P = .05) (Figure 2A).
We identified no definitive difference when we compared monthly ranibizumab and aflibercept subgroups for death or any other outcome in this study. In all analyses within this study, outcome measures were consistent across all trials without significant heterogeneity as measured by the I2 statistic and χ2 tests.
The evaluation of vascular mortality is shown in eFigure 1 in the Supplement. For this analysis, the APTC definition of vascular death was applied to the reported causes of death. A disparity was noted in the VISTA findings. Three vascular deaths were reported in the monthly aflibercept arm; however, in another part of the report, causes of death of 6 patients in this arm were listed as cardiac arrest, acute cardiac failure, MI, pulseless electrical activity, CVA, and “death.”13 After discussion with the VISTA sponsors, part of the disparity is explained by their prespecified definition requiring that treatment-emergent events must occur within 30 days of the previous injection. Only treatment-emergent events were presented in their safety analysis set and, apparently, 2 patients had vascular deaths more than 30 days after the last aflibercept injection despite being in the monthly arm. In addition, the VISTA Study team and Regeneron will reclassify 1 patient as having a vascular-related death.13 For our analysis, we considered these 6 deaths to be vascular per the APTC definition, but alternative calculations are discussed in the Sensitivity Analyses subsection below. The treatment-emergent ATEs reported in the ranibizumab trials included events throughout the trials, including those occurring more than 30 days after an injection. The meta-analysis including monthly aflibercept and the 0.5-mg or combined ranibizumab dose arms showed an increased risk for vascular-related death in the 0.5-mg ranibizumab group (OR, 2.51; 95% CI, 1.08-5.82; P = .03) and in the combined-dose ranibizumab group (OR, 2.23; 95% CI, 1.01-4.89; P = .05) compared with sham/laser treatment.
The analysis of CVAs is shown in eFigure 2 in the Supplement. In the VIVID study, one patient died of brain herniation, which was confirmed to have been related to a CVA. Analysis of monthly aflibercept and the 0.5-mg ranibizumab dose revealed an increased risk for CVA compared with laser/sham treatment (OR, 2.33; 95% CI, 1.04-5.22; P = .04).
The meta-analysis with respect to MI and ATEs is shown in eFigures 3 and 4 in the Supplement. In the groups receiving 0.5-mg, 0.3-mg, and combined doses, no definitive increase in the risk for MI or ATE compared with sham/laser treatment was identified.
Outcome measures were consistent across all trials; however, in the RISE study, 1 patient assigned to sham/laser treatment was mistakenly treated with 1 dose of ranibizumab. This patient had a CVA and later died of an unknown cause that was classified as a vascular-related death. For safety purposes, the RISE study included this patient in the 0.5-mg ranibizumab arm. However, we also repeated the analysis placing this patient in the sham/laser treatment arm and found an increased risk for death (OR, 2.60; 95% CI, 1.26-5.38; P = .01) for the 0.5-mg dose group (eFigure 5A in the Supplement), whereas the risk for CVA had an OR of 1.97 (95% CI, 0.88-4.42; P = .10) (eFigure 6A in the Supplement) and the risk for vascular death had an OR of 2.09 (95% CI, 0.90-4.86; P = .09) (eFigure 7A in the Supplement).
Given the disparity in the number of vascular-related deaths in the VISTA study, we applied the APTC criteria to the listed cause of death. However, if fewer of these deaths had vascular causes, the analysis of the increased risk for vascular-related death decreases. If there are only 5 vascular-related deaths in the VISTA monthly arm instead of 6, the increased risk for vascular-related death in the high-dose analysis is an OR of 2.37 (95% CI, 1.00-5.63; P = .05). If only 4 deaths have vascular causes, the OR is 2.24 (95% CI, 0.93-5.41; P = .07). Also, if all 3 deaths in the sham/laser treatment arm are considered vascular related, including the patient with multiple-organ failure, the risk for vascular-related death has an OR of 2.21 (95% CI, 0.97-5.04; P = .06) (eFigure 8 in the Supplement). No other changes in results were noted among the remaining comparisons.
