Patients with a first injection of IVT anti-VEGF between January 1, 2014, and December 31, 2018, were identified from the French National Health Insurance Databases, i.e., Système National des Données de Santé (SNDS), covering 99% of the French population (i.e., more than 66 million inhabitants). The SNDS contains anonymized individual data on all reimbursed health expenditures, including drugs (coded according to the Anatomical Therapeutic Classification [ATC]), ambulatory health care, and outpatient laboratory tests. The SNDS does not contain the medical indications of the reimbursements recorded, but the existence of 100% coverage of the patient for care related to a serious and costly long-term illness (cancer, diabetes, ischemic heart disease, etc.) is indicated and coded according to the International Classification of Diseases, 10th Revision (ICD-10), as well as the date of the onset of the illness. The SNDS also contains sociodemographic data, such as age, sex, area of residence (zip code), and deprivation index, corresponding to an index of the participant’s area of residence calculated from socioeconomic data . This information is linked, via a unique and anonymous individual identifier, to the national hospital database, which contains data on all stays in public and private hospitals (i.e., admission and discharge dates, ICD-10 coded diagnoses, care and surgical procedures performed, costly drugs, or medical devices provided). The SNDS has been described and used in other pharmacoepidemiological studies [18,19,20,21,22]. This research adheres to the Declaration of Helsinki of 1964 and its later amendments. Institutional review board approval was not required because the data in the SNDS are fully anonymous.
Individuals 18 years of age or older with at least one IVT injection of aflibercept (ATC code S01LA05) or ranibizumab (S01LA04) between 2014 and 2018 registered in the SNDS databases were initially identified. IVT bevacizumab users were not included because in France this drug is only prescribed in hospitals and its use is rare (n = 180 patients who initiated treatment between 2014 and 2018). Patients with a first injection of IVT ranibizumab or aflibercept (i.e., new users) were included. New use was defined as the absence of any IVT anti-VEGF (i.e., ranibizumab, aflibercept, pegaptanib-S01LA03, and bevacizumab-L01XC07) injection in the last 2 years. The index date corresponds to the first IVT anti-VEGF dispensing date recorded during the inclusion period.
Patients who initiated aflibercept therapy were compared to those who initiated ranibizumab therapy.
In the main analyses, we hypothesized that only sufficient anti-VEGF accumulation in the blood could be consistent with the occurrence of a systemic adverse outcome . Thus, patients who received a single IVT anti-VEGF injection, those who received different molecules, or those who were lost to follow-up or died within 3 months of the index date were secondarily excluded. These patients were re-included in a secondary analysis described below.
Patients who initiated treatment were followed from the index date plus 3 months until one of the following events, whichever came first: occurrence of an outcome of interest or cancer, end of study (12-31-2019), loss to follow-up, change of IVT anti-VEGF molecule, or treatment interruption. A treatment interruption was considered to have occurred when a period of more than 6 months was observed between two IVT anti-VEGF injections or between the last IVT anti-VEGF injection and the end of follow-up. The interruption date corresponded to the most recent date in these intervals, and patients were censored 9 months after the interruption date.
The outcomes were defined as the occurrence of MI or stroke resulting in hospitalization during the follow-up or death. MI and stroke were identified by using the main or related diagnoses of incident hospital stays registered in the national hospital database (ICD-10 codes for MI: I21, I22; ICD-10 codes for stroke: I60, I61, I62, I63, I64). In the main analyses, MI and stroke that occurred between the index date and the start of follow-up (i.e., 3 months after the index date) were considered as medical history.
Covariates: Sociodemographic Characteristics and Medical History of Intravitreal Anti-VEGF Users
Age, sex, and deprivation index were measured at the index date. History of ischemic heart disease (including MI), arrhythmia or conduction disorders (including non-valvular atrial fibrillation), heart failure, stroke, chronic smoking index, morbid obesity index, and chronic alcoholism index were measured within 6 years before the index date. Dispensing of antihypertensive, lipid-lowering, antiplatelet, antidiabetic, anticoagulant, and antiarrhythmic treatments was measured during the year preceding the index date. Likely and possible indications for IVT anti-VEGF were inferred on the basis of medical information recorded in the 2 years before and after the index date. Medical history was identified on the basis of ICD-10 codes recorded on the day of a hospitalization or diagnosis of a long-term illness, or reimbursements for medications or disease-specific care. Medical history and indications for IVT anti-VEGF therapy were identified using the algorithms provided in the electronic Supplementary Material, Table S1.
New users of aflibercept and ranibizumab were described according to the distribution of covariates mentioned previously. Incidence of MI, stroke, and death was calculated for each group. Kaplan–Meier analyses were used to assess the 6-year risk of MI, stroke, or death in new users of aflibercept and ranibizumab.
The risks of MI, stroke, or death in new users of aflibercept compared to new users of ranibizumab were assessed using univariate and multivariate Cox proportional hazards regression models. Multivariate models were adjusted on year of inclusion, sociodemographic variables (i.e., age, sex, deprivation index, and region of residence), cardiovascular diseases or risk factors (i.e., history of ischemic heart disease including MI, heart failure, stroke, arrhythmia diagnosis or antiarrhythmic treatment, history of treatment by antihypertensive, lipid-lowering, antiplatelet, oral anticoagulant, and antidiabetic drugs, history of chronic smoking, morbid obesity, and chronic alcoholism), and origin of the prescription (i.e., private or hospital practitioner).
The risks of MI, stroke, or death in patients initiating treatment with aflibercept versus ranibizumab were also assessed using multivariate models weighted by the inverse propensity score to validate the findings of the main multivariate models. Weights were derived from the reciprocal of the propensity scores containing the same covariates as those described in the main multivariate models. To reduce instability induced by large weights, the stabilized weights were computed and then were truncated at the first and 99th percentiles. The standardized differences method for assessing balance in observed baseline covariates between new users of aflibercept and new users of ranibizumab was applied to compare the prevalence of covariates in the stabilized weighted sample. Imbalances below 10% were considered negligible. Adjustments on age, sex, and region of residence were added in multivariate models weighted by the inverse propensity score.
Subgroup analyses were conducted according to history of ischemic heart disease (including MI) or stroke, treatment indications (AMD or DME), history of diabetes, age groups (≤ 70, 71–80 or > 80 years), sex, and number of IVT anti-VEGF injections received during the follow-up (2–3, 4–8, or > 8).
Secondary analyses were conducted to assess potential variations in outcomes by (1) changing the definition of censoring (i.e., 1, 3, or 6 months after treatment interruption versus 9 months in the main analyses), (2) considering only ischemic stroke as an event (ICD-10 code: I63) instead of all strokes (i.e., ischemic and hemorrhagic), and (3) reintroducing patients who received a single injection of IVT anti-VEGF. In the latter analysis, patients were followed from the index date until the occurrence of one of the censoring events.
All statistical tests were two-tailed, and we considered p values less than 0.01 to be statistically significant. Statistical analyses were performed using SAS Enterprise Guide software, version 7.15.