The Comparative Risk of Endophthalmitis Following Intravitreal Injection with Bevacizumab, Aflibercept, and Ranibizumab

Abstract

Purpose:

To determine if sterile pre-loading of anti-VEGF agents reduces the risk of post-intravitreal injection endophthalmitis.

Methods:

This is a retrospective cohort study using medical claims data from a large, national U.S. insurer. Cohorts were created using intravitreal injections of anti-VEGF injections from 2005–2016. For inclusion, patients had to have at least 6 months of data prior to the injection and were excluded for any previous diagnosis of endophthalmitis, multiple injected drugs on the day of injection, or intraocular surgery within 15 days of the injection or between an injection and a diagnosis of endophthalmitis. The primary outcome was the odds of endophthalmitis after an intravitreal injection.

Results:

706,725 bevacizumab, 210,849 ranibizumab, and 177,731 aflibercept injections were given to 130,327 patients. Multivariate analysis showed that ranibizumab and aflibercept together had an increased odds of endophthalmitis (OR=1.29,95%CI:1.04–1.59,p=0.02) compared to bevacizumab. Individually, ranibizumab(OR=1.25,95%CI:0.97–1.61,p=0.08) and aflibercept(OR=1.34,95%CI:0.99–1.81,p=0.06) each had higher odds of endophthalmitis, but neither result met significance. Also, when compared to males, females had a higher odds of getting endophthalmitis (OR:1.30,95%CI:1.05–1.61,p=0.02).

Conclusions:

The odds of endophthalmitis with aflibercept and ranibizumab combined were higher compared to the sterilely pre-loaded bevacizumab, arguing for a safety advantage of sterile pre-loading of anti-VEGF injections.

Summary Statement:

Sterile pre-packaging of anti-VEGF agents into syringes reduces the risk of post-injection endophthalmitis. Women may also be at higher risk for post-injection endophthalmitis compared to men.

Introduction

Intravitreal injections of anti-vascular endothelial growth factor (anti-VEGF) agents are an effective and common treatment for numerous retinal conditions.^1–4 Although typically well-tolerated, post-injection endophthalmitis is one of the most serious complications that can occur and is reported to have an incidence rate of 0.02–0.05%.^5–7 Earlier this decade, bevacizumab-associated outbreaks of clusters of endophthalmitis made national news.^8–11 Studies of these outbreaks concluded that the most likely cause was contamination during the compounding process.^12,13 In response, The United States Food and Drug Administration (FDA) released draft guidance in 2015 that would have limited the use of bevacizumab to within 5 days of compounding.¹⁴ This strict day count would have made proper sterility testing of bevacizumab impossible, and a large concern arose that it would have essentially ended the use of bevacizumab for intravitreal injections.^15,16 Fortunately, the American Academy of Ophthalmology and the American Society of Retina Specialists, among others, appear to have worked as the FDA did not implement the draft guidance.

In addition to the lobbying efforts, the FDA was possibly influenced to not implement the restrictive policy by contemporaneous reports demonstrating similar rates of endophthalmitis between bevacizumab and ranibizumab.^7,17 Despite the release of these and other recent reports showing bevacizumab was not associated with an increased risk of endophthalmitis^18,19, apprehension over the safety of compounding persists.²⁰ Further adding to these concerns are recent reports of infections from use of counterfeit bevacizumab in China and India^21,22 and increased rates of sterile inflammation.^20,23,24

However, overlooked in this discussion is the real possibility that the loading of syringes in proper sterile conditions as performed during the repackaging of bevacizumab may lower the incidence of endophthalmitis. As used across the United States, most bevacizumab injections are aliquoted out from a much larger single chemotherapeutic dose under sterile conditions typically in a negative pressure hood at a compounding pharmacy. Consistent with this, several large studies have shown rates of endophthalmitis to be lower in bevacizumab treated eyes, but none have had enough statistical power to detect a significant difference.^7,17,25

Previously, we studied intravitreal injections from 2005 to 2012 and found that among 296,565 bevacizumab and 87,245 ranibizumab injections, the odds of endophthalmitis for ranibizumab was 1.51 (p=0.11) compared to bevacizumab.⁷ Aflibercept was not studied at that time as the database only had 44 injections in it through the end of 2012. Since that publication, 4 additional years of data have been added to the medical claims database used in that study. Given the availability of these new data and the expanded use of aflibercept since 2012, we wanted to reassess this comparison. The goal of this study is to evaluate the risk of endophthalmitis for intravitreal injections compared between each of the anti-VEGF agents.