This study was designed to evaluate prolonged monthly anti-VEGF therapy in a high-risk group of patients to assess any potential systemic safety signal. We started by evaluating a group of patients at high risk for arteriothrombotic disease—those with diabetes mellitus. No excess risk for arteriothrombotic disease or mortality was seen in the recent overall Cochrane meta-analysis of patients with DME.2 However, when we focused on the group with the highest level of exposure, that is, a monthly injection of ranibizumab or aflibercept for 2 years, we found an increased risk for death when the analysis included aflibercept and the 0.3- or 0.5-mg ranibizumab dose. In the main high-dose analysis, we also found a borderline increased risk for CVA and vascular-related death. The authors of the RISE and RIDE studies10,11 noted that deaths and CVAs were numerically higher in ranibizumab groups but that this imbalance had not been seen in other studies, such as the Diabetic Retinopathy Clinical Research Network protocol I,42 which evaluated ranibizumab, laser, and triamcinolone for DME, or the RESTORE Extension Study, which compared ranibizumab injections with laser treatment for DME.11,45,49 However, these other trials did not require monthly injections, and the present analysis implies that this difference could be related to overall drug exposure. In fact, several meta-analyses of anti-VEGF agents for DME2,15,50 have not demonstrated a definitive safety signal, but these studies have predominantly evaluated trials with less frequent dosing or of shorter duration and thereby with less drug exposure. The recent Cochrane DME analysis2 did not use the monthly dosing arms from the VIVID or VISTA studies in the main analysis, and only 1-year data were available. Another meta-analysis of ranibizumab for DME20 did not find an increased risk for CVA, MI, ATEs, or mortality across all trials but noted a dose-dependent increased risk for mortality in the largest trial that used monthly ranibizumab. A recent meta-analysis of phase 2 and 3 aflibercept trials for multiple diseases,51 which accounted for duration of treatment, found the risk for death in treated patients with DME to be 1.70 per 100 patient-years at risk vs 0.55 per 100 patient-years at risk for sham/laser treatment.
In patients with AMD, 2 meta-analyses17,27 have observed an increased risk for CVA with the higher dose of ranibizumab. This increase was only seen in the high-risk patients when stratified for baseline CVA risk by Bressler et al.27 In the study by Ueta et al,17 the increased risk was seen not only with the higher dose but also with monthly dosing vs less frequent or as-needed treatment, although no increased risk for mortality was seen.
One potential explanation for the finding of an increased mortality risk in the present analysis but not in other larger meta-analyses is that this analysis focuses on a population with DME receiving monthly injections for 2 years, who thereby have a much higher level of drug exposure than patients in other studies. The finding of a possibly increased risk for CVA and vascular-related death in the higher-dose arm provides biological plausibility that the increased risk for all-cause mortality could be related to systemic vascular effects of the small doses of these agents rather than random events. Several studies21,23- 26 have documented that these agents pass into the bloodstream after intravitreous injection and reduce systemic VEGF levels, sometimes for prolonged periods. In 1 AMD study, 2 years of ranibizumab or bevacizumab injections significantly reduced serum VEGF levels, but more so in the monthly treatment arms.25 Persons with diabetes mellitus are particularly susceptible to ATEs, and long-term blockade of VEGF in the bloodstream could exacerbate this susceptibility and increase the risk for CVA and death.14,52 Concern over a potential safety signal in this population is not new because both trial sponsors14,52 excluded patients with recent ATEs from enrollment, the VISTA study12,13 stratified randomization by history of MI or CVA, and Genentech presumably sought approval of the lower ranibizumab dose for DME to lower the risk for a safety issue. Furthermore, any risk could be increased in patients undergoing intensive bilateral treatment.
This study has several limitations. First, we studied a subgroup of a subgroup (patients with diabetes mellitus), and only 4 trials are included. Individually, the trials are not large enough to show a difference in ATEs or death. However, in the meta-analysis of those with the highest levels of anti-VEGF exposure, increased risks are suggested. Although the risk for death was higher in this select subpopulation, in the evaluation of the risk for CVA or vascular-related death, only a borderline difference was found, and reclassification of a single patient can alter the confidence in the difference identified. Second, the 2-year data from 2 of the 4 studies have not yet been published. Fortunately, the first-year data have been published, including a detailed description of the study design. The main results of the second year have been presented and are available from the sponsor (Regeneron, unpublished data, December 4, 2014), who is submitting the study for publication.
