Similar risk of kidney failure among patients with blinding diseases who receive ranibizumab, aflibercept, and bevacizumab: an OHDSI Network Study

Abstract

Objective or Purpose:

A) To characterize the incidence of kidney failure associated with intravitreal anti-vascular endothelial growth factor (VEGF) exposure, and B) compare the risk of kidney failure in patients treated with ranibizumab, aflibercept, or bevacizumab.

Design:

Retrospective cohort study across 12 databases in the Observational Health Data Sciences and Informatics (OHDSI) network.

Subjects, Participants, and/or Controls:

Subjects aged ≥18 years with ≥3 monthly intravitreal anti-VEGF medications for a blinding disease (diabetic retinopathy, diabetic macular edema, exudative age-related macular degeneration, or retinal vein occlusion).

Methods, Intervention, or Testing:

A) The standardized incidence proportions and rates of kidney failure while on treatment with anti-VEGF were calculated. B) For each comparison (e.g., aflibercept versus ranibizumab), patients from each group were matched 1:1 using propensity scores. Cox proportional hazards models were used to estimate the risk of kidney failure while on treatment. A random-effects meta-analysis was performed to combine each database’s hazard ratio (HR) estimate into a single network-wide estimate.

Main Outcome Measures:

Incidence of kidney failure while on anti-VEGF treatment, and time from cohort entry to kidney failure.

Results:

Of the 6.1 million patients with blinding diseases, 37,189 who received ranibizumab, 39,447 aflibercept, and 163,611 bevacizumab were included; the total treatment exposure time was 161,724 person-years. The average standardized incidence proportion of kidney failure was 678 per 100,000 persons (range 0 to 2389), and incidence rate 743 per 100,000 person-years (0 to 2661). The meta-analysis HR of kidney failure comparing aflibercept to ranibizumab was 1.01 (95% confidence interval (CI) 0.70, 1.47, p=0.45), ranibizumab to bevacizumab 0.95 (95% CI 0.68, 1.32, p=0.62), and aflibercept to bevacizumab 0.95 (95% CI 0.65, 1.39, p=0.60).

Conclusions:

There was no substantially different relative risk for kidney failure between those who received ranibizumab, bevacizumab, or aflibercept. Practicing ophthalmologists and nephrologists should be aware of the risk for kidney failure among patients receiving intravitreal anti-VEGF medications and that there is little empirical evidence to preferentially choose among the specific intravitreal anti-VEGF agents.

Introduction

Intravitreal anti-vascular endothelial growth factor (anti-VEGF) medications have revolutionized the treatment of blinding diseases.^1,2 However, anti-VEGF agents are known to have adverse kidney effects. Systemic administration of anti-VEGF leads to well-documented proteinuria and acute kidney injury, which are major risk factors for kidney failure.^3–8 Intravitreal administration of anti-VEGF uses 1/150 to 1/400 the medication dose of systemic administration.⁹ Despite the lower dose, systemic absorption has been demonstrated after intravitreal administration that differs by medication.^10,11 Comparative studies show that ranibizumab has the shortest systemic half-life of elimination and leads to the least elevation of serum drug levels and suppression of systemic VEGF, as compared to aflibercept and bevacizumab.^10,11

There is conflicting evidence as to whether intravitreal administration of anti-VEGF medications, like systemic administration, are also toxic to the kidneys. Kidney biopsies of patients who have received intravitreal anti-VEGF have demonstrated thrombotic microangiopathy, a pathognomonic finding of systemic VEGF blockade, and associated collapsing focal segmental glomerulosclerosis.¹² Multiple studies have shown worsening hypertension, proteinuria, and kidney disease after intravitreal anti-VEGF.^12,13 However, other retrospective studies and systematic reviews have not shown declines in kidney function or increased risk for kidney disease after intravitreal anti-VEGF medication exposure.^14–18 The concern for kidney toxicity from intravitreal anti-VEGF has led some groups to advocate for the use of ranibizumab, over aflibercept and bevacizumab, in those who are at risk for or have kidney disease.^3,19

