Clinical Characteristics, Oral Anticoagulation Patterns, and Outcomes of Medicaid Patients With Atrial Fibrillation: Insights From the Outcomes Registry for Better Informed Treatment of Atrial Fibrillation (ORBIT‐AF I) Registry

We used data from the Outcomes Registry for Better Informed Treatment of Atrial Fibrillation (ORBIT‐AF). The ORBIT‐AF study design has been published.⁸ Briefly, ORBIT‐AF is a prospective, US‐based outpatient registry of AF management. Patients ages 18 years or older provided informed consent and were enrolled at 176 clinical sites. ORBIT‐AF sites represent a geographically diverse set of providers from multiple specialties, including cardiology, electrophysiology, and primary care. Information on demographics, medical comorbidities, AF history, procedures, medications, and provider characteristics are abstracted and entered into a web‐based data collection form. Prospective data on hospitalizations, disease progression, medications, procedures, and vital status is collected at ≈6 month intervals for 2 years after initial enrollment. The Duke Clinical Research Institute coordinates and manages the study.

The ORBIT‐AF Registry enrolled 10 135 AF patients between June 2010 and August 2011. After excluding patients who were missing information on insurance or OAC status (n=2), we were left with a final analytic population of N=10 133 patients who completed the baseline study visit.

The primary outcome of interest for this study was the proportion of patients who were using OAC (warfarin, dabigatran, or rivaroxaban) at baseline or during follow‐up. Current medications at each visit were abstracted from medication lists in the medical record and entered into online data collection forms. Because ORBIT‐AF enrolled some patients before the approval of rivaroxaban and Apixaban was not approved until after the enrollment period (2010–2011), OAC at baseline was defined as warfarin or dabigatran use. As a secondary endpoint, we also examined the proportion of patients who were using novel anticoagulants (dabigatran or rivaroxaban) at baseline or at any visit during the 2‐year follow‐up period. Secondary outcomes of interest included time in therapeutic range (TTR) for warfarin users, all‐cause mortality, first stroke/systemic embolism, and first major bleed. TTR was calculated using the Rosendaal method. Daily INR values were imputed between the first and last INR measured during follow‐up. For patients with 60 days or more between any 2 INR measurements, this time span was excluded from the TTR calculation. Using the Rosendaal method of linear interpolation between each pair of measured INR values, we calculated TTR as the percent of days with INR between 2 and 3 during this period. Stroke/systemic embolism was defined as the composite of all stroke (both ischemic and hemorrhagic) and systemic (non–central nervous system) embolism at any point during follow‐up. Major bleeding was defined as any bleeding event meeting International Society of Thrombosis and Hemostatis major bleeding criteria at any point during follow‐up.

Patients were classified as Medicaid or non‐Medicaid based on insurance status as recorded at the baseline visit. Because Medicaid eligibility is based on income and is therefore a proxy for lower SES,⁹ we classified Medicaid patients as any patient with Medicaid insurance at baseline, regardless of additional forms of insurance. Of 470 patients insured by Medicaid, N=47 (10%) also had private insurance, n=267 (56.8%) also had Medicare insurance, and n=5 (1.1%) also had military health care.

