Vascular Safety of Ranibizumab in Patients With Diabetic Macular Edema: A Pooled Analysis of Patient-Level Data From Randomized Clinical Trials

Key Points

Question  What is the cardiovascular and cerebrovascular safety profile of ranibizumab, 0.5 mg and 0.3 mg, in patients with diabetic macular edema treated with various dosing regimens compared with sham and/or laser?

Findings  Pooled analysis of individual patient-level data involving 1767 patients from 6 phase 2 and 3 clinical trials in patients with diabetic macular edema showed that rates of cardiovascular and cerebrovascular events were comparably low in all arms.

Meaning  This analysis suggests that intravitreous administration of ranibizumab does not increase the risk of systemic vascular events in patients with diabetic macular edema; however, it is uncertain whether this conclusion applies to patients at high risk for vascular disease who were not included in these trials.

Importance  Patients with diabetic macular edema (DME) are at high risk of vascular complications, including stroke and myocardial infarction (MI). Concerns have been raised that intravitreal dosing of vascular endothelial growth factor inhibitors in DME could be associated with an increase in cardiovascular and cerebrovascular adverse events.

Objective  To evaluate the cardiovascular and cerebrovascular safety of ranibizumab, 0.5 mg and 0.3 mg, compared with sham with and without laser in DME.

Data Sources  Patient-level data from 6 randomized, double-masked, sham- and laser-controlled clinical trials.

Study Selection  Company-sponsored (Genentech or Novartis) studies in DME completed as of December 31, 2013.

Data Extraction and Synthesis  Pairwise comparisons (ranibizumab, 0.5 mg, vs sham and laser; ranibizumab, 0.3 mg, vs sham) were performed using Cox proportional hazard regression (hazard ratios, 95% CIs) and rates per 100 person-years. Data analysis was conducted from June 1 to July 15, 2015.

Main Outcomes and Measures  Standardized Medical Dictionary for Regulatory Activities queries and extended searches were prospectively defined to identify relevant safety end points, including arterial thromboembolic events, MI, stroke or transient ischemic attack, vascular deaths, and major vascular events as defined by the Antiplatelet Trialists’ Collaboration (APTC).

Results  Overall, 936 patients were treated with ranibizumab, 0.5 mg; 250 patients with ranibizumab, 0.3 mg; and 581 patients with sham/laser. The hazard ratios associated with all pairwise comparisons included 1 for all key cardiovascular and cerebrovascular safety end points. For ranibizumab, 0.5 mg, vs sham/laser and ranibizumab, 0.3 mg, vs sham, the hazard ratios were, respectively, arterial thromboembolic events, 1.05 (95% CI, 0.66-1.68) and 0.78 (95% CI, 0.43-1.40); MI, 0.84 (95% CI, 0.41-1.72) and 0.94 (95% CI, 0.43-2.06); stroke or transient ischemic attack, 0.94 (95% CI, 0.44-1.99) and 0.53 (95% CI, 0.19-1.42); stroke (excluding transient ischemic attack), 1.63 (95% CI, 0.65-4.07) and 0.59 (95% CI, 0.14-2.46); vascular death, 2.17 (95% CI, 0.57-8.29) and 2.51 (95% CI, 0.49-12.94); and APTC-defined events, 1.09 (95% CI, 0.63-1.88) and 1.00 (95% CI, 0.51-1.96).

Conclusions and Relevance  This pooled analysis includes 1 of the largest patient-level data sets on treatment of DME with ranibizumab. Although still underpowered to detect small differences for infrequent events, such as stroke, the findings suggest that intravitreous ranibizumab does not increase the risk of systemic vascular events. However, uncertainty remains for patients with DME who are at high risk for vascular disease and were not included in these trials.