Methods

Dataset

The methodology for this study closely parallels that used for our previous study.⁷ A retrospective cohort study was performed using the Clinformatics™ Data Mart Database (OptumInsight, Eden Prairie, MN). This de-identified administrative medical claims database contains the billing claims of all beneficiaries from a large national insurance company covering patients in each of the 50 United States. All outpatient medical claims (office visits, procedures and medications given) and demographic data for each beneficiary during their enrollment are included. The subset of data available for this study included all patients in the database from January 1, 2005 to December 31, 2016. This study has been deemed exempt from review by the University of Pennsylvania’s Institutional Review Board due to the de-identified nature of the database.

Study cohorts

All intravitreal injections (CPT code 67028) using either bevacizumab, ranibizumab or aflibercept were identified and divided into cohorts based on the agent injected. Each intravitreal injection was considered a unique observation and the date of injection was counted as the index date. Each injection was included in the study if the patient had at least 6 consecutive months in the insurance plan before and 1 month following the injection. Any injection coded with a bilateral code (CPT 50) or was billed for both a right and left injection on the same day was considered to be 2 injections. (SeeSupplemental Table 1 for complete list of codes used during this study.)

At any point that an individual was given a diagnosis code for endophthalmitis, all future injections were removed from the analysis. Additionally, in an effort to isolate only cases of endophthalmitis related strictly to intravitreal injections, all index dates that occurred less than 15 days after an intraocular surgery or had an intraocular surgery occur between the index date and a diagnosis of endophthalmitis were also excluded. Any intravitreal injections involving more than one drug or if the patient had an injection of steroid within 15 days of the injection (before or after) were excluded as well.

Outcome measures

The primary outcome measure for this study was the odds of developing endophthalmitis after an intravitreal injection for each of the anti-VEGF agents compared against each other. Furthermore, under the hypothesis that a pre-filled syringe would have a lower risk than those filled in office, bevacizumab was compared to a combined ranibizumab and aflibercept cohort. Cases of endophthalmitis were defined by having a new endophthalmitis diagnosis code in conjunction with a code for a “tap and inject” procedure, a vitrectomy or an intravitreal injection. This outcome definition is a slight modification from the previous study, which also required a code for intravitreal antibiotics on the same date. The requirement for antibiotics was dropped after a sensitivity analysis in our previous study revealed that using the above definition much more closely tracked with endophthalmitis rates seen in clinical trials and other chart-based reviews. Cases were required to occur from 1–14 days after index date. Those that occurred on the same day as the index date were excluded due to an inability to distinguish if the intravitreal injection preceded the endophthalmitis defining procedure.

To account for possible nonindependence and clustering of observations from a single patient, general estimating equations were used in all regression analyses to generate odds ratios. Covariates of interest were collected at the time of the index date and included basic demographic information; age, race and gender. Additionally, the injection–associated diagnosis was categorized as either age-related macular degeneration (AMD), diabetic macular edema (DME), retinal vein occlusions with cystoid macular edema (RVO/CME) or other. To account for any variation in the rates of endophthalmitis over the 11-year length of the study, the year the injection was performed was also included as a variable. Statistical analysis was performed using SAS (version 9.4; SAS Institute Inc., Cary, NC).

Results

Our cohort consisted of 130,327 patients who had 1,095,305 intravitreal injections during the study period. Of these, bevacizumab constituted the majority of injections at 706,725, followed by ranibizumab at 210,849 and aflibercept at 177,731.(Figure 1) On average aflibercept injections were given to older patients who were more likely to be white and less likely to be given for AMD than the bevacizumab or ranibizumab injections (p<0.01 for all comparisons) (See Table 1 for baseline characteristics).