There are only a handful of studies directly comparing the kidney effects of ranibizumab with aflibercept and bevacizumab. The prospective Diabetic Retinopathy Clinical Research (DRCR) Network Protocol T clinical trial identified 26 renal events in the ranibizumab group, 26 in the aflibercept group, and 16 in the bevacizumab group among a total of 218, 224, and 218 patients, respectively, treated for diabetic macular edema; the differences were not statistically significant.²⁰ The pre-planned post hoc analysis from the same study did not identify differences in changes in blood pressure or urine albumin-creatinine ratio in patients treated with any of the three anti-VEGF drugs.²¹ However, patients enrolled in clinical trials are often younger with well-controlled diabetes and may not be generalizable to patients treated with intravitreal anti-VEGF in routine clinical practice. Large-scale studies using routine clinical data comparing the kidney effects of anti-VEGF medications are rare. One study using the FDA’s Adverse Event Reporting System (FAERS) found no differences in kidney events between drugs, but was unable to account for baseline demographic and medical characteristics due to constraints of the database.⁹

There are major limitations in the adverse kidney effects examined in the existing studies comparing intravitreal anti-VEGF medications. Due to the relatively small sample sizes of the populations studied, many studies examined kidney effects as a group and did not distinguish between severity of kidney events. Specifically, studies have grouped severe conditions like kidney failure, a leading cause of morbidity and mortality, with transient issues such as acute kidney injury.^9,13,20 There is a need to understand whether there are differences in the severe phenotype of kidney failure between the three commonly used intravitreal anti-VEGF medications.

The Observational Health Data Sciences and Informatics (OHDSI, pronounced “Odyssey”) network is well poised to address this gap in knowledge. The OHDSI network is an international open-science collaborative centered around the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM).²² By leveraging the vast data network and standardized analytic pipeline of the OHDSI community, we can assess the risk of kidney failure with intravitreal anti-VEGF exposure across an international population. The purposes of this study were to: A) characterize the incidence of kidney failure associated with intravitreal anti-VEGF exposure, and B) compare the risk of kidney failure in patients treated with ranibizumab, aflibercept, and bevacizumab in the OHDSI network.

Methods

Study Design

This was a retrospective cohort study across 12 databases (6 administrative claims and 6 electronic health records (EHR)) standardized to OHDSI’s OMOP CDM.²³ In this CDM, local clinical concepts (e.g., International Classification of Diseases diagnosis codes, or Current Procedural Terminology codes) are mapped to OMOP concepts through an extract-transform-load (ETL) process. The OMOP CDM normalizes the structure and content of source data which allows disparate healthcare databases to be queried in a standardized manner. Database details are included in the appendix. (Table S1) All data partners had local institutional review board approval or exemption for their participation. The study adhered to the tenets of the Declaration of Helsinki and complied with Health Insurance Portability and Accountability Act (HIPAA).

Subjects / Exposure

Adult patients aged ≥18 years who were newly treated with monthly intravitreal anti-VEGF medications (ranibizumab, aflibercept, or bevacizumab) for at least 3 months for a blinding disease with at least 365 days of prior observation were included in the study. Blinding diseases included diagnoses for diabetic retinopathy, diabetic macular edema, exudative age related macular degeneration, and retinal vein occlusion. The minimum number of anti-VEGF exposures was chosen based on prior studies demonstrating suppression of systemic plasma VEGF levels after three monthly intravitreal injections.^10,11 Patients with cancer diagnoses (e.g., metastatic colorectal cancer) that could warrant systemic anti-VEGF were excluded. Patients with pre-existing kidney failure (defined further below) were also excluded. If patients switched anti-VEGF medications, only the first exposure was included in the analysis, after which the patients were right censored from the cohort. Cohorts were created using ATLAS, a publicly available web-based software application developed by the OHDSI community.²⁴ Details of the cohorts and the codes used to define each phenotype can be found in the Supplemental Tables.^25,26 (Table S2, Table S3)

Time-at-Risk

Patients were assumed at-risk for kidney failure after the third anti-VEGF exposure until the end of continuous drug exposure or of the study period. The end of continuous exposure was defined as a gap of over 180 days between injections. The cutoff of 180 days was chosen because the need for anti-VEGF typically decreases over time. In the Diabetic Retinopathy Clinical Research (DRCR) Retina Network’s clinical trials for treatment of diabetic macular edema, a median of 8–9 ranibizumab injections are given in the first year and that decreases to 2–3 in the second year across most studies.²⁷ The cutoff of 180 days would categorize most patients with diabetic macular edema being treated per the DRCR protocol as having a “continuous” drug exposure in the first two years of treatment.