Intravitreal use of vascular endothelial growth factor (VEGF) inhibitors during the past decade has established a new standard of care for patients with a range of exudative and neovascular chorioretinal diseases. Large randomized clinical trials have demonstrated the efficacy and safety of these drugs in neovascular age-related macular degeneration, retinal vein occlusion, myopic choroidal neovascularization, diabetic retinopathy, and diabetic macular edema (DME).^1,2 Ranibizumab (Lucentis; Genentech, Inc) was the first targeted biological molecule approved for intravitreal use in all ophthalmology indications, starting with neovascular age-related macular degeneration in 2006. Of all available VEGF inhibitors, ranibizumab has the most comprehensive efficacy and safety record. Data from both clinical studies and postmarketing use include clinical development programs that have enrolled more than 55 400 patients in interventional and observational studies across indications, with a cumulative estimated 4.3 million patient treatment-years of exposure and more than 25.8 million injections (L. Zuurman, B. Dandotikar; written communication, November 22, 2016).

Systemic safety monitoring of anti-VEGF therapies is important because VEGF plays a crucial role within the vascular system as a regulator of developmental and pathologic angiogenesis, vascular permeability, and inflammation.³ An increased risk of various vascular events has been observed in patients treated systemically with a full-length anti-VEGF antibody, bevacizumab, in oncology.⁴ Intravitreal doses are lower; nevertheless, all anti-VEGF agents move rapidly from the eye into the bloodstream, and some persist in the systemic circulation up to 1 month after intravitreal injection.⁵ Although ranibizumab was engineered for delivery into the eye with a fragment crystallizable region–free design, which substantially reduces systemic exposure compared with the other fragment crystallizable–containing anti-VEGF molecules,^5-8 a potential for systemic effects following this route of administration cannot be excluded.

Concern regarding systemic adverse effects is even more relevant in patients with DME,^9,10 who have an elevated risk of macrovascular and additional microvascular complications compared with patients with diabetes who did not have DME.¹¹ Results from a health care claims analysis showed rates of hospitalized cerebrovascular accident and myocardial infarction to be 2.0 and 2.5 times higher, respectively, in patients with DME than in patients with diabetes who did not have DME.¹²

Although powered to evaluate efficacy, clinical trials typically are underpowered to evaluate low-frequency safety events. Combining information across trials increases the power to identify potential treatment-associated safety events. The objective of this analysis was to provide additional insight to the retinal community on the safety of ranibizumab by pooling results across Novartis and Genentech company-sponsored clinical trials in DME.

This pooled analysis using patient-level data allows more in-depth analyses than examination of published study-level data (typical of most published meta-analyses) by incorporating the per-patient duration of exposure to treatment, adjustment for imbalances in predefined baseline risk factors, and the effect of results from single studies on the overall result. A formal statistical analysis plan was specified before the start of this project, defining study inclusion criteria, end points, potential risk factors, and analysis methods. This analysis is thus considered to provide the most comprehensive evidence that we can collate at the present time to assess the systemic risks associated with intravitreal ranibizumab at doses of 0.5 and 0.3 mg.

The studies evaluated in this pooled analysis include all randomized, double-masked phase 2 and 3 studies in DME that met the following criteria: company-sponsored (Genentech or Novartis) trials with a control arm (sham with or without laser) and at least 1 ranibizumab arm (0.3 mg or 0.5 mg with or without laser) completed by December 31, 2013. Six clinical trials met these criteria: RELATION,¹³ RESOLVE,¹⁴ RESTORE,¹⁵ REVEAL,¹⁶ RISE,¹⁷ and RIDE.¹⁷ All 6 studies were conducted in compliance with the tenets of the Declaration of Helsinki.¹⁸ Approval was obtained from the designated independent ethics committee or institutional review boards, and all patients provided written informed consent before enrollment into the trials.

The pooled safety database included all adverse events (AEs), timing of AEs relative to study initiation or first dose of ranibizumab, dosing information (drug exposure), demographic data, key potential baseline risk factors classified from medical history, and treatment arm. Strategies to identify relevant events were defined prospectively. The selected safety end points (grouping of AEs) represent the medical concepts previously identified as potential risks for the systemic use of anti-VEGF agents.^1,2,4,19 The main focus of this study is key cardiovascular and cerebrovascular end points (Figure 1), but additional prespecified end points, also considered as potentially caused by VEGF therapy, were assessed, with the results presented herein (Figure 2).

