Characteristics and outcomes in patients with atrial fibrillation receiving direct oral anticoagulants in off-label doses

Background We evaluated adherence to dosing criteria for patients with atrial fibrillation (AF) taking dabigatran or rivaroxaban and the impact of off-label dosing on thromboembolic and bleeding risk. Methods We used data for a retrospective cohort from a large U.S. health plan for Medicare beneficiaries age > =65 years with AF who initiated dabigatran or rivaroxaban during 2010–2016. Stroke and major bleeding were quantified in patients who were eligible for low dose but received standard dose, and in patients who were eligible for standard dose but received low dose. Results We identified 8035 and 19,712 patients who initiated dabigatran or rivaroxaban, respectively. Overall, 1401 (17.4%) and 7820 (39.7%) patients who received dabigatran and rivaroxaban met criteria for low dose, respectively. Of those, 959 (68.5%) and 3904 (49.9%) received standard dose. In contrast, 1013 (15.3%) and 2551 (21.5%) of patients eligible for standard dose dabigatran and rivaroxaban received low dose. Mean follow-up for patients eligible for low and standard dose dabigatran and rivaroxaban were 13.9, 15.1, 10.1, and 12.3 months, respectively. In unadjusted analyses, patients eligible for low or standard dose dabigatran and rivaroxaban but receiving off-label dose, had no differences in the rates of ischemic stroke. Among patients who met criteria for standard dose direct oral anticoagulants (DOAC), use of low dose was associated with significantly higher risk of any major bleeding (Dabigatran: HR = 1.44; 95% CI 1.14–1.8, P = 0.002, Rivaroxaban HR 1.34, 95% CI 1.11–1.6, P = 0.002) and gastrointestinal bleeding (Dabigatran: HR = 1.48; 95% CI 1.08–2, P = 0.016). In patients who met criteria for low dose DOACs, there was lower risk of major bleeding (Dabigatran: HR = 0.59; 95% CI 0.43–0.8, P < 0.001), gastrointestinal (Rivaroxaban: HR 0.79; 95% CI 0.64–0.98, P = 0.03) and intracranial bleeding (Dabigatran: HR = 0.33; 95% CI 0.12–0.9, P = 0.001) with standard dosing. After propensity matching, use of off-label doses was not associated with stroke, major, gastrointestinal or intracranial bleeding for either dabigatran or rivaroxaban. Conclusions While a significant number of patients receive higher or lower dose of dabigatran and rivaroxaban than recommended, we found no evidence of significant impact on thromboembolic or hemorrhagic outcomes.


Background
Patients with Atrial fibrillation (AF) have a higher risk for stroke or systemic embolism, death and disability [1]. Oral anticoagulants, either vitamin K antagonist (VKA) or direct oral anticoagulants (DOACs) reduce that thromboembolic risk by about two-thirds irrespective of baseline risk [2]. However, the use of anticoagulation is associated with increased risk of bleeding, with intracranial hemorrhage (ICH) being the most serious bleeding complication [3]. Randomized controlled trials (RCT) of DOACs [Dabigatran, Rivaroxaban, Apixaban, and Edoxaban] have demonstrated similar protection against ischemic stroke but lower rates of ICH compared with VKAs [4][5][6][7].
The RCTs of DOACs in AF patients used dose adjustments based on patient characteristics such age, weight, renal function and the use of concomitant medications. A reduced dose of 75 mg twice daily of dabigatran is recommended to decrease bleeding risk in patients with creatinine clearance (CrCl) 15-30 mL/minute, or co-administration of a strong P-glycoprotein [P-gp] inhibitor (e.g., dronedarone) in patients with CrCl 30-50 mL/minute [8,9]. With regards to rivaroxaban, a dose reduction to 15 mg daily is recommended in patients with CrCl 15-50 mL/minute, and concomitant use of a dual P-gp and cytochrome-3A4 [P-gp-Cyp3A4] inhibitor should be avoided to prevent potential increased rivaroxaban concentration [10]. Administration of lower apixaban dose 2.5 mg twice daily is indicated if 2 of the following 3 criteria are met: age > 80 years, weight < than 60 kg, and serum Creatinine > 1.5 mg/dl [11].
Since the use of DOACs became widespread, deviations from the recommended dosing are not infrequent [12,13]. Analysis of 5738 patients treated with DOACs from the ORBITA-AF II registry showed that 9.4% of patients were under-dosed and 3.4% were overdosed. Overdosing was associated with significantly increased all-cause mortality whereas under-dosing was associated with increased cardiovascular hospitalizations [14]. A subsequent analysis of 7925 AF patients treated with DOACs from the same registry showed that 16% of patients were on reduced doses with many of these doses adjustments (57%) not following the recommended doses [15]. Nevertheless, after risk-adjustment, the use of lower-than-recommended dose resulted in similar thromboembolic and bleeding risk compared to appropriately dosed DOAC use [15].
We hypothesized that a sizeable number of DOAC prescriptions do not adhere to the Food and Drug Administration (FDA) dosing criteria and may increase thromboembolic and bleeding events. The purpose of our study was to: 1) examine characteristics and predictors of low dose use among patients who meet FDA criteria for standard dose, or standard dose use among patients who meet FDA criteria for low dose, among patients who initiate dabigatran and rivaroxaban, ii) compare the risk of ischemic stroke and bleeding events in patients receiving off-label low dose or off-label standard dose to patients receiving FDA-recommended doses in a community-based sample of elderly Medicare beneficiaries with AF enrolled in a large U.S. health plan.