Outcome

The outcome of interest was kidney failure, or end-stage kidney disease in which the operational definitions were based on Kidney Disease Improving Global Outcomes (KDIGO) stage 5 chronic kidney disease (CKD) treated with dialysis or kidney transplantation.^28,29 Based on our previous research, we implemented and tested two versions of the kidney failure phenotype.³⁰ The first kidney failure phenotype definition was aligned with KDIGO Stage 5 CKD, and was defined as having either a condition, observation, or procedure code related to end-stage kidney disease or kidney transplantation, two measurements of estimated glomerular filtration rate (eGFR) <15 mL/min/1.73m² at least 90 days apart, or two observation or procedure codes of kidney dialysis spaced at least 90 days apart. The Supplemental Tables contain the logic and full list of codes used to represent the definition. (Table S2, Table S4) The second kidney failure phenotype definition used an administrative code only limited definition that included the presence of a condition, observation, or procedure related to end-stage kidney disease. (Table S2, Table S4) The sensitivity, specificity, positive predictive value, and negative predictive value of the performance of each kidney failure definition was evaluated on a subset of databases using PheValuator, an R package for evaluating phenotype algorithms.^30,31 In brief, PheValuator uses machine learning to develop predictive models that serve as a probabilistic reference standard, computes the probability of having a disease of interest for each subject in the cohort, and uses that result to determine the rule-based phenotype performance metrics including sensitivity, specificity, positive predictive value, and negative predictive value. Based on the output from PheValuator, it was determined that the second kidney failure phenotype had better positive predictive value with similar sensitivity to the “complex” definition, thus the simpler kidney failure phenotype was used in this study. (Table S5)

Statistical Analysis

All analyses were performed in R using the Strategus execution pipeline.³² Strategus is a standardized method to call Health Analytics Data-to-Evidence Suite (HADES) open-source R packages developed by the OHDSI community for large scale analytics.^33,34 The pre-specified study protocol and end-to-end open executable source code are available on GitHub.²⁵ We developed an interactive website to promote transparency and allow for sharing and exploration of the results.³⁵

A). Incidence of Kidney Failure

Summary statistics were calculated to describe the baseline characteristics of patients in each anti-VEGF exposure cohort by database. These characteristics included demographic information (age, sex, race, ethnicity), blinding diseases (retinopathy due to diabetes mellitus, retinal vascular occlusion, age related macular degeneration, macular retinal edema), kidney diseases (renal impairment, chronic kidney disease, chronic disease of genitourinary system, kidney disease), Diabetes Comorbidity Severity Index (DCSI) score, and Charlson Comorbidity Index-Romano adaptation (CCI).^36,37 The baseline characteristics of patients were also stratified by the outcome of kidney failure. Differences between the groups were represented by the standardized mean difference and significant differences reported using a t-test.³⁸

The incidence proportion (number of outcomes divided by the total number of people at risk) and incidence rate (number of outcomes during the time-at-risk divided by the number of total person days) of kidney failure were calculated for each exposure cohort. Only kidney failure events while on treatment contributed to this analysis. Crude incidence proportions for each year were standardized to the 2015 U.S. population by age and sex using direct standardization and then averaged.^39,40

B). Comparative Risk of Kidney Failure

A series of study diagnostics were performed and the meta-analysis of comparative risks only included the results from databases that passed all evaluations. The study diagnostics are described in detail elsewhere and the thresholds used in this study are located in the study GitHub.^23,41 We used the large-scale propensity score method to match patients in each target and comparator exposure cohort comparison (aflibercept versus ranibizumab, bevacizumab versus ranibizumab, and bevacizumab versus aflibercept) using 1:1 propensity score matching. The propensity score model included a large number of baseline covariates (e.g., demographic characteristics, pre-existing conditions, measurements, procedures) as potential confounders and used the L₁-regularization technique to avoid model overfitting.^42,43 The outcome was time from cohort entry to kidney failure while on treatment. Cox proportional hazards models were used to estimate the risk of kidney failure while on treatment. A random effects meta-analysis was performed to combine per site hazard ratio estimates into a single network-wide estimate.⁴⁴