Figure 1:

Numbers of injections remaining after exclusion criteria are applied

Table 1:

Baseline Characteristics

	Ranibizumab	Bevacizumab	Aflibercept	Totals
Number of injections	210,849	706,725	177,731	1,095,305
Number of patients*	26,150	107,171	24,133	130,327
Number of Endophthalmitis	85	230	65	380
Characteristics**				p-value
Age (mean (SD))	76.7 (9.7)	75.9 (10.7)	78.0 (9.5)	<0.011
Gender (F)	126768 (60%)	417575 (59%)	106384 (60%)	<0.012
Race				<0.012
White	165069 (78%)	542239 (77%)	135989 (77%)
Black	13343 (6%)	47460 (7%)	10236 (6%)
Hispanic	9817 (5%)	39270 (6%)	7935 (4%)
Asian	3078 (1%)	12990 (2%)	3795 (2%)
Unknown	19542 (9%)	64766 (9%)	19776 (11%)
Diagnosis				<0.012
Diabetic Retinopathy	25293 (12%)	86933 (12%)	17657 (10%)
AMD/CNVM	123041 (58%)	384589 (54%)	72622 (41%)
Vein occlusion/CME	20391 (10%)	74024 (10%)	14694 (8%)
Multiple	10466 (5%)	37385 (5%)	6419 (4%)
Other/unknown***	31658 (15%)	123794 (18%)	66339 (37%)

¹ANOVA F-Test

²Chi-Square

^*Patients may be included in more than one cohort as patients could have had multiple types of anti-VEGF medications

^**statistics based on number of injections

^***Patients who did not have codes for any of the other listed diagnoses

Over the course of the study, 380 incident cases of endophthalmitis were found; 85 after ranibizumab, 230 after bevacizumab and 65 after aflibercept injections. The overall rate of endophthalmitis per anti-VEGF injection was 0.035% (or 1 infection every 2857 injections). The individual rates of infection by drug were 0.040% for ranibizumab (1 every 2526 injections), 0.033% for bevacizumab (1 every 3030 injections), and 0.037% for aflibercept (1 every 2703 injections). The highest and lowest yearly rates of endophthalmitis were 2006 (0.105% or 1 every 956 injections) and 2016 (0.026% or 1 every 3886 injections), respectively. (Table 2)

Table 2:

Breakdown of injection and outcome by drug

	Ranibizumab			Bevacizumab			Aflibercept			Year Total
Cohort Year	# Injection	# Endoph	Rate (%)	# Injection	# Endoph	Rate (%)	# Injection	# Endoph	Rate (%)	# Injection	# Endoph	Rate (%)
2005	0	0	0.000	2,251	1	0.044	0	0	0.000	2251	1	0.044
2006	0	0	0.000	7,649	8	0.105	0	0	0.000	7,649	8	0.105
2007	203	0	0.000	18,127	9	0.050	0	0	0.000	18,330	9	0.049
2008	7,831	4	0.051	19,557	11	0.056	0	0	0.000	27,388	15	0.055
2009	10,968	5	0.046	31,849	12	0.038	0	0	0.000	42,817	17	0.040
2010	15,894	8	0.050	49,038	18	0.037	0	0	0.000	64,932	26	0.040
2011	21,743	9	0.041	72,002	22	0.031	0	0	0.000	93,745	31	0.033
2012	24,306	12	0.049	92,715	32	0.035	44	0	0.000	117,065	44	0.038
2013	34,386	14	0.041	93,827	24	0.026	23,177	14	0.060	151,390	52	0.034
2014	33,402	9	0.027	94,399	34	0.036	30,324	10	0.033	158,125	53	0.034
2015	31,524	18	0.057	108,495	31	0.029	53,982	19	0.035	194,001	68	0.035
2016	30,592	6	0.020	116,816	28	0.024	70,204	22	0.031	217,612	56	0.026
Totals	210,849	85	0.040	706,725	230	0.033	177,731	65	0.037	1,095,305	380	0.035