Data source
We designed a new user retrospective cohort study using data for Medicare beneficiaries enrolled in a large U.S. health plan with prescription drug coverage. Medical (inpatient visit, outpatient physician visits) and pharmacy claims with detailed prescription fill information from October 1, 2010 through December 31, 2016 were analyzed. In addition, the data also includes laboratory test results (such as serum creatinine) for Medicare beneficiaries enrolled in managed care plans. The study was nonhuman subject research by the University of Iowa institutional review board because it involved analysis of an existing database that was fully de-identified.

Patient population
We analyzed claims for Medicare beneficiaries age > = 65 years with newly diagnosed atrial fibrillation (AF) between 2010 and 2016 (dabigatran was approved by FDA in October 2010 followed by rivaroxaban approval in November 2011). We identified patients who initiated dabigatran 150 mg twice daily (standard dose) or 75 mg twice daily (low dose), or rivaroxaban 20 mg daily (standard dose) or 15 mg daily. We did not include patients receiving apixaban as information to assess dosing criteria such as patient weight was not available in our data, and we did not include patients receiving edoxaban due to relatively low use of this drug during our time frame. Patients were categorized into mutually exclusive groups according to the first DOAC and DOAC dose received.
Patients were excluded if they did not have a diagnosis of AF during the 12 months prior to initiating the DOAC, where AF was defined as International Classification of Diseases-Ninth Revision-Clinical Modification [ICD-9-CM] code 427.31 or ICD-Tenth Revision [ICD-10] code I48.0, I48.1, I48.2, I48.3, I48.4, or I48.91, as primary or secondary diagnosis. Additionally, we excluded patients who: i) were younger than 65 years at the time of diagnosis, ii) had a diagnosis indicating pulmonary embolism or deep vein thrombosis within 8 weeks prior to initiating the DOAC, iii) underwent hip surgery within 6 weeks prior to initiating the DOAC, or iv) were not enrolled in the health care plan for at least 1 year prior to initiating the DOAC.
Patients who initiated dabigatran or rivaroxaban were categorized to approximate dosing criteria, using the most recent estimated glomerular filtration rate (eGFR) available prior to initiating the DOAC as a proxy for creatinine clearance. eGFR was calculated based on Modification of Diet in Renal Disease (MDRD) Study equation or Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation [16,17]. Patients who received dabigatran were deemed to meet criteria for low dose if they had severe renal disease (defined as eGFR < 30 mL/minute/ 1.73 m2) or had moderate renal disease and concurrent use of a p-gp inhibitor (where moderate renal disease was defined as eGFR 30-50 mL/minute/1.73 m2 and p-gp inhibitors included dronedarone, cyclosporine, itraconazole, tacrolimus, ketoconazole). Patients who received rivaroxaban were deemed to meet criteria for low dose if they had eGFR< 50 or concomitant use of a dual P-gp-Cyp3A4 inhibitor (including ketoconazole, fluconazole, itraconazole, cobicistat, conivaptan, indinavir, voriconazole, posaconazole, nefazodone HCL, ritonavir, saquinavir, telithromycin). Patients with no valid GFR for assessing renal function were excluded. Among patients eligible for low doses of dabigatran or rivaroxaban, 87.4 and 86.5% had available eGFR whereas among those eligible for standard dose dabigatran or rivaroxaban, 54.9 and 56% had available eGFR respectively. We performed sensitivity analysis for DOAC dose adjustments based on eGFR only, excluding drug interactions.