Sensitivity Analysis to Methodological Choices

Analyses were done to test the sensitivity of our design to methodological choices. In the main analysis, only kidney failure events that occurred while the patient was on-treatment with intravitreal anti-VEGF contributed to the calculations. As a sensitivity analysis, all calculations were repeated under an intent-to-treat design, in which all kidney failure events after cohort entry contributed to the calculations even if they occurred while the patient was not receiving intravitreal anti-VEGF. There was also no right-censoring; patients who switched anti-VEGF medications were analyzed according to the first anti-VEGF they received. The intent-to-treat sensitivity analysis was chosen for several reasons. Because kidney failure is a long-term outcome, potential adverse kidney impacts of the intravitreal anti-VEGF could be cumulative and manifest years after medication exposure. Furthermore, it is possible that patients who experienced an adverse kidney failure, for example proteinuria, were preferentially switched to another anti-VEGF medication or stopped therapy altogether thus biasing the main on-treatment study design. Since there was no right-censoring in the sensitivity analysis, patients who switched or stopped anti-VEGF were still included in the intent-to-treat analysis. The sensitivity analysis was performed on a subset of databases. (Table S1)

Results

In total, 12 unique databases within the OHDSI network were evaluated for inclusion in this study; data from 11 databases were included in the analysis of A) incidence of kidney failure, and data from 6 for B) comparative risk of kidney failure. (Table S1)

A). Incidence of Kidney Failure

Across all databases, there were 6,102,794 patients with blinding diseases, among whom 37,189 received ranibizumab, 39,447 aflibercept, and 163,611 bevacizumab. These patients contributed a total of 161,724 person-years of treatment exposure time. Baseline characteristics of patients included in each exposure cohort in the Optum’s de-identified Clinformatics^® Data Mart Database - Socio-economic Status database are presented in Table 6 and for all other databases in Table S7. Across databases, 3% to 22% of participants in all exposure groups were aged 70–74, 26% to 97% were male, 45% to 89% were White, 3% to 27% Black, 1% to 20% Asian, and 2% to 37% Hispanic. The average DCSI score ranged from 1.70 to 5.59, and the average CCI ranged from 1.37 to 4.98.

Table 6:

Baseline characteristics of patients in each exposure cohort (ranibizumab, aflibercept, and bevacizumab) in the Optum’s de-identified Clinformatics^® Data Mart Database - Socio-economic Status database (Clinformatics^®) (before propensity score matching).^*

	Ranibizumab N (%)	Aflibercept N (%)	Bevacizumab N (%)
Age Group (years)
<29	14 (0)	19 (0)	102 (0)
30–49	307 (3)	324 (3)	1919 (3)
50–69	2466 (25)	2534 (26)	16433 (23)
70–89	7205 (72)	6887 (70)	53036 (74)
>90	59 (1)	53 (1)	426 (1)
Sex
Male	3878 (39)	4012 (41)	28282 (39)
Female	6173 (61)	5805 (59)	43634 (61)
Race
White	7607 (76)	7109 (72)	52109 (72)
Black	1025 (10)	991 (10)	5956 (8)
Asian	254 (3)	308 (3)	1990 (3)
Ethnicity
Hispanic	750 (7)	923 (9)	8326 (12)
Non-Hispanic	8886 (88)	8408 (86)	60055 (84)
Diabetes Comorbidity Severity Index (DCSI) score, mean	3.8 (N=8504)	3.85 (N=8259)	3.99 (N=61182)
Charlson Comorbidity Index (CCI), mean	3.29 (N=8319)	3.69 (N=8353)	3.65 (N=60952)
Retinopathy due to diabetes mellitus	2537 (25)	3202 (33)	20236 (28)
Retinal vascular occlusion	1998 (20)	2056 (21)	14517 (20)
Age related macular degeneration	6752 (64)	5878 (57)	46496 (61)
Macular retinal edema	2047 (20)	2888 (29)	16488 (23)

^*Includes patients who had pre-existing kidney failure who were excluded in subsequent analyses

The baseline characteristics of patients in the Optum’s de-identified Clinformatics^® Data Mart Database, stratified by whether or not they developed the outcome during the time-at-risk, are shown in Table S8. A higher proportion of patients who developed kidney failure across all databases and exposure groups had baseline conditions including chronic kidney disease, kidney disease, renal impairment, and chronic disease of the genitourinary system. (Figure 1)

Figure 1:

Proportion of patients with baseline kidney diseases comparing those who did and did not develop kidney failure in each exposure group (ranibizumab, aflibercept, and bevacizumab) within each database.

