Data Sources

We obtained outpatient and carrier standard analytic files and the corresponding denominator files from the US Centers for Medicare & Medicaid Services for January 1, 2004, through December 31, 2009, for all Medicare beneficiaries having any claim with a diagnosis for neovascular AMD (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] diagnosis code 362.52). This time frame allowed for a 3-year study period with 1 year of history and 1 year of follow-up. The standard analytic files contain encrypted identifiers that allow longitudinal follow-up of beneficiaries. The outpatient files contain claims data for outpatient institutional care covered under Medicare Part B. The carrier files contain claims data for noninstitutional physician services also covered under Medicare Part B. The denominator files include demographic characteristics, information about Medicare eligibility and enrollment, and death date, if applicable. Medicare data are made available for research purposes 12 to 18 months after the claim is filed.

We limited the study population to beneficiaries 65 years or older living in the United States. We restricted the analyses to include beneficiaries only during periods of fee-for-service enrollment, because claims data from Medicare managed care plans were not available. The institutional review board of the Duke University Health System approved this study.

Study Population

We selected all Medicare beneficiaries with a diagnosis of neovascular AMD from 2005 to 2008. For each beneficiary, we identified the earliest outpatient or carrier claim that contained a diagnosis of neovascular AMD. We considered the date of this claim to be the index date. We deemed a beneficiary to be newly diagnosed if we found a subsequent AMD claim in the year following the index date and if no AMD diagnosis was recorded in the year prior to the index date. We required beneficiaries to be enrolled in fee-for-service Medicare throughout the ascertainment period. To coincide with the onset of the use of bevacizumab for neovascular AMD, we limited the sample to patients with an index date between January 1, 2006, and December 31, 2008. We used data from 2009 for follow-up.

To enable comparisons with a recently published analysis in the same database, we assembled the source population of fee-for-service Medicare beneficiaries with at least 1 diagnosis of neovascular AMD on an outpatient or carrier claim.

Diagnostic Procedures, Treatments, and Physician Visits

We searched outpatient and carrier claims in the year after the index date for procedures related to the treatment and management of neovascular AMD. Diagnostic procedures of interest included fluorescein angiography (Current Procedural Terminology [CPT] code 92235) and optical coherence tomography (92135). Therapies of interest included thermal laser photocoagulation (67220, 0017T), photodynamic therapy with intravenous verteporfin (67221, 67225), and intravitreous injection (67028) of pegaptanib, bevacizumab, or ranibizumab. We used 2 methods to identify the medications associated with intravitreous injections ( Table 1 ) . First, we searched for medication-specific Healthcare Common Procedure Coding System (HCPCS) medication reimbursement codes. Second, because medication-specific codes are not available immediately after FDA approval, we identified medications during these periods by using nonspecific reimbursement codes in combination with reimbursement amounts from the Medicare quarterly pricing reports. We defined discontinuation of anti-VEGF therapy as the absence of treatment for 12 months, with the first day of the 12-month period serving as the discontinuation date. For all of the procedures listed above, we used HCPCS modifier codes on each claim to determine which eye (ie, left, right, or both) was involved.

We searched carrier claims in the year after the index date for evidence of different types of evaluation and management visits. We defined evaluation and management visits coded with CPT codes 99201 through 99215 (ie, ophthalmologic evaluation and management) or 92002 through 92014 (ie, general evaluation and management) as ophthalmology visits if the physician specialty was ophthalmology.

Other Variables

We obtained demographic characteristics including age, sex, and race from the denominator files. Comorbid conditions were based on the presence of relevant ICD-9-CM diagnosis and procedure codes on any inpatient, outpatient, or carrier claim in the year of the index diagnosis. Specifically, we searched for evidence of atrial fibrillation, cancer, cataract, chronic kidney disease, chronic obstructive pulmonary disorder, dementia, diabetes mellitus, glaucoma, heart failure, ischemic heart disease, myocardial infarction, and stroke or transient ischemic attack using definitions developed by the Chronic Condition Warehouse.

Statistical Analysis

In the source population and in the derived study population, we describe the number of claims and unique patients with a neovascular AMD diagnosis in each year. We describe the baseline characteristics of the study population by presenting frequencies with percentages for categorical variables and medians with interquartile ranges for continuous variables. We tested for differences between annual cohorts using χ ² tests for categorical variables and Kruskal-Wallis tests for continuous variables. In the source population, we calculated the proportion of beneficiaries who received at least 1 intravitreous injection. In the study population, we estimated the proportions of beneficiaries who underwent diagnostic procedures or were treated for age-related macular degeneration, both overall and with specific procedures, using the cumulative incidence function. Among beneficiaries treated with an anti-VEGF agent, we used the mean frequency function to estimate the cumulative mean number of injections per eye during the year after the initial injection. We used the cumulative incidence function and the mean frequency function to account for the competing risk of mortality and the fact that follow-up differs across individuals. Length of follow-up is not uniform because beneficiaries may die or switch to managed care (at which time fee-for-service claims are no longer filed). In all cases, the data were censored at the time of the event. Because of changes in treatment availability during the study period, we present the results of the analyses stratified by the year of the index diagnosis.

We used a Cox proportional hazards regression model to estimate the relative risks of treatment with intravitreous injection within 1 year after the index date. The model adjusted for age, sex, race, comorbid conditions, and the date of the index diagnosis. We estimated rates of discontinuation of anti-VEGF therapy using the Kaplan-Meier method. To ensure that these estimates were not artificially inflated by right-censoring, we used discontinuation data only through 12 months before the date on which each patient’s data were censored. For example, for a patient with data censored after June 30, 2009, the latest possible discontinuation date was June 30, 2008. Therefore, we only used that patient’s data through June 2008 in the discontinuation estimator. We used a series of Kaplan-Meier estimators to describe the distribution of the number of anti-VEGF injections received within 12 months of the first injection. We used SAS statistical software version 9.2 for all analyses (SAS Institute Inc, Cary, North Carolina, USA).