Whenever possible, standardized Medical Dictionary for Regulatory Activities (MedDRA) queries (SMQs) were used (ie, externally validated, standard sets of MedDRA terms that relate to a defined medical condition or area of interest) to create safety end points. When specific SMQs were not available, composite safety end point searches were prospectively developed, based on several SMQs or SMQs and preferred terms. Such composite safety end points were developed for arterial thromboembolic events (ATEs), stroke, stroke and/or transient ischemic attack (TIA), and wound healing complications. Although stroke and TIA are considered to be of the same etiology and legitimately combined, in patients with diabetes, TIA might be overestimated owing to confounding symptoms; therefore, analyses of both stroke and stroke and/or TIA are presented.

Vascular deaths were evaluated by reviewing all reported deaths to identify those with potential vascular and unknown causes. The Antiplatelet Trialists’ Collaboration (APTC) end point, a predefined end point consisting of death from a vascular or unknown cause, nonfatal MI, and nonfatal stroke, was also assessed.²⁰ The detailed definitions for all end points are given in eTable 1 in the Supplement. Particular MedDRA-preferred terms may be included in more than 1 end point.

An additional end point was created following publication of the 1-year results of the Diabetic Retinopathy Clinical Research Network (DRCR.net) Protocol T study,²¹ which reported an imbalance in any cardiovascular events. The any cardiovascular events end point was a nonstandardized safety end point developed by DRCR.net based on data reported in the Protocol T trial and did not use MedDRA hierarchy or SMQs. We worked with DRCR.net to develop a composite end point of MedDRA-preferred terms to apply to this pooled data set that mirrors the events included by DRCR.net in their any cardiovascular events analysis (eFigure 1 in the Supplement).

Patients were classified into 3 groups for the purpose of analysis, based on the treatment they received: (1) ranibizumab, 0.5 mg, monthly or as needed, with or without laser; (2) ranibizumab, 0.3 mg, monthly or as needed; and (3) sham or laser (cohorts subsequently referred to as 0.5 mg, 0.3 mg, and sham, respectively).

The analyses focused on 2 primary pairwise comparisons: ranibizumab, 0.5 mg, vs sham and ranibizumab, 0.3 mg, vs sham. In addition, comparisons of ranibizumab, 0.5 mg, vs ranibizumab, 0.3 mg, were performed (eFigure 2 in the Supplement). To be included in a pairwise comparison, each study was required to have both treatment arms compared. Therefore, all 6 studies were included in the ranibizumab, 0.5 mg, vs sham comparisons, but only RIDE and RISE were included in the pairwise comparisons of ranibizumab, 0.3 mg, vs sham or ranibizumab, 0.3 mg, vs ranibizumab, 0.5 mg.

Treatment arms were defined based on the initial randomized dose regardless of dosing regimen. Therefore, regimens of fixed monthly injections (RIDE and RISE), 3 loading doses followed by monthly as-needed injections (RESTORE, RESOLVE, and REVEAL), dose doubling allowed (RESOLVE), and 4 loading doses followed by monthly as-needed injections (RELATION) were pooled (Table).

Event rates are reported as percentage of patients with events for each end point and as rates per 100 person-years of exposure to accommodate different durations of exposure across studies. Only the first event of each type was recorded, and its timing was entered into the analysis. For a composite end point, such as APTC, a patient was counted only once even if multiple events (eg, stroke and MI) occurred, and the timing of the first event was included in the analysis. Safety events were included only up to the point when the protocol allowed patients receiving sham treatment to cross over to active treatment. Sham patients who received ranibizumab rescue therapy before protocol-defined crossover were censored at the point of active treatment. Methods incorporated the timing of the events and the length of the observational period to allow for the pooling of studies of differing durations.

Treatment differences for each of the pairwise comparisons were analyzed in 2 different ways for each end point: (1) separate Cox proportional hazards regression models were fit for each pairwise comparison unadjusted for baseline risk factors and (2) a global Cox proportional hazards regression model incorporating all pairwise comparisons adjusting for relevant baseline risk factors (listed in eTable 2 in the Supplement) was fit. All models included a stratification factor for study. To evaluate the appropriateness of pooling estimates across studies, an interaction term (study by treatment) was included in each Cox proportional hazards regression model. Estimated hazard ratios (HRs) and 95% CIs are provided in the figures containing forest plots, and the results of the interactions tests are provided in the legends of those figures. Cox proportional hazards regression models allowed for combining studies of different durations and incorporate the timing of each end point for each patient. Selective cumulative Kaplan-Meier plots over time categorized by treatment are also provided. Statistical analysis was conducted using SAS, version 9.2 (SAS Institute). Data analysis was conducted from June 1 to July 15, 2015.