Abbreviations:

CUMC = Columbia University Medical Center, CCAE = IBM Health MarketScan Commercial Claims and Encounters Database, MDCD = IBM Health MarketScan Multi-State Medicaid Database, MDCR = IBM Health MarketScan Medicare Supplemental and Coordination of Benefits Database, JHME = Johns Hopkins Medical Enterprise, JMDC = Japan Medical Data Center, USC = University of Southern California, Optum^® EHR = Optum^® de-identified Electronic Health Record data set, Clinformatics^® = Optum’s de-identified Clinformatics^® Data Mart Database - Socio-economic Status, NEU = IQVIA PharMetrics^® Plus for Academics Database, VA = Department of Veterans Affairs

The number of kidney failure outcomes range from 0 to 317 in each exposure group across all databases. (Table 9) The crude incidence proportion of kidney failure per 100,000 persons ranged from 0 to 1864 across the databases with an average of 699. The age-sex standardized incidence proportion per 100,000 persons ranged from 0 to 2389 across the databases with an average of 678. Incorporating person-time, the incidence rate of kidney failure ranged from 0 to 2661 per 100,000 person-years, with an average of 742. Among patients with a blinding disease, whether or not they were exposed to anti-VEGF, the average incidence proportion of kidney failure per 100,000 persons across all databases was 2921, the average standardized incidence proportion per 100,000 persons was 4822, and the average incidence rate was 782 per 100,000 person-years. (Table S10)

Table 9:

The crude incidence proportion per 100,000 persons, age-sex standardized incidence proportion per 100,000 persons, and incidence rate per 100,000 person-years in each exposure cohort across all databases.

	Patients at Risk	On Treatment Time (person-years)	Number of Outcomes	Incidence proportion per 100,000 persons	Age-sex standardized incidence proportion per 100,000 persons§	Incidence rate per 100,000 person-years
Ranibizumab*
Total	29609	29929.8	224
CCAE1	3799	3081.6	40	1053	682	1298
MDCR2	7604	8406.9	48	631	821	571
MDCD3	1265	1121.5	17	1344	2389	1516
Optum® EHR4	2520	2489.3	15	595	611	603
Clinformatics®5	8048	9841.5	65	808	838	660
JMDC6	203	133.5	0	0	0	0
JHME7	19	12.9	0	0	0	0
NEU8	2084	2179.8	14	672	399	642
CUMC9	117	114.3	0	0	0	0
VA10	3943	2538.6	25	630	250	980
USC11	<10	9.9	0	0	0	0
Aflibercept
Total	32153	37633.3	290
CCAE	3319	3249.4	56	1687	1659	1723
MDCR	4644	5532.9	23	495	162	416
MDCD	1717	1794.8	32	1864	1447	1783
Optum® EHR4	3282	4695.3	24	731	394	511
Clinformatics®	8056	11004.0	72	894	1409	654
JMDC	205	190.6	<10	488	335	525
JHME	574	655.6	8	1394	1830	1220
NEU	3696	3710.5	17	460	414	458
CUMC	335	466.0	0	0	0	0
VA	6266	6245.2	56	890	790	900
USC	59	89.0	<10	1695	320	1123
Bevacizumab
Total	108308	94160.8	695
CCAE	10508	6772.5	104	990	679	1536
MDCR	10625	9043.8	50	471	106	553
MDCD	3845	2630.9	70	1821	2203	2661
Optum® EHR4	11933	12639.4	69	578	1341	546
Clinformatics®	52642	50581.2	317	602	1072	627
JMDC	0	0.0	0	NA	NA	NA
JHME	286	226.0	<5	699	926	885
NEU	8331	6275.6	25	300	328	398
CUMC	74	41.1	0	0	0	0
VA	10037	5930.2	58	580	300	980
USC	27	20.1	0	0	0	0

^*Only patients who contribute at least 1 day to time-at-risk are included in this analysis.