Covariates
Data on patient-level characteristics such as demographics, comorbid conditions, concurrent medication use, and prior health services utilization were extracted from health care plan enrollment data and inpatient, outpatient, and physician claims. Comorbid diseases were identified by ICD-9-CM/ICD-10 diagnoses on claims during the 12 months preceding the date of first DOAC fill. We used the Charlson Comorbidity Index to estimate patients' overall comorbidity status [18]. We also identified all conditions in the CHA2DS2-VASc stroke risk score (congestive heart failure diagnosis, female sex, hypertension diagnosis, diabetes, age, prior stroke or transient ischemic attack, and vascular disease diagnosis). History of major bleeding was defined as any prior major bleeding, gastrointestinal bleeding, intracranial hemorrhage, or prior receipt of transfusion. Additional conditions included liver disease, alcohol abuse, obesity, chronic obstructive pulmonary disease, peripheral vascular disease, pulmonary circulation disease, heart valve disease, history of coronary revascularization, history of implantable cardiac device, and prior pulmonary embolism or deep vein thrombosis. We also extracted data on medication use at the time of DOAC initiation (p-glycoprotein inhibitors, cytochrome 3A4 inhibitors, insulin, statins, beta blockers, angiotensin-converting-enzyme (ACE) inhibitors, angiotensin-receptor blockers (ARBs), calcium channel blockers, prescription antiplatelets (e.g., clopidogrel), proton pump inhibitors and non-steroidal anti-inflammatory drugs. (A list of included drugs and  ICD9/ICD10 codes for comorbid conditions is provided  in Additional file 1: Table S1). Medications were considered concomitant if the patient had supply within 90 days from the DOAC prescription. We also identified patients with a history of warfarin use prior to initiating the DOAC.

Endpoints
We selected the following clinical endpoints: i) ischemic stroke, ii) any major bleeding, iii) gastrointestinal bleeding (GI), iv) intracranial bleeding based on the primary ICD-9-CM/ICD-10 diagnosis on inpatient claims for acute care stays (definitions provided in Additional file 1: Table S1). We also examined drug discontinuation, as defined by the date of last fill for the original DOAC and dose. Patients were followed from the date of the initial DOAC prescription until December 31, 2016 or lapse of health plan enrollment (due to death or other reason), or cessation of the initial DOAC dose.

Statistical analysis
Analyses were conducted separately for patients who initiated dabigatran and rivaroxaban, and by dose eligibility. For example, among patients who received dabigatran and met criteria for low dose dabigatran, we compared characteristics and outcomes among patients who received standard dose in contrast to dosing criteria vs. those who received the recommended low dose. Similarly, among patients who received dabigatran and met criteria for standard dose, we compared patients who received low dose vs. those who received the recommended standard dose. We compared demographic characteristics, comorbid diseases, and concurrent medication use among patients on different doses using the chi-square test.
We examined rates of ischemic stroke, any major bleeding, GI hemorrhage, and intracranial hemorrhage per patient-year of follow-up in the full sample and in propensity matched patients. Specifically, we performed 2-way nearest-neighbor propensity-matching to create groups of patients receiving low dose or standard dose that were balanced with respect to observed patient characteristics. Propensity matching was conducted separately for patients qualifying for low or standard dose, and for patients on dabigatran or rivaroxaban (i.e., four separate propensity matched samples). We assessed covariate balance in propensity-matched samples using standardized differences between patients receiving low or standard dose, where differences less than 10% indicate satisfactory balance. Because standardized differences remained greater than 10% for a small number of covariates, we further adjusted for unbalanced covariates if they were related to the outcome of interest using Cox proportional hazards regression models. We then calculated rates of each endpoint per patient year of followup in unmatched and propensity-matched samples. Statistical significance was assessed using Cox proportional hazards regression models that censored for end of follow-up (December 31, 2016), medication cessation, or disenrollment from the managed care plan for any reason. Results were reported as hazard ratios (HR) with 95% confidence intervals (CI) for each drug and recommended dose, where the HR reflects the outcome rate for off-label dosing relative to FDA-recommended dose. Finally, one sensitivity analysis was performed in which drug dosing criteria was based on GFR measurements only, without considering pharmacologic interactions reflected in FDA criteria.
All analyses were conducted using SAS with 2-tailed level of significance set at 0.05.