The pooled DME data set from the 6 trials comprised 1767 patients, including 1186 who received ranibizumab and 581 who received sham treatment (Table). As is typical for many pooled analyses, studies differed by design, patient population (eg, white, Asian), region (non–US-only vs US-only settings), inclusion and exclusion criteria, treatment duration (1-2 years), ranibizumab dose, treatment regimen, and as-needed retreatment criteria (Table). Patient inclusion and exclusion criteria with respect to prior cardiovascular and cerebrovascular history varied. For RESOLVE,¹⁴ patients with a history of stroke or MI could be included. For RISE and RIDE,¹⁷ patients who had a stroke (or other cerebrovascular accident) or MI within 3 months prior to enrollment were excluded. For the remaining studies, all patients with a prior stroke were excluded. Frequencies for baseline risk factors are provided in eTable 2 in the Supplement. Generally, treatment groups were well balanced for each of the potential risk factors.

The overall rates of ATE, MI, stroke or TIA, strokes (TIA excluded), vascular deaths, and APTC events are listed in eTable 3 in the Supplement, and the related rates per 100 person-years of exposure are presented in Figure 1. The 95% CIs for HRs from the Cox proportional hazards regression models adjusted for baseline risk factors included 1 and were comparable (within end points) for all pairwise comparisons (Figure 1). The study-level forest plots, showing HRs from the constituent studies for the ATE-related end points, are displayed in eFigure 2 in the Supplement. The Cox proportional hazards regression results for the pairwise (unadjusted for baseline risk factors) model were very similar to results from the global (adjusting for baseline risk factors) model, as shown in eFigure 2 in the Supplement. Kaplan-Meier estimates over time were similar or lower in the ranibizumab arm across treatment groups compared with sham for all ATE-related end points (eFigure 3 in the Supplement). In addition, the analysis performed to mirror the DRCR.net any CV event showed no meaningful differences between the ranibizumab and sham groups, regardless of whether hypertension was included (Figure 3).

Hazard ratios from both the pairwise and global Cox proportional hazards regression models for hypertension, proteinuria, congestive heart failure, cardiac arrhythmia, renal failure, gastrointestinal hemorrhage or perforation, venous thromboembolic event, and death from all causes are presented in Figure 2. These were also similar, with 95% CIs including 1 for all pairwise comparisons.

A number of AEs were identified under the prespecified safety end point of wound healing complications. These events were reported at similar rates in the ranibizumab, 0.3-mg, and sham groups (HR, 1.00; 95% CIs, 0.14-7.08), but were higher in patients who received ranibizumab, 0.5 mg, compared with sham (HR, 8.07; 95% CI, 1.88-34.74) (Figure 2 and eFigure 4 in the Supplement). A total of 20 patients (1.85 per 100 person-years) reported such AEs in the ranibizumab, 0.5-mg, group across the 6 studies (all nonserious except 1 serious AE of wound infection) vs 2 (0.43 per 100 person-years) in the ranibizumab, 0.3-mg, group and 2 (0.27 per 100 person-years) in the sham group. Information on these AEs is limited, with 9 reported as wound and 7 reported as wound infection/inflammation. No events consistent with impaired healing of surgical wounds were identified. The rate of AEs under the MedDRA-preferred term of skin ulcer was lower in the ranibizumab, 0.5-mg, group (0.37 per 100 person-years) than in the sham group (0.81 per 100 person-years).

This pooled analysis brings together patient-level data from 6 Genentech- and Novartis-sponsored phase 2 and 3 studies in DME that included 1767 patients (1186 received ranibizumab). The rates of ATE events, including MIs, strokes or TIAs, strokes (TIA excluded), vascular deaths, and APTC events during 1 to 2 years of treatment with ranibizumab, 0.5 mg and 0.3 mg, vs sham were found to be comparably low in all treatment groups.