^§Standardized to the 2015 U.S. Population by age and sex

¹CCAE = IBM Health MarketScan Commercial Claims and Encounters Database

²MDCR = IBM Health MarketScan Medicare Supplemental and Coordination of Benefits Database

³MDCD = IBM Health MarketScan Multi-State Medicaid Database

⁴Optum^® EHR = Optum^® de-identified Electronic Health Record data set

⁵Clinformatics^® = Optum’s de-identified Clinformatics^® Data Mart Database - Socio-economic Status

⁶JMDC = Japan Medical Data Center

⁷JHME = Johns Hopkins Medical Enterprise

⁸NEU = IQVIA PharMetrics^® Plus for Academics Database

⁹CUMC = Columbia University Medical Center

¹⁰VA = Department of Veterans Affairs

¹¹USC = University of Southern California

In the sensitivity analysis using the intent-to-treat study design, the incidence proportions and rates of kidney failure were higher than the main on-treatment analysis. The average crude incidence proportion of kidney failure among the anti-VEGF exposure groups was 3445 per 100,000 persons, the standardized incidence proportion was 3821 per 100,000 persons, and the incidence rate was 1192 per 100,000 person-years. (Table S11)

B). Comparative Risk of Kidney Failure

Half of the databases (N=6) passed study diagnostics for this part of the analysis. (Table S1) Results from the study diagnostics are shown in Table S12.

The hazard ratio (HR) estimates for risk of kidney failure across databases are shown in Figure 2. Comparing aflibercept to ranibizumab, the meta-analysis HR combining the estimates across all databases was 1.01 (95% confidence interval (CI) 0.70, 1.47, p=0.45). Comparing ranibizumab to bevacizumab, the meta-analysis HR was 0.95 (95% CI 0.68, 1.32, p=0.62), and comparing aflibercept to bevacizumab HR was 0.95 (95% CI 0.65, 1.39, p=0.60).

Figure 2:

Hazard ratio (HR) estimates for the risk of kidney failure among new users of monthly anti-VEGF medications while on treatment comparing ranibizumab, aflibercept, and bevacizumab. Results from each database are provided as well as the meta-analytic estimates.

Abbreviations:

Clinformatics^® = Optum’s de-identified Clinformatics^® Data Mart Database – Socio-economic Status, MDCR = IBM Health MarketScan Medicare Supplemental and Coordination of Benefits Database, MDCD = IBM Health MarketScan Multi-State Medicaid Database, CCAE = IBM Health MarketScan Commercial Claims and Encounters Database, NEU = IQVIA PharMetrics^® Plus for Academics Database, VA = Department of Veterans Affairs

The incidence of kidney failure is 678 per 100,000 patients who receive intravitreal anti-vascular endothelial growth factor agents. The risk of kidney failure appears similar among those who receive ranibizumab, aflibercept, or bevacizumab.

Results were similar on sensitivity analysis using the intent-to-treat study design. The meta-analysis HR comparing aflibercept with ranibizumab was 1.02 (95% CI 0.78, 1.34, p=0.43), ranibizumab with bevacizumab was 0.90 (95% CI 0.76, 1.07, p=0.89), and aflibercept with bevacizumab was 0.99 (95% CI 0.80, 1.22, p=0.56). (Table S13)

Discussion

In this retrospective cohort study across 12 databases in the OHDSI network evaluating 6.1 million patients with blinding diseases, the overall incidence rate of kidney failure while on treatment among patients who received at least 3 monthly intravitreal anti-VEGF was 743 per 100,000 person-years with a standardized incidence proportion of 678 per 100,000 persons. A greater proportion of patients who developed kidney failure, compared to those who did not, had pre-existing kidney disease and impairment. We did not find differences in the risk for developing kidney failure when comparing those who were exposed to intravitreal ranibizumab, aflibercept, and bevacizumab.