Patient characteristic
We identified 8035 patients with valid GFR measurements including 6580 on standard dabigatran dose and 1455 patients on low dose (18.1%); 19,712 patients were rivaroxaban including 13,245 on standard dose rivaroxaban and 6467 on low dose (32.8%). We identified significant differences in baseline characteristics between different does of dabigatran and rivaroxaban before propensity matching (Tables 1 and 2). Among dabigatran patients, those taking low dose were more likely to be over 75 years (P < 0.001), female (P = 0.01), have CKD III or more advanced (P < 0.001) and higher burden of comorbidities as indicated by higher Charlson comorbidity index (low dose: 5.52 ± 3.75 vs. high dose: 4.17 ± 3.47, P < 0.001). Among rivaroxaban patients, low dose was more frequently prescribed to patients over 75 years (P < 0.001), females (P = 0.03), patients with CKD III or more advanced (P < 0.001) and patients with higher comorbidity burden (low dose: 5.05 ± 3.65 vs. high dose: 3.71 ± 3.34, P < 0.001).
As shown on Tables 3, 4, 5 and 6, 1401 (17.4%) and 7820 (39.7%) patients who received dabigatran and rivaroxaban met criteria for low dose, respectively. Of those, 959 (68.5%) and 3904 (49.9%) received standard dose of dabigatran and rivaroxaban respectively. In contrast, 1013 (15.3%) and 2551 (21.5%) of patients eligible for standard dose dabigatran and rivaroxaban received low dose. Patients older than 75 years, females, African Americans, and patients with history of major bleeding or heart failure were more likely to receive lower than recommended dose of dabigatran or rivaroxaban (Tables 3, 4, 5 and 6). Conversely, patients eligible for low dose dabigatran or rivaroxaban that received standard dose were more likely younger with lower rates of advanced CKD.

Outcomes Stroke
The absolute event rates and event rates/year for ischemic stroke in each dosing category are presented in Table 7. Before adjustment for patient characteristics or propensity-match analysis, use of low dose dabigatran among patients eligible for standard dose dabigatran did not affect ischemic stroke risk. Among those eligible for standard dose rivaroxaban but receiving low dose, no significantly different risk of ischemic stroke was found. Among patients eligible for low dose dabigatran who received standard dose, we did not identify any relationship to ischemic stroke risk (Table 7). Also, among patients eligible for low dose rivaroxaban, use of standard dose rivaroxaban was not associated with increased risk of ischemic stroke. After propensity matching, we found no difference in risk of ischemic stroke in 732 patients eligible for low dose dabigatran who received low dose compared to 732 matched patients eligible for low dose dabigatran who received standard dose, or among propensity-matched patients eligible for standard dose dabigatran who received either standard dose (n = 1960) or low dose (n = 1960). Similarly, analysis of propensity matched samples of patients eligible for low dose rivaroxaban (5328 on low dose and 5328 on standard dose) or patients eligible for standard dose rivaroxaban (4500 on standard dose and 4500 on low dose) found no significant association of dose to risk of ischemic stroke (Table 7).