We also sought to mirror the analysis of the nonstandard any cardiovascular event end point performed by DRCR.net in Protocol T.²¹ The higher rate observed in Protocol T 1-year results with ranibizumab compared with aflibercept or bevacizumab was not seen in the present pooled analysis. However, Protocol T did not include a sham injection or laser control group. In a post hoc analysis of Protocol T 2-year data, no significant differences were identified among treatment groups in the proportion of participants who had at least 1 such event.

The 2-year analysis of Protocol T reported an increased rate of APTC events in patients who received ranibizumab compared with aflibercept or bevacizumab.²² This higher rate was also not observed in the present analysis, which found low APTC event rates (3.72 per 100 person-years in the ranibizumab-0.3 mg cohort and 2.96 per 100 person-years in the ranibizumab-0.5 mg cohort) in line with results from other DME trials, including DRCR.net Protocol I.²³ Protocol T included 218 ranibizumab-treated patients with DME vs 1186 patients in the present pooled analysis.

Analyses conducted for congestive heart failure, hypertension, cardiac arrhythmia, renal failure, proteinuria, gastrointestinal hemorrhage, venous thromboembolic event, and deaths from all causes also showed comparable rates for both ranibizumab doses vs sham. An imbalance in the prespecified wound healing complications end point was identified, with higher rates reported in the ranibizumab, 0.5-mg, group vs the sham group. The reporting of these wound healing AEs was low considering the DME patient population, and ophthalmologists were not requested to prospectively assess peripheral microvascular disease in these studies. The rate of skin ulcers, an AE that may be expected if there was compromised wound healing, was lower with ranibizumab, 0.5 mg, than with sham. Thus, the relevance of this observed imbalance remains unclear, and monitoring through routine pharmacovigilance continues.

A major advantage of the present pooled analysis compared with most previous meta-analyses is that it incorporates patient-level data for each included trial, thereby allowing a greater depth of analysis.²⁴ This advantage included time to event analyses allowing adjustment for study differences and potential differences in demographics and baseline risk factors. Adjusting for these baseline risk factors yielded results similar to those obtained without adjustment. Overall, our finding is in line with that of previous DME meta-analyses conducted on study-level published results.^25,26 Despite overlaps in the studies included, these meta-analyses have used different event classification definitions and/or analysis methods.

Studies with available patient-level data were selected, which permitted a prospectively defined systematic approach that allowed us to combine heterogeneous studies using the same standardized, prespecified end points and methods of analysis across all studies. Adjustments for multiple doses and events were not made to allow for a nominal 95% CI–based interpretation for each comparison, which could have increased the probability of chance findings. This pooled analysis was performed on one of the largest ranibizumab clinical trial safety databases of anti-VEGF treatment in DME of which we are aware.

Most patients with DME who have a history of stroke or MI were excluded from these trials (exclusion criteria varied by trial). This exclusion limits extrapolation of the present findings to the wider patient population, especially to patients at high risk of vascular events. In addition, as with many meta-analyses based on published study results, this pooled analysis of patient-level data brings together information from heterogeneous clinical studies. In particular, the analysis combines data from trials with monthly treatment, accounting for 42% of the study population, and with as-needed regimens, accounting for 58% of the study population.

Previous meta-analyses of studies with monthly dosing have suggested the potential occurrence of vascular effects of monthly anti-VEGF treatment in DME.^26,27 None of the included studies compared monthly and as-needed regimens, and we did not perform cross-study comparisons of safety end point rates between these 2 regimens owing to difference in study methodologies. This analysis, comprising a majority of patients who followed a less-than-monthly regimen, reflects the posology that most physicians use in clinical practice.

Even pooled analyses and meta-analyses may lack the power to assess safety, depending on the event rates and the magnitude of the treatment effect.²⁸ A prospective study with a sample size of 1767 (as we have for this pooled analysis) may be too small to appropriately evaluate the rates of infrequent events, such as stroke. For example, with a rate of 1% in the sham group, such a sample size confers a 25% power (using a 2-sided 5% significance level) to detect a doubling of the stroke risk in the ranibizumab, 0.5-mg, group. When combined into the APTC end point, events of similar causes (MI, stroke, and vascular death) result in higher event rates (sham rates of 3.5%), providing adequate power to detect a doubling of the risk in a prospective study. However, as observed here and noted in previous studies,²⁸ HRs below 1 (MI) and above 1 (stroke, vascular death) are combined, which can make the results more difficult to interpret.