In the univariate analysis, there was no difference seen between medications in any comparison (bevacizumab vs. combined ranibizumab/aflibercept - OR: 1.19, 95% CI: 0.97–1.46, p=0.10; ranibizumab vs. bevacizumab - OR: 1.24, 95% CI: 0.97–1.59, p=0.09; aflibercept vs. bevacizumab - OR: 1.12, 95% CI: 0.85 −1.48, p=0.41; aflibercept vs. ranibizumab OR: 0.91, 95% CI: 0.66–1.25, p=0.55). Each progressive year showed a significantly decreased odds of endophthalmitis (OR: 0.93, 95% CI: 0.90–0.97, p<0.001), and females had significantly increased odds of endophthalmitis compared to men (OR: 1.31, 95% CI:1.06–1.62, p=0.01). Unknown race was associated with significantly lower odds compared to whites (OR: 0.58, 95% CI: 0.38–0.90, p=0.01). No other race/ethnicities were significantly associated with endophthalmitis (p>0.08 for all comparisons). Age (1 year increments) and disease indication were also not associated with endophthalmitis (p>0.37 for all comparisons). (See Table 3 for unadjusted results)

Table 3:

Unadjusted and Adjusted ORs for Developing Post-injection Endophthalmitis^*

	Unadjusted OR (95% CI)	p-value	Adjusted OR (95% CI)	p-value
Age**	1.00 (0.99 – 1.01)	0.88	1.00 (0.99 – 1.01)	0.90
Year of injection**	0.93 (0.90 – 0.97)	<0.001	0.92 (0.90 – 0.98)	0.002
Gender
Male	Ref	Ref	Ref	Ref
Female	1.31 (1.06 – 1.62)	0.01	1.30 (1.05 – 1.61)	0.02
Race
White	Ref	Ref	Ref	Ref
Black	1.36 (0.96 – 1.93)	0.08	1.39 (0.97 – 1.98)	0.07
Hispanic	0.63 (0.36 – 1.10)	0.10	0.67 (0.38 – 1.16)	0.15
Asian	0.70 (0.29 – 1.68)	0.42	0.75 (0.31 – 1.81)	0.53
Unknown	0.58 (0.38 – 0.90)	0.01	0.61 (0.40 – 0.94)	0.03
Diagnosis
Diabetic Retinopathy	Ref	Ref	Ref	Ref
AMD/CNVM	1.16 (0.84 – 1.61)	0.37	0.98 (0.69 – 1.41)	0.92
Vein occlusion	0.80 (0.50 – 1.29)	0.36	0.78 (0.48 – 1.26)	0.31
Multiple	0.84 (0.46 – 1.50)	0.55	0.74 (0.41 – 1.34)	0.32
Other/unknown	0.95 (0.65 – 1.39)	0.81	1.03 (0.69 – 1.55)	0.89
Bevacizumab (Ref) vs.
Ranib & Aflib	1.19 (0.97 – 1.46)	0.10	1.29 (1.04 – 1.59)	0.02
Bevacizumab (Ref) vs.
Ranibizumab	1.24 (0.97 – 1.59)	0.09	1.25 (0.97 – 1.61)	0.08
Aflibercept	1.12 (0.85 – 1.48)	0.41	1.34 (0.99 – 1.81)	0.06
Ranibizumab (Ref) vs.
Aflibercept	0.91 (0.66 – 1.25)	0.55	1.07 (0.76 – 1.51)	0.69

^*All variables listed in the table were included in the adjusted analysis.