Major bleeding
The absolute event rates and event rates/year for bleeding complications in each dosing category are presented in Table 7. Among patients on dabigatran who met criteria for standard dose, use of low dose was associated with significantly higher risk of any major bleeding (HR = 1.44; 95% CI 1.14-1.8, P = 0.002, Table 7), and higher risk of GI bleeding (HR 1.48, 95% CI 1,08-2, P = 0.016) but not intracranial bleeding compared with patients on standard doses of dabigatran. A similar pattern of increased major bleeding risk (HR 1.34, 95% CI 1.11-1.6, P = 0.002) was identified among patients on rivaroxaban who met criteria for standard dose but received low dose, along with a trend towards increased risk of GI bleeding (HR 1.26, 95% CI 0.99-1.6, P = 0.06) but not intracranial bleeding. In patients who met criteria for low dose dabigatran, there was lower risk of major bleeding (HR = 0.59; 95% CI 0.43-0.8, P < 0.001, Table 7) and intracranial bleeding (HR = 0.33; 95% CI 0.12-0.9, P = 0.03, Table 7) but not GI bleeding in patients who received standard compared to low dose dabigatran. Among patients who met criteria for low dose rivaroxaban, there was lower risk of GI bleeding (HR = 0.79; 95% CI 0.64-0.98, P = 0.03, Table 7) without differences in the risk of major, and intracranial bleeding. After controlling for patient characteristics in propensity-matched samples, we did not find any association of off-label use of low dose or standard dose and the risk of any bleeding events for either dabigatran or rivaroxaban.

Sensitivity analysis
We performed sensitivity analysis among patients with dose adjustments based on valid GFR measurements only and not based on pharmacologic interactions. The propensity matched analysis showed that standard dose dabigatran is associated with higher risk of stroke among patients eligible for low dose according to GFR (HR 2.6, 95% CI 1.03-6.7; p = 0.04). The analysis did not suggest any other significant differences in stroke and bleeding risks between off-label and standard dosing of dabigatran and rivaroxaban. The results are presented in Additional file 2: Table S2.