The results of this pooled analysis show that, for patients with DME similar to those included in these trials and treated predominantly with an as-needed regimen, the rates of vascular events remain low and are outweighed by the potential benefits. To establish the safety of anti-VEGF therapies in patients with DME who have a history of stroke or MI, additional large, active controlled trials and/or real-world evidence including many more such patients are needed. Treatment choice for patients at high risk should thus be based on careful evaluation of the potential benefits and risks, taking into account the safety and efficacy of alternative treatments, including laser and corticosteroids.

Accepted for Publication: February 19, 2017.

Corresponding Author: Marco A. Zarbin, MD, PhD, Institute of Ophthalmology and Visual Science, New Jersey Medical School, Rutgers University, 90 Bergen St, Room 6156, Newark, NJ 07103 (zarbin@earthlink.net).

Published Online: April 6, 2017. doi:10.1001/jamaophthalmol.2017.0455

Author Contributions: Dr Francom 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: Zarbin, Dunger-Baldauf, Haskova, Snow, Francom.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Zarbin, Haskova, Koovejee, Mousseau, Snow, Francom.

Critical revision of the manuscript for important intellectual content: Zarbin, Dunger-Baldauf, Haskova, Koovejee, Margaron, Snow, Beaumont, Staurenghi, Francom.

Statistical analysis: Zarbin, Dunger-Baldauf, Snow, Francom.

Administrative, technical, or material support: Zarbin, Mousseau, Snow.

Study supervision: Zarbin, Haskova, Koovejee, Snow, Francom.

Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Dr Zarbin is a consultant for Calhoun Vision, Coherus BioSciences, Frequency Therapeutics, Genentech/Roche, Healios KK, Isarna Therapeutics, Makindus, Novartis, Ophthotech, and Percept Corp. Dr Beaumont is on the advisory boards of Novartis Allergan and for education funding by Bayer. Dr Staurenghi is a consultant for Bayer, Novartis, Genentech/Roche, Boehringer, Allergan, and Alcon. Drs Dunger-Baldauf, Mousseau, Margaron, and Snow are employees of Novartis Pharma. Drs Haskova, Koovejee, and Francom are employees of Genentech Inc. No other disclosures were reported.

Funding/Support: This study was funded by Novartis Pharma AG, Basel, Switzerland, and Genentech, Inc, South San Francisco, California.

Role of the Funder/Sponsor: Genentech and Novartis were involved in the design of the analysis plan; the interpretation of the data; the preparation, review, and approval of the manuscript, as well as the decision to submit the manuscript for publication; and the decision on the journal to which the manuscript was submitted.

Previous Presentations: The study was presented orally at the annual meeting of American Academy of Ophthalmology, November 14-17, 2015; Las Vegas, Nevada; the annual meeting of American Society of Retina Specialists; July 14, 2015; Vienna, Austria; and the MaculArt meeting; June 29, 2015; Paris, France.

Additional Contributions: Assistance in the analysis and/or interpretation of the data was given by Chad Melson, MS (Experis); Christine Thorburn, PhD, Vladimir Bezlyak, PhD, Eli Zangvil, MD, and Soumil Parikh, MD (Novartis); Natasha Singh, PharmD, Ronald Cantrell, PhD, Lisa Tuomi, PharmD, Flavia Di Nucci, MD, Michael Rea, MS, and Susanna Grzeschik, PhD (all with Genentech at the time of the study). We thank the advisory committee who approved the original design of this pooled analysis and gave input into the initial interpretation, including Jean-Francois Korobelnik, MD (Service d'Ophtalmologie, Centre Hospitalier Universitaire de Bordeaux, France) and Ingram Olkin, PhD (Department of Statistics, Stanford University). Members of the advisory committee received compensation for their contribution.