^**Variables was treated as continuous, but evaluated at 1 year increments

After controlling for potential confounders, the multi-variate analysis showed that when combined, ranibizumab and aflibercept had higher odds of endophthalmitis (OR: 1.29, 95% CI: 1.04–1.59, p=0.02) compared to bevacizumab. Individually, aflibercept was marginally associated with increased odds (OR: 1.34, 95% CI: 0.99–1.81, p=0.06) of endophthalmitis compared to bevacizumab. The odds of endophthalmitis was also higher with ranibizumab compared to bevacizumab, but this comparison did not meet significance (OR: 1.25, 95% CI: 0.97–1.61, p=0.08). There was no difference in the odds between aflibercept and ranibizumab (OR:1.07, 95% CI: 0.76–1.51, p=0.69). Again, the year the injection was performed was associated with endophthalmitis, as the odds decreased more recently an injection was given with each successive year being more protective (OR:0.92, 95% CI: 0.90–0.98, p=0.002). Similarly, when compared to males, females had higher odds of getting endophthalmitis (OR: 1.30, 95% CI: 1.05–1.61, p=0.02). Within the categories of Race/Ethnicity, only those patients that had unknown race had a significant association, and they were less likely to get endophthalmitis (OR:0.61, 95%CI: 0.40–0.94, p=0.03). Disease indication for having an injection was not associated with endophthalmitis (p>0.31 for all categories). (See Table 3 for complete adjusted results)

Discussion:

This study evaluated 1,095,305 intravitreal injections and found that when grouped together office filled medications (ranibizumab and aflibercept) had higher odds of post-injection endophthalmitis than bevacizumab (OR=1.29, p=0.02). Individually, aflibercept was marginally associated with a higher odds of post-injection endophthalmitis compared to bevacizumab (OR:1.34, p=0.06). Ranibizumab may have a higher odds ratio for endophthalmitis compared to bevacizumab (OR:1.25), but the statistical support for this possibility was not strong (p=0.08). When initially undertaking the updated version of this study, we hypothesized that an additional 4 years of anti-VEGF injections would provide a large enough sample size to detect a modest, but real difference in a rare and potentially disastrous outcome. While our study did accrue nearly 1.1 million injections, our power to detect individual differences between the medications was still limited by “only” having 210,849 ranibizumab injections and 177,731 aflibercept injections. Our current findings are consistent with our earlier study in which ranibizumab was associated with a higher odds ratio (OR=1.51, 95% CI:0.91–2.56) compared to bevacizumab, but did not meet statistical significance (p=0.11).⁷ However, our results do corroborate the previous studies demonstrating no increased risk for endophthalmitis due to bevacizumab injections.^7,17–19

To be clear, we are not suggesting the higher/lower odds of endophthalmitis should be considered a property of the anti-VEGF agents themselves. Instead, we believe bevacizumab offers a lower risk for endophthalmitis strictly due to the standard protocol most compounding pharmacies use to aliquot the medication under sterile conditions into pre-loaded syringes, compared to the in-office loading for the other two agents. Numerous studies looking to take advantage of the cost savings associated with repackaging have demonstrated the stability of each of the anti-VEGF agents in plastic syringes.^26–29 Furthermore, pre-packaged syringes have demonstrated a 40% reduction in office preparation time, offering a clinical efficiency argument as well.^30,31 Although the odds ratio of 1.29 for endophthalmitis is a modest difference, if the disparity in absolute risk difference seen in our study is the true difference in risk, this would suggest that if every injection was sterilely loaded, one case of endophthalmitis could be prevented for every 8,847 injections performed. Considering our study found 1 case occurs every 2857 injections, this means 1 out of every 3 cases could potentially be avoided. If our results can be corroborated, the potential 33% reduction in endophthalmitis, the proven molecular stability in plastic syringes and reduced preparation time, are consistent with a premise that each anti-VEGF agent should be sterilely pre-filled into syringes. In possible recognition of these benefits, Genentech recently started offering both the 0.5mg and 0.3mg doses of ranibizumab in factory filled syringes.

Due to a lack of culture results within claims data, we are unable to determine which cases were culture positive and which were negative. Because of this, misclassification bias is a potential concern, especially in light of the difficulty in clinically distinguishing between noninfectious vitritis and true endophthalmitis. Historically, bevacizumab has been associated with higher rates of noninfectious vitritis^23,24, which would argue the lower odds of endophthalmitis for compounded bevacizumab is actually being under-represented. Complicating this picture, however, are reports of high noninfectious vitritis rates with aflibercept.^32,33 Regeneron released a statement in late February 2018 that recent clusters of noninfectious vitritis were associated with specific batches of syringes shipped with the medication.³⁴ Another report also noted an exceptionally high rate of noninfectious vitritis in aflibercept, but 85% (17/20) of those cases were associated with a single provider.³² A third report showed a significant difference in noninfectious vitritis between anti-VEGF agents, but this result was due to the starkly higher rate in aflibercept compared to ranbizumab³³, not between bevacizumab and aflibercept.