Discussion
The findings of this retrospective cohort analysis of Medicare beneficiaries with AF treated with dabigatran or rivaroxaban between 2010 and 2016 can be summarized as follows: i) among patients on dabigatran or rivaroxaban who met criteria for low dose, the majority received standard dose; ii) among patients on dabigatran or rivaroxaban who met criteria for standard dose, less than one fourth received the low dose; iii) older age, female sex, black race, bleeding history, and heart failure were associated with receipt of lower than recommended dose for patients receiving dabigatran or rivaroxaban; iv) unadjusted analysis suggested that in patients receiving lower dose than recommended, the risk of any major bleeding was increased, likely reflecting higher baseline bleeding risk, while in patients receiving higher dose than recommended, the risk of bleeding was decreased; v) after risk adjusting using multivariable models or propensity-matching, off-label dosing of dabigatran or rivaroxaban was not associated with increased risk of stroke or bleeding compared to recommended dosing. An increased risk of ischemic stroke with standard dose dabigatran was found among patients eligible for low dose based on eGFR only.  The results of our analysis are in accordance with findings of previous studies which demonstrated that a significant part of AF population on DOACs receive an off-label dose [14,15,19]. An updated analysis of the ORBITA-AF II registry from 2013 until 2016, including 7925 AF patients treated with DOACs, showed that 84% received DOACs at standard dose (mainly rivaroxaban and apixaban, only 451 patients on dabigatran), which was consistent with FDA labeling in 96% of cases [ 15]. Reduced DOAC dose was prescribed to 16% of patients, which was consistent with FDA labeling in 43%. In unadjusted analysis, under-dosing was associated with higher rates of all-cause mortality and major bleeding [15]. Nevertheless, after risk-adjustment, the use of lower-thanrecommended dose resulted in similar thromboembolic and bleeding risk compared to appropriately dosed DOAC use [15]. Our cohort included a larger sample size than ORBITA-AF II and focused on dabigatran and rivaroxaban as opposed to apixaban and rivaroxaban in ORBITA-AF II. Another methodological difference is the use of calculated  creatinine clearance with the Cockcroft-Gault formula instead of the MDRD or the CKD-EPI equations for eGFR calculation that we applied. Although in clinical practice, eGFR by MDRD or CKD-EPI is more commonly used than calculated creatinine clearance, discordances in dabigatran and rivaroxaban doses may occur in up to 30% of elderly patients with creatinine clearance < 60 ml/min [ 19]. Despite these methodological differences both our analysis and the previous report from the ORBITA-AF II registry suggest that among patients on dabigatran or rivaroxaban who met criteria for low dose, the majority received standard dose. Moreover, unadjusted analysis suggested that in patients  receiving lower dose than recommended, the risk of any major bleeding was increased. In both analyses, propensity matching did not reveal any significant differences in stroke and bleeding. Yao et al. 19 previously evaluated potential over-and under-dosing of DOACs based only on renal indication for dose reduction using the data for privately insured and Medicare Advantage enrollees. Like our study, Yao et al. found no significant relationship between risk of stroke or bleeding and dose in dabigatran or rivaroxaban-treated patients with renal indication for dose reduction. However, in aggregate analyses of patients taking dabigatran, rivaroxaban or apixaban with renal indication for low dose, patients had significantly higher bleeding risk if they received standard dose. Consistent with our study, Yao et al. also found no statistically significant relationship between dose reduction and risk of stroke or bleeding in the dabigatran-or rivaroxaban-treated patients who did not have a renal indication for low dose. In contrast to our study, Yao et al. evaluated renal indications for dose Abbreviations: ACE Angiotensin converting enzyme, AMI Acute myocardial infarction, COPD Chronic obstructive pulmonary disease, ICD International classification of diseases, NSAIDS Non-steroidal antiinflammatory drugs, P-GP P-glycoprotein, CYP Cytochrome reductions only, and did not consider use of p-gp inhibitors or dual p-gp and CYP3A4 inhibitors in assessing dosing criteria.
In aggregate our findings demonstrate that decisions by healthcare providers about DOAC dosing may be based on patient clinical conditions not reflected in FDA dosing recommendations. Our analysis suggests that patients deemed by providers to be at higher bleeding risk may have received low dose DOACs even though FDA criteria suggest that they qualified for standard dose. It is noteworthy that our unadjusted analyses found higher risk of bleeding in patients who met criteria for standard dose rivaroxaban or dabigatran but received low dose, suggesting that the perception of higher bleeding risk by physicians was warranted, and may point to important patient characteristics not reflected in FDA criteria. Similarly, patients qualifying for low dose may have been prescribed standard doses if providers deemed their bleeding risk to be low.
DOAC-specific factors should predominantly affect dosing decisions. Renal function is the main indicator for low dose dabigatran and rivaroxaban. Nearly~80% of ingested dabigatran is metabolized by the kidney, while~30% of rivaroxaban is metabolized by the kidney. With chronic kidney disease, the half-lives of these medications are extended, leading to potentially high plasma concentrations of the medications and increased bleeding risk [20,21]. DOAC dose adjustments based on renal function therefore reflect the increased bleeding risk in patients with compromised renal function. Several drug interactions also affect dosing of DOACs. Dabigatran is a substrate for P-glycoprotein. Concomitant use of dabigatran with P-gp inducers such as rifampin reduces the anticoagulant effect of dabigatran and should be avoided Abbreviations: ACE Angiotensin converting enzyme, AMI Acute myocardial infarction, COPD Chronic obstructive pulmonary disease, ICD International classification of diseases, NSAIDS Non-steroidal antiinflammatory drugs, P-GP P-glycoprotein, CYP Cytochrome whereas use of dabigatran with P-gp inhibitors (eg, ketoconazole, dronedarone) in patients with renal disease may increase the anticoagulant effect, hence dose adjustment is required. Combined P-gp and CYP3A4 inhibitors (ketoconazole, protease inhibitors) increase the anticoagulant effects of rivaroxaban and should not be used concomitantly with rivaroxaban. In our analysis, we found a significant percentage of patients on medications interfering with the metabolism of dabigatran and rivaroxaban. Notably, receipt of standard dose dabigatran among patients qualifying for low dose dabigatran due to eGFR 30-60 ml/min with concomitant use of a p-gp inhibitor was particularly common, suggesting that some providers may not recognize the potential interaction. We also note, however, that our pharmacy data only permit evaluation of prescription fills, thus it is possible that some patients receiving concomitant p-gp inhibitors were instructed not to take them or switched to an alternative drug, in which case they would not meet criteria for low dose dabigatran. Other combinations that increase the risk of bleeding with DOACs are antiplatelet agents and non-steroidal anti-inflammatory drugs (NSAIDs), although no specific dosing adjustments are recommended for patients on these drugs. In our study, we found that use of prescription anti-platelet use among patients who met criteria for standard dose dabigatran or rivaroxaban was associated with a modestly higher use of low dose dabigatran and rivaroxaban. Similarly, use of low dose rivaroxaban was also more frequent among rivaroxaban patients taking NSAIDS.

