Both bivariate and multivariate analytic approaches were used to examine 2010 NAMCS data in order to answer the research question. The NAMCS was designed and is used to collect data on the actual use and provision of ambulatory care services throughout the US. A national sample of ambulatory care visits are surveyed in order to collect the data. A complex four-stage probability sampling design is employed in the data collection process. A description of the sampling strategy is discussed elsewhere [25]. Because they were the most recently available data at the onset of this study, the 2010 NAMCS data were used. These data are weighted, by the survey designers, to be nationally representative of patient health records.
As recommended by the Center for Disease Control and Prevention’s (CDC) National Center for Health Statistics (NCHS), all analyses were performed on weighted data. The weighting, as calculated, uses the most recently available census data to provide a stratified representation of the nation’s primary care patient population. Only weighted data are reported in the results.
The Patient Record Form is the survey instrument for data abstraction from the medical record. The NAMCS patient record form is completed by ambulatory care clinic staff for a systematic random sample of patient visits during a randomly assigned 1-week reporting period. The data obtained include demographic characteristics of patients, expected source(s) of payment, patients’ complaints, diagnoses, diagnostic/screening services, procedures, medication therapy, disposition, types of providers seen, causes of injury, and certain characteristics of the clinic facility, such as geographic region and metropolitan status.
For this research project, the study population was US adults with a diagnosis of AF. The ICD-9 code for AF is 427.31. The covariates or independent variables for this research were: CHADS2 score levels (low and high), sex (male/female), race (Caucasian/Non-Caucasian), geographic locale (rural/non-rural), education attainment in a patient’s zip code (20% of adults with a university degree/? 20% of adults with a university degree), poverty level in patient’s zip code (10%/? 10%), health insurance status (insured/noninsured), and primary health care provider (HCP) seen (yes/no). Patient age was also a covariate in this study. For some of the analyses performed, CHADS2 age ranges are reported (75 years/?75 years), in other analyses three age-ranges were included (18–39, 40–64, and???65). For the logistic regression analysis performed the two age ranges used were 40–64 years and???65 years. All of the study covariates were recoded from their original configuration for analyses. Re-coding entailed either collapsing categories and/or removing unknown responses.
CHADS2, one of the study covariates or independent variables, is a clinical prediction tool for estimating the risk of stroke for patients with non-valvular AF. AF is associated with thromboembolic stroke. A CHADS2 score ranges from 0 to 6 and is computed from the following variables: congestive heart failure (1 point), hypertension (1 point), age ?75 years (1 point), diabetes (1 point), and prior stroke or transient ischemic attack (TIA) (2 points). Patients with a score ranging from 0 to 1 were coded as having a low CHADS2 score and those with a score???2 were coded as having a high CHADS2 score. The stroke/TIA variable was computed from six, three digit ICD-9 codes for stroke (434.00, 434.01, 434.10, 434.11, 434.90, 434.91) and six, three digit ICD-9 codes for TIA (435.0, 435.1, 435.2, 435.3, 435.8, 435.9). Those patients with any of the 12 possible conditions were coded as 1 and those without any of the conditions were coded as 0. In contrast, the CHA2DS2-VASc includes the additional criteria: 2 points for age ?75 years, 1 point for vascular disease (previous MI, peripheral artery disease, aortic plaque), 1 point for female sex, and 1 point for ages 65 to 74 years. CHA2DS2-VASc scores range from 0 to 9. Scores ?2 points were considered to be high risk and 0 was considered a low risk for stroke.
The findings from an earlier study comparing the newer CHA2DS2-VASc to the older CHADS2 [13] found no statistically significant differences between the risk stratification results for different population groups and the associated warfarin prescribing patterns. Further, Odum, et al., [17] noted that the implementation of this risk schema warrants further and continual assessment. With these considerations in mind, we examined differences in prescribing patterns and risk stratification comparing both tools.
Warfarin prescription was the dependent variable for this study. Using the CDC’s New Ambulatory Care Drug Database System for NAMCS data, prescribed drugs were classified as warfarin or other. Six separate drugs were re-coded as the variable warfarin. These were: Jantoven, Athrombin K, Coumadin, Panwarfin, Warcef, and warfarin.
In order to identify patients at risk for bleeding, the three digit ICD-9 codes associated with GI hemorrhage were merged into a variable called bleed risk. All patients with a risk for GI bleeding were excluded from the study.
Statistical Package for Social Scientists (SPSS, IBM, Chicago, IL, version 23.0) was used to complete all statistical analyses and alpha was set at p???0.05. Bivariate contingency table analysis was conducted to establish the relationships between each of the covariates and the dependent variable. Bivariate analysis tests whether or not a statistically significant relationship exists between an outcome or dependent variable and a predictor or independent variable. Bivariate analysis is not a stratified analysis. A chi square was computed as the test statistic for differences between percentages. Multivariate logistic regression analysis, to produce adjusted measures and eliminate confounding, was also performed using SPSS (version 23.0).
The Institutional Review Board (IRB) at the researchers’ institutions recognize that the analysis of de-identified and publicly available data does not constitute human subjects research as defined in federal regulations and as such does not require IRB review. Hence, human subjects’ approval was not necessary nor sought since this was a de-identified data only study.