Third, only study-level data are available rather than patient-level data. Therefore, we cannot adjust for baseline patient characteristics, which may not be balanced, leading to possible confounding. In addition, calculations of time to death cannot be performed; however, these imbalances did not materialize in the trials at 1 year, and Kaplan-Meier calculations have shown the separation of curves for CVA and death mostly during year 2 of the RISE and RIDE trials.2,12,53 Hence, most of the deaths occurred after a full year of monthly drug exposure, although this finding could be related to selection of patients in better health at study entry, such as excluding those with recent arteriothrombotic disease.
Fourth, all studies were sponsored by pharmaceutical companies and could have an increased risk for bias.54 Fifth, we compared across 2 different anti-VEGF agents. However, this comparison is common in meta-analyses, including those by the Cochrane Collaboration.2,19,50,55 In the current meta-analysis, the trials were similarly designed and registered with the US Food and Drug Administration, and outcomes were consistent across all trials as measured by heterogeneity tests. In all trials, patients with a recent MI or CVA were excluded, although with a different definition of recent (3 vs 6 months), which could introduce bias.11,12
This study looked at a limited subset of patients in clinical practice—the subgroup of a high-risk group with the highest level of anti-VEGF exposure and specifically persons with diabetes mellitus receiving monthly injections for 2 years. Many studies have shown excellent effects on DME with far fewer injections, such as 2 to 3 in the second year of protocol I in the Diabetic Retinopathy Clinical Research Network Study.42,56 Although patients in protocol T, which compared aflibercept, bevacizumab, and ranibizumab for DME, received 9 to 10 injections in the first year, year 2 will likely have fewer injections.48 The concern about a potential increased risk for death may not apply to most of the patients who are undergoing less intensive therapy, which is the standard of care in many clinical practices.
This study suggests a safety signal for patients at high risk for vascular disease who received intensive treatment for DME with monthly anti-VEGF therapy for 2 years. Further evaluation of these studies with patient-specific data and future larger studies are needed to better evaluate this signal. These increased risks were noted although patients with uncontrolled diabetes mellitus, hypertension, or recent CVA or MI were excluded from these trials. Furthermore, patients enrolling in clinical trials tend to be healthier and more adherent to treatment regimens than the general population. Hence, as clinicians, we should be increasingly cognizant of our patients’ medical histories when we decide how to treat them. Because these data provide possible evidence of a systemic adverse effect in aggressive treatment of at-risk patients, further studies of other at-risk populations, such as patients with AMD and previous CVAs, are indicated to assess whether this biologically plausible risk is seen in other high-risk, high-exposure groups.
Corresponding Author: Robert L. Avery, MD, California Retina Consultants and Research Foundation, 525 E Micheltorena, Ste A, Santa Barbara, CA 93103 (firstname.lastname@example.org).
Submitted for Publication: March 20, 2015; final revision received August 28, 2015; accepted September 3, 2015.
Published Online: October 29, 2015. doi:10.1001/jamaophthalmol.2015.4070.
Author Contributions: Drs Avery and Gordon had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Avery.
Acquisition, analysis, or interpretation of data: Both authors.
Drafting of the manuscript: Avery.
Critical revision of the manuscript for important intellectual content: Both authors.
Statistical analysis: Both authors.
Administrative, technical, or material support: Gordon.
Study supervision: Avery.
Conflict of Interest Disclosures: Both authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Dr Avery reported receiving personal fees from Alcon, Allergan, and Bausch and Lomb; personal fees and other from Alimera, Novartis, and Replenish; grants, personal fees, and nonfinancial support from Genentech; and grants, personal fees, and other from Regeneron, outside the submitted work. In addition, Dr Avery reports a patent intravitreous drug delivery licensed to Replenish. No other disclosures were reported.
Previous Presentations: This paper was presented in part at the 2015 Meeting of the Association for Research in Vision and Ophthalmology; May 4, 2015; Denver, Colorado; and at the annual meeting of the American Society of Retinal Specialists; July 14, 2015; Vienna, Austria.
Additional Contributions: Maureen Maguire, PhD, Perelman School of Medicine, University of Pennsylvania, Philadelphia, provided expertise in critically reviewing this manuscript. She did not receive any compensation for this role.