Limitations
An important limitation of this paper is our inability to measure GFR using the Cockcroft [15,16,22]. The MDRD and CKD-EPI equations use serum creatinine in combination with age, sex, and race to estimate GFR, while CG also uses patient weight. The use of alternative equations for dosing decisions has been the topic of considerable debate. In clinical practice, physicians rarely use the CG-estimated GFR and instead rely on the MDRD or CKD-EPI equations, which are easy to calculate and often automatically reported with serum creatinine laboratory tests. Notably, the National Kidney Disease Education Program (NKDEP) previously indicated that either the MDRD or CG equation may be used for drug dosing decisions [23], while more recently, the National Institute of Diabetes and Digestive and Kidney Diseases suggested that either CKD-EPI or CG equations are appropriate for drug dosing purpose [24]. In contrast, a previous review of FDA-recommended drug dosing showed that the CG equation is historically the most common renal function equation cited in drug dosing recommendations [25]. As noted by Yao et al. [26], it is likely that FDA drug labels historically relied on CG-estimated GFR because this method was available before the MDRD or CKD-EPI equations were developed and widely adopted, and drug dosage recommendations have not caught up to standard clinical practice with respect to assessing renal function. Nevertheless, this poses inconsistencies with clinical practice. While for most patients, GFR estimated by the MDRD or CKD-EPI equations has reasonable concordance with CG-estimated GFR, for older patients and patients with significant comorbidity, malnutrition leading to decreased muscle mass, or morbid obesity, there may be less agreement and dosing of medications based on the former may not be consistent with FDA recommendations [27][28][29]. Schwartz [30] found that use of the MDRD or CKD-EPI equations rather than the CG equation for estimating GFR may fail to identify 20 to 50% of patients for whom reduced dabigatran and rivaroxaban doses are recommended. Thus, our analysis likely underestimates the number of patients who qualify for low dose rivaroxaban and dabigatran if decisions are based on the CG equation, thereby underestimating the proportion of patients who are overdosed, or overestimating the proportion of patients who are underdosed, relative to FDA criteria. However, they do likely reflect dosing decisions based on GFR estimates typically used in clinical practice.
Other potential limitations of this study should also be considered. First, due to the observational nature of the study, it is possible that unmeasured confounders could have affected our results in spite of using propensity matched analysis. Second, our analysis included patients over the age of 65 and the findings require validation in younger patients. Finally, we lacked detailed evidence on AF burden and estimation of thromboembolic risk. Strengths our study are the large sample size, the availability of laboratory results (GFR) for most patients, incorporating concomitant medication use for assessing dosing criteria, and application of risk adjustment methods including propensity-matched analyses.

Conclusion
The purpose of our study was to improve understanding of safety and efficacy of DOACs in AF patients receiving low or standard dosing of these medications that was inconsistent with FDA criteria. The majority of patients qualifying for low dose DOACs received standard doses and a percentage of patients qualifying for standard dosing received low dose. After adjustment for comorbidities the risk of stroke and major bleeding was not affected by use of dose inconsistent with FDA criteria. Further validation of our results is warranted especially in patients at high thromboembolic or bleeding risk.
Additional file 1: Table S1. ICD 9 and 10 codes of comorbidities and outcomes included in the analysis and list of medications used for dose adjustments of dabigatran and rivaroxaban. Description of Data: All ICD 9 and 10 codes included in the analysis.