Data sources
This study used five different anonymized, integrated databases including medical and pharmacy claims. Diagnoses and procedures were identified based on International Classification of Diseases Ninth Revision Clinical Modification (ICD-9-CM) and Current Procedural Terminology (CPT) codes from patients’ medical claims, while medication use was assessed based on National Drug Codes (NDC) from patients’ pharmacy claims.
The IMS LifeLink® Health Plan Claims Database is a commercial database consisting of approximately 55 million patients from over 75 managed care organizations across the U.S. and several million Medicare managed-care enrollees from four U.S. geographical regions. This database consists of two files, including a claims file and an eligibility file. The claims file contains details on: medical and pharmacy claims, including date of service, place of service, ICD-9-CM codes, CPT codes, physician specialty, NDCs, drug quantity dispensed, days supplied, charged and paid amounts, and copayments. The eligibility file includes monthly medical and pharmacy eligibility flags as well as patient demographic data.
The MarketScan® Commercial Claims and Encounter and Medicare Supplemental and Coordination of Benefits Databases are constructed from privately insured paid medical and prescription drug claims for approximately 30 million employees and their dependents (in 2010) [
15]. “Commercial Claims and Encounter” and “Medicare Supplemental and Coordination of Benefits” are provided as separate databases. The MarketScan Medicare Database contains the healthcare experience of individuals with Medicare supplemental insurance paid by employers for approximately 3.42 million retirees (in 2010) [
15]. Medical claims capture details regarding dates of service, place of service, physician specialty, up to four ICD-9-CM diagnosis codes, CPT codes, charges, and health plan payments (both the Medicare-paid and employer-paid supplemental amounts are included). Pharmacy claims include details on dispense date, NDCs, quantity of medication dispensed, days supplied, and health plan payments. Eligibility file contains details on monthly medical and pharmacy eligibility, age, sex, and geographical region for individuals who are present in the claims file.
Clinformatics™ DataMart, a product of OptumInsight Life Sciences, Inc. (Eden Prairie, MN) (Optum), consists of a commercially insured population from a diverse group of health plans in the United States including 30 million individuals (during 2002 to 2007) [
16]. The medical and pharmacy claims files contain details on date of service, place of service, ICD-9-CM codes, CPT codes, provider type, NDCs, drug quantity dispensed, days supplied, charges, deductibles and copayments. The member file includes information on eligibility periods as well as patient demographic data.
The southern U.S. Medicaid program covers low-income or disabled individuals and consists of two files: a claims file, with details on medical and pharmacy utilization, including date of service, place of service, ICD-9-CM codes, CPT codes, physician specialty, NDCs, drug quantity dispensed, days supplied, and paid amounts; as well as an eligibility file, with details on monthly enrollment and patient demographics.
The most recent two-years of claims available from each database were used; IMS LifeLink Database (IMS)- including commercially insured claims, 07/2009-06/2011 MarketScan Commercial Database (MarketScanCommercial)- including commercially insured claims, 07/2009-06/2011; MarketScan Medicare Supplemental Database (MarketScanMedicare)- including employer-sponsored Medicare Supplemental plans only, 07/2009-06/2011; Clinformatics™ DataMart, a product of OptumInsight Life Sciences, Inc. (Eden Prairie, MN) (Optum)- including commercially insured claims, 04/2010-03/2012, and a Medicaid Database for a southern US state (Medicaid), 07/2008-06/2010. Given the large sample sizes in the IMS, MarketScanCommercial, MarketScanMedicare and Optum databases, a 10% random sample was selected from each of these, while the full Medicaid sample was used.
Patient selection
Patients were included in the study if they were ≥18 years of age and had at least one primary or secondary diagnosis of AF, determined based on the ICD-9-CM code 427.31, within a two-year period.
The index date was defined as the date of the first AF diagnosis. Patients were followed over a period of one year after the index date (i.e., the study period). All demographic and outcomes data were evaluated during the study period. Additional criteria for continuous eligibility was not applied; however, patients who were Medicare and/or health maintenance organization (HMO) eligible anytime during the two-year period were excluded in the Medicaid database, in order to ensure the availability of all claims within this database.
Study measures
Demographics (i.e., age and sex) of patients, stroke risk scores, comorbidities, use of anticoagulants, stroke related hospitalizations, frequency of INR tests and all-cause resource use were assessed during the one year period following the first occurrence of the diagnosis of AF.