The incidence of kidney failure characterized in this study is higher than the overall population. The United States Renal Data System (USRDS) reports a standardized incidence proportion of 36.3 per 100,000 persons in 2020, that peaked at 43.1 per 100,000 persons in 2006.⁴⁵ Among the US databases, we find an incidence of kidney failure of 678 per 100,000 persons in the intravitreal anti-VEGF exposure cohorts, and 4821 per 100,000 persons among patients with blinding diseases. Similar trends were seen in international databases. In Japan, the standardized incidence proportion of kidney failure was estimated at 29.6 per 100,000 persons in 2016, which is again lower than the rates identified in this study using the Japan Medical Data Center.⁴⁶ The higher incidence of kidney failure identified in this study is likely due to differences in the underlying study population when compared to the overall population. Patients treated with anti-VEGF are older and often have cardiometabolic risk factors like diabetes that put them at higher risk for kidney disease. Although the standardized incidence proportion that we calculate controls for the age and sex of the underlying population, it does not take into account medical risk factors.

Although there is strong evidence for differing suppression of systemic plasma VEGF by the three commonly used intravitreal anti-VEGF medications, the estimated risks of kidney failure were not different among the three treatment groups. Differences in risk of kidney failure between intravitreal anti-VEGF, if present, are likely to be small. There could be several explanations for not finding a statistically significant difference in the risk of kidney failure between the three anti-VEGF medications. In the course of kidney disease progression from kidney damage to kidney failure, death is a major competing risk whereby patients with chronic kidney disease are more likely to die than to develop kidney failure.⁴⁷ Censoring bias from the competing risk of death could be masking differences between intravitreal anti-VEGF medications. Future studies could specifically focus on the outcome of death when comparing intravitreal anti-VEGF medications. Our study design may have impacted our results. For our main analysis, we chose an on-treatment study design where we captured cases of kidney failure while patients were actively receiving intravitreal anti-VEGF. This study design was chosen to best isolate the effects of intravitreal anti-VEGF on the outcome of kidney failure. It is harder to causally link kidney failure events that occur years after exposure to intravitreal anti-VEGF. However, as a result of the on-treatment study design, patients who switched therapies, for example from bevacizumab to ranibizumab, were right-censored and not included in the analysis. It is possible that patients who had experienced an adverse kidney event, for example acute kidney injury or proteinuria, were preferentially switched to another intravitreal anti-VEGF or stopped therapy altogether. This would imply that patients who were at higher risk for kidney failure were preferentially excluded from the analysis. We examined this potential source of bias by performing a sensitivity analysis using an intent-to-treat analysis study design. The intent-to-treat study design captured all cases of kidney failure, whether or not they occurred while on intravitreal anti-VEGF, and did not include any right-censoring. We found similar findings with the intent-to-treat analysis as the on-treatment study design, suggesting that selection bias was not a major driver of our results.

The lack of difference in risk of kidney failure by intravitreal anti-VEGF medication obviates the need to preferentially choose ranibizumab over aflibercept and bevacizumab, as was recommended by prior groups.^3,19 There are several implications for these findings. Since ranibizumab costs nearly 20 times more than bevacizumab, this could represent potential cost savings for the healthcare system.⁴⁸ There are racial and ethnic disparities in intravitreal anti-VEGF medication use. For example, studies have shown that Hispanic patients are less likely to receive ranibizumab and more likely to receive bevacizumab compared to otherwise similar non-Hispanic White patients.^49,50 There are also racial and ethnic disparities in the risk for kidney disease. Minoritized populations, including non-Hispanic Black and Hispanic patients, have nearly a four-fold and two-fold, respectively, higher prevalence of kidney failure compared to non-Hispanic White populations.⁵¹ By demonstrating that ranibizumab does not lower the risk of kidney failure, this frees the treating physician to choose from any of the anti-VEGF medications and not exacerbate existing health disparities.