We compared baseline characteristics by Medicaid insurance status at the baseline visit (yes or no) among those N=10 133 patients who completed the baseline visit. Continuous variables are presented as medians (interquartile range [IQR], 25th–75th percentiles) and categorical variables are presented as counts (proportions). Differences between Medicaid and non‐Medicaid patients were compared using Wilcoxon rank‐sum tests for continuous variables and chi‐square test for categorical variables. We used similar methods to compare unadjusted antithrombotic treatment rates at baseline and over 2 years of follow‐up among Medicaid and non‐Medicaid patients. To characterize OAC use over follow‐up, we used data from all available visits until the patient was censored because of death or loss‐to‐follow‐up. Of N=10 133 patients, N=6935 (68.4%) had 24 full months of follow‐up. As a sensitivity analysis, we examined unadjusted OAC and NOAC use among those patients with a full 24 months of follow‐up and found that patterns of OAC and NOAC treatment were similar. Therefore, we present estimates of OAC use over follow‐up using all available visit data over the 2‐year study period. We repeated these analyses stratifying by stroke risk as estimated by CHADS₂ (<2, ≥2) and CHA₂DS₂‐VASc (<2, ≥2). The CHADS₂ score is calculated by assigning 2 points for a history of past stroke or transient ischemic attack (TIA) and 1 point each for congestive heart failure (CHF)/left ventricular (LV) dysfunction, hypertension, older age (≥75 years), and diabetes (range, 0–6).¹⁰ The CHA₂DS₂‐VASc score assigns 2 points each for age ≥75 years and history of stroke or TIA, and 1 point each for CHF/LV dysfunction, hypertension, diabetes, female sex, 65≤age≤74, and vascular disease (range, 0–9).¹¹ We also examined patterns of reported contraindications among those patients definitively indicated for OAC (CHADS₂ ≥2) because of the possibility that reduced access to care and fewer resources may be associated with less‐consistent monitoring and/or medication nonadherence. We limited this analysis to those patients who were OAC indicated because of stroke risk because we were most interested in reasons for nontreatment among the subgroup of patients with the greatest potential stroke reduction benefit with OAC treatment. Among patients who were not treated with OAC at baseline, we compared frequencies of contraindications to OAC by Medicaid status overall and stratified by CHADS₂ score using chi‐squared tests.

We assessed the association between receiving baseline OAC and Medicaid status using logistic regression with generalized estimating equations (GEEs) to account for the correlation of patients within sites. All multivariable models were adjusted for the following variables as documented at the baseline visit: demographics (age, race [African American/Hispanic/white/other], and sex); medical history (smoking, cancer, hypertension, osteoporosis/hip fracture, diabetes, hypothyroidism, gastrointestinal bleed, obstructive sleep apnea, insufficient kidney function [dialysis or estimated glomerular filtration rate <60 mL/min per 1.73 m²], hyperlipidemia, anemia, cognitive impairment/dementia, frailty, coronary artery disease, chronic obstructive pulmonary disease [COPD], alcohol or drug abuse, peripheral vascular disease, sinus node dysfunction, stroke/TIA, CHF, and valvular disease [all yes/no]), vital signs (heart rate [beats/min], blood pressure [mm Hg], and body mass index [BMI; kg/m²]), left atrial diameter, AF history (type of AF [new onset/paroxysmal AF/persistent AF/permanent AF], past cardioversions, past antiarrhythmic drugs, past AF interventional therapy), and functional status (living independently/living with assistance/residence in assisted living/residence in skilled nursing home/bedbound). All continuous variables were tested for linearity. Nonlinear relationships were found for age, systolic blood pressure, and BMI, which were accounted for using linear splines and truncation. Missing data (<0.2% except left atrial diameter enlargement type, which has 14% missing) on the covariates used in the modeling were handled using multiple imputation. Among patients who were taking OAC, we compared receipt of novel oral anticoagulants (dabigatran or rivaroxaban) versus warfarin by Medicaid status during 2 years of follow‐up using the same modeling approach. The association between TTR and Medicaid status was assessed using multivariable‐adjusted GEE linear regression modeling. Finally, we compared clinical outcomes by Medicaid status using Cox proportional hazards regression with a robust sandwich covariance estimate.

All P values presented are 2‐sided, and we considered P<0.05 to be statistically significant for all analyses. Statistical analysis was performed using SAS software (version 9.3; SAS Institute Inc., Cary, NC). All ORBIT‐AF study participants gave written informed consent before enrollment. The Duke Institutional Review Board (IRB) approved the ORBIT‐AF Registry, and all participating sites obtained approval from local IRBs before entering patient data.

Among 10 133 patients enrolled in ORBIT‐AF, n=470 (4.6%) had Medicaid insurance at baseline; N=95 patients (0.98%) were uninsured. Distribution of baseline characteristics of the study population by Medicaid status is shown in Table 1. Compared with patients who did not have Medicaid insurance, Medicaid patients were, on average, younger, more likely to be female, less likely to be white, and more likely to have less than a high school education. Burden of comorbidities was greater in the Medicaid population, with higher rates of CHF, past stroke/ TIA, COPD, smoking, diabetes, and hypertension than among non‐Medicaid patients. High stroke risk as estimated by CHADS₂ ≥2 was more common among Medicaid patients (78.7%) than non‐Medicaid patients (71.5%; P=0.0004).