The level of stroke risk (i.e., low, medium, high) was assessed using CHADS
2 and CHA
2DS
2-VASc scores. For determination of the CHADS
2 score, one point each was assigned for the presence of Congestive heart failure (CHF), Hypertension (HTN), Age ≥75 years, or diabetes (ICD-9-CM codes available in Additional file
1: Table S1). Two points were assigned for a history of stroke or transient ischemic attack (TIA). For determination of the CHA
2DS
2-VASc score [C = CHF/Left ventricular dysfunction (LVH), H = HTN, A = Age (≥75), D = Diabetes, S2 = Stroke/TIA, V = Vascular disease, A = Age 65–74, and Sc = Sex category], one point each was assigned for the presence of CHF/ LVH, HTN (systolic blood pressure >160 mmHg), age being 65–74 years, diabetes, vascular disease (coronary artery disease, heart attack, peripheral artery disease, aortic plaque) and sex category being female. Two points were assigned for the presence of each of the following factors: Age ≥75 and history of stroke, TIA, or thromboembolism. Patients with AF were subsequently assigned to one of the following categories based on their risk factors for stroke; low risk (0 points), moderate risk (1 point), or high risk (≥2 points).
The percentage of patients using anticoagulants was determined based on prescription claims with national drug codes (NDCs). Patients with at least one prescription claim for an anticoagulant medication were categorized as receiving treatment. The number and percentage of patients with a gap in anticoagulant therapy and the time to the first gap in anticoagulant therapy (i.e., the number of days from the start of an anticoagulant drug to the start of a gap in anticoagulant therapy defined as 60 days or more) were determined using pharmacy claims. Anticoagulant medical possession ratio (MPR) was calculated as follows: ([Days supply any anticoagulant] - [Last fill days supply])/([Last prescription fill date in data set] - [First prescription fill date]). In cases where the days supply of any two anticoagulant medications overlapped by more than 25% of their total supply, only the unique days were included in the numerator (i.e., overlapping prescriptions with the same date of service were not double counted).
Percentage of patients hospitalized for stroke was identified based on ICD-9-CM (primary code indicative of ischemic and/or hemorrhagic stroke) and CPT/UB-92 (uniform billing) codes (indicative of hospitalization or outpatient visits) from medical claims. Among patients hospitalized for stroke, outpatient use of anticoagulants was also determined. The number and percentage of patients with a bleeding event were identified based on inpatient hospitalizations associated with a primary ICD-9-CM code indicative of bleeding. Condition specific ICD-9-CM codes are provided in the Appendix (Additional file
2: Table S2, Additional file
3: Table S3).
INR frequency of occurrence was evaluated following warfarin use. The number of average unique INRs per month of warfarin treatment was calculated as number of unique INRs during treatment months divided by number of months taking warfarin (calculated based on dispense days and days supplied). The monthly INR quality ratio was calculated as the number of months with one or more INRs during months taking warfarin divided by the number of months taking warfarin. Ratios range from 0–1, with higher values representing better INR quality ratios. INRs reflect only those captured in claims data. If an INR was performed during a visit and not recorded on a claim, it is not captured in the tool.
The percentage of patients with an inpatient hospitalization, an ER visit and an outpatient visit were reported for the study period, as well as the mean number of visits per patient for each all-cause resource category.
Data analyses
All analyses were descriptive in nature with no multivariate analyses performed. Categorical variables were summarized using counts and sample proportions. Mean values were reported for continuous measures. Analyses stratified by stroke risk level, anticoagulant treatment status and by age group were conducted.
SAS software (Version 9.3, SAS Institute, Cary, NC) was used for extracting medical/pharmacy claims and demographic information from all databases for AF patients, and for organizing this information so that the data were in the proper format to be utilized by the software tool, which carried out the analyses that produced the study outcomes. The software tool is condition-specific and Health Insurance Portability and Accountability Act–compliant, and enables the uploading of pharmacy and medical claims data via a simple point-and-click method to produce results for a series of predetermined and user-defined measures and to generate sample-specific reports. This study did not directly involve human subjects, and all study data were anonymized prior to being received; therefore, this study did not require ethical review or approval in order to be conducted.