Noninfectious vitritis is reported to occur at roughly the same frequency as endophthalmitis.^20,33 To help counter the possibility of misclassification, our study mandated the use of a therapeutic procedure in addition to a new diagnosis (and not just a diagnosis) for a case of endophthalmitis to be considered to have occurred. We believe using this case definition reduces the likelihood of misclassification bias impacting our results. This is in part due to the frequency that noninfectious vitritis is clinically monitored without a therapeutic procedure, and also because of the similarities between our rate of endophthalmitis and the rates reported from other clinical data sources.

This study also found that even after controlling for type of medication, the odds of endophthalmitis after intravitreal injections has decreased since 2005 with the lowest rate of endophthalmitis occurring in 2016. The exact cause for the decrease in risk is unknown. However, an expert panel released an updated best practice for intravitreal injections in 2014.³⁵ This panel recommended the discontinuation of pre- and post-injection antibiotics, stressed the importance of limiting aerosolized oral bacterial contamination and emphasized the need for povidone iodine.³⁵ All of these best practices have evolved over time, and it is possible that the reduced rate seen over the course of this study is the culmination of the increased utilization of these practices across the country. Our results further strengthen the conclusion of others³⁶, that the overall rate of post-injection endophthalmitis is decreasing.

The other important finding among our results was that women were associated with a higher OR of endophthalmitis compared to men (OR=1.30, p=0.02). Surprisingly, few other studies have evaluated the effect of gender on the risk of endophthalmitis with most large-scale studies focusing on injection procedure related differences (speculum use, antibiotics, face mask use, etc.) A previous study using this same database did not find an association but was performed with a much smaller sample size. Reasons for this gender risk are not readily apparent, but may be related to the higher incidence of dry eye in older women compared to men, and the anti-infective properties a normal, functioning tear film is known to confer.³⁷

Although the medical claims data used for this analysis provides several strengths as noted above, the limitations of this study also need to be considered. First, due to the de-identified nature of administrative medical claims data, we are unable review a specific patient’s chart to verify clinical data. Also due to the de-identified nature of the data, we are unable to identify which compounding pharmacies were used to supply the bevacizumab in this study, and therefore, are unable to verify the “safety” of using one pharmacy over another. Next, the data used in this study comes from a single medical claims database and may not represent the results if another medical claims database (e.g. Medicare) was used instead. Also, in October of 2016 pre-filled syringes of 0.5mg ranibizumab were approved for sale by the FDA. Given the lateness of appearance within the study period, it is unlikely that a significant portion of the pre-filled version entered our dataset. However, if it had, this would mean sterilely loaded injections were counted as office loaded. This in turn would lead to a bias to the null and that the true odds of endophthalmitis with office-loaded syringes is higher than reported here.

Next, due to the nature of claims data we are unable to account for presumed risk factors for endophthalmitis like use of povidone-iodine, topical antibiotics or injection setting (operative theater vs. office). While differences in all of these factors may have impacted with results over the course of the study, for them to have affected the main outcome, they would have had to occurred differentially between agents, which seems unlikely. Lastly, no specific billing code exists for single use vials of bevacizumab, preventing the assessment of the use of these vials versus those injections derived from a compounding pharmacy. While some single use vials may have been included in our bevacizumab injections group, given the increased cost associated with this use, the likelihood of single bevacizumab vials constituting more than a minimal portion of bevacizumab injections in our data is very low.

In summary, these results offer evidence that the use of pre-loaded intravitreal anti-VEGF agents may decrease the incidence of endophthalmitis. This study also demonstrates that the overall rate of post-injection endophthalmitis has decreased over the years and that women may have higher odds of endophthalmitis.