The unadjusted rates of antithrombotic treatment at baseline and during follow‐up are shown in Table 2. The overall rate of OAC use was 76.1% at baseline and 82.0% over follow‐up. The overall rate of NOAC use at baseline or follow‐up was 16.1%. Medicaid patients were slightly less likely to receive OAC at baseline (72.8% vs 76.3%; P=0.08), but this difference was not statistically significant. Among patients with high stroke risk (CHADS₂ ≥2), Medicaid status was associated with significantly lower rates of OAC receipt (73.2% vs 80.5%; P<0.001). This difference persisted when defining high stroke risk according to CHA₂DS₂‐VASc >2, with lower rates of OAC among Medicaid patients versus non‐Medicaid patients (73.7% vs 78.3%; P=0.02). Results from multivariable regression models of the association between Medicaid status and OAC receipt are shown in Table 2. Among patients with high stroke risk (CHADS₂ ≥2), Medicaid patients were 32% less likely to receive OAC at baseline than non‐Medicaid patients after adjustment. Whereas Medicaid patients overall were less likely to receive OAC at baseline or during follow‐up, these differences were not statistically significant after adjustment for clinical covariates. Among warfarin users, TTR was higher for Non‐Medicaid patients than for Medicaid patients (median=68 vs 60; P<0.0001). This difference persisted after multivariable adjustment (TTR difference of −2.93; 95% CI, −5.67, −0.19).

Among patients who were taking OAC, Medicaid patients were less likely to receive NOACs than non‐Medicaid patients at baseline or during follow‐up (12.1% vs 16.3%; P=0.016). Medicaid patients were less likely to receive NOACs in unadjusted models, but this association was attenuated after adjustment. Similar results were found when stratifying patients by stroke risk according to CHA₂DS₂‐VASc, with lower rates of NOAC receipt in Medicaid patients (15.9% vs 17.6%; P=0.46) and attenuation of results after multivariable modeling adjustment.

Figure displays the reported OAC contraindications by Medicaid status, among patients with high stroke risk (CHADS₂ ≥2; n=1449). Medicaid patients were more commonly described as unable to adhere/monitor warfarin as a reported contraindication to OAC, whereas non‐Medicaid patients were more likely to have past bleed, patient refusal, and frequent falls/frailty listed as a contraindication to OAC.

Unadjusted and adjusted rates of adverse clinical outcomes by Medicaid status are shown in Table 3. Medicaid patients had higher unadjusted rates of adverse events compared with non‐Medicaid patients, including mortality (6.9% vs 5.6%), stroke/systemic embolism (1.5% vs 1.0%), and major bleeding (4.3% vs 3.8%). However, these differences were attenuated after adjustment for clinical characteristics.

There are several limitations to our study. First, the number of Medicaid patients was somewhat limited, and thus we had limited power to detect small differences in clinical outcomes or treatment receipt by insurance groups. Second, the ORBIT‐AF registry is voluntary, and Medicaid patients in our study may not be representative of all Medicaid patients with AF. Consistent with past work,^{19, 20} we classified all patients with Medicaid insurance as “Medicaid,” regardless of additional forms of insurance. Because we were unable to further stratify the Medicaid population by other forms of insurance attributed to small numbers of patients in each strata, it is possible that treatment patterns differ within the Medicaid population by those who had additional insurance versus those who did not. Third, we did not have information on other socioeconomic variables that may influence treatment decisions, such as median household income. Residual measured and unmeasured confounding variables may have influenced these findings. Fourth, this study was conducted during the earlier period of NOAC approval (2009–2010). Because uptake of NOAC is a dynamic phenomenon, recent treatment patterns may differ. Additionally, because the ORBIT‐AF follow‐up period was limited to 2 years, it is possible that different results would have been observed with longer follow‐up. Finally, we did not have data on actual prescription drug and/or INR monitoring costs, which may differ by geographical region and could influence choice of OAC therapy.