There are still many lingering questions, specifically whether the receipt of intravitreal anti-VEGF increases the risk of kidney failure compared to not receiving anti-VEGF. There is both evidence for and against the increased risk of kidney failure and kidney disease associated with intravitreal anti-VEGF use.^13,17,52,53 A major challenge in many of these studies is the issue of indication bias–by comparing patients who receive anti-VEGF with patients who do not, it is hard to isolate the effects of the medication compared to the disease itself.⁵⁴ By choosing an active comparator design and comparing ranibizumab, aflibercept, and bevacizumab to one another, we mitigated the effects of indication bias. However, we lose the opportunity to specifically address the question of whether receipt of intravitreal anti-VEGF itself increases the risk of kidney failure. The finding that the baseline incidence of kidney failure among patients with blinding disease is similar or even higher to those who receive anti-VEGF suggests there might not be an increased risk. However, more work should be done to definitively address whether intravitreal anti-VEGF increases the risk of kidney failure.

Conducting this study in the OHDSI network allowed us to confidently estimate the comparative risk of kidney failure between intravitreal anti-VEGF medications. The large federated data network of the OHDSI community enabled us to identify 240,247 new users of monthly anti-VEGF — a sample size that would not be feasible in clinical trials or single center retrospective studies. Further, applying the same analytic procedure across different databases allowed us to directly compare results from different centers. Each database reported a different incidence of kidney failure, reflecting the diversity in underlying patient populations. We also found different practice patterns in use of anti-VEGF across our global data partners. For example, there were no instances of intravitreal bevacizumab use for blinding diseases in the Japan Medical Data Center since that medication has not been approved by Japan’s regulatory authority for drugs, Pharmaceuticals and Medical Devices Agency (PMDA), for intraocular use. The variability of findings across the databases, even when studying the same question, highlights the importance of applying a standard protocol across multiple data sources.²²

We employed best practices and standardized analytics developed by the OHDSI community. OHDSI uses large-scale propensity score estimation,^42,43 adjusting for thousands of variables (after excluding instruments, colliders, and mediators). This has been shown to be superior to manual variable selection,^43,55,56 and superior to attempts to empirically select confounders.⁴² We also employed a robust set of study diagnostics to each available database to ensure that valid conclusions were drawn from the comparative effect estimation analysis. Study diagnostics were evaluated with the results blinded and only databases that passed diagnostics had the results unblinded. For example, we used a set of negative controls (e.g., epidermoid cyst, impacted cerumen) where no causal effect is believed to exist with intravitreal anti-VEGF to assess potential systematic error.^25,35,57,58 Databases that exhibited potential systematic error were excluded from the population-level effect estimation analysis. Another study diagnostic evaluated baseline covariate balance after propensity score adjustment between the two comparative anti-VEGF medication groups. Again, databases that did not have sufficient covariate balance between comparison groups were excluded from the estimation of comparative risk of kidney failure as they are unlikely to produce a scientifically sound result.

There are several limitations to this work. Using data from the EHR or administrative claims to define phenotypes is a well-established methodology and we used OHDSI best practices to carefully design and validate these phenotypes. However the degree to which these phenotypes represent the real-world clinical state of patients is unknown. Although there were many advantages to using best practice analytics developed by the OHDSI community, there are limitations in their application to ophthalmic research. We were unable to consider the dose of medication that any individual received due to limitations in the current standardized analytics pipeline. For example, in our analysis, patients who receive bilateral intravitreal anti-VEGF medications monthly for a year were analyzed in the same way as someone who received unilateral intravitreal anti-VEGF medications despite being exposed to 2x the amount of medication. We do not believe this played an important role in the main results as the issue of dosage would be expected to occur in all anti-VEGF comparison groups. Despite these limitations, this is the largest study examining the association between intravitreal anti-VEGF and kidney failure and the largest application of the OHDSI network to an ophthalmic question to date.

In conclusion, we find an incidence rate of 743 per 100,000 person-years with a standardized incidence proportion of 678 per 100,000 persons of kidney failure, which is higher than the overall population, among patients who are on treatment with intravitreal anti-VEGF medications. The risk for progression to kidney failure is not different when comparing those who received ranibizumab, bevacizumab, or aflibercept. Practicing ophthalmologists should be aware that patients needing intravitreal anti-VEGF medications are at high risk for kidney failure, particularly those with pre-existing kidney disease and impairment, and might benefit from ongoing kidney health monitoring. There is no evidence in this study to support the practice of preferentially selecting ranibizumab over aflibercept and bevacizumab to avoid inducing kidney failure in patients with blinding diseases.