Background

Atrial fibrillation and flutter (hereafter denoted as AF for brevity) is a common arrhythmia with significant morbidity and mortality, and increases the lifetime risk of stroke more than five-fold compared with a non-AF reference population [1, 2]. Findings from large prospective randomized controlled clinical trials demonstrate that oral anticoagulation (OAC) reduces the risk of AF-associated thromboembolic events [2,3,4,5,6]. The importance of timely OAC prescription in appropriate patients is recommended in guidelines globally [2, 7,8,9]. However, among those at highest risk based on guidelines, “real-life” data show that up to two-thirds of eligible AF outpatients at risk for thromboembolic events are not prescribed OAC as indicated [9,10,11]. Additionally, prior studies have indicated that physician characteristics, such as race, if concordant with the patient, can impact guideline adherence [12,13,14]. Furthermore, a Medicare study of emergency department (ED) patients found that less than 20% of patients with new-onset AF who were eligible for stroke prophylaxis were prescribed an OAC, regardless of stroke risk [15].

With more than a half-million patients with acute AF presenting to U.S. EDs per year, and over 25% of all new AF diagnoses made in the ED, the trajectory of AF care often begins in the ED [16,17,18,19,20,21,22]. With the increased availability of direct OACs (DOACs) and the ease of management associated with their use, the ED is an overlooked opportunity to improve that trajectory of care. Understanding OAC prescriber characteristics, prescribing practices, and time delays to appropriate care in an integrated system, where follow-up is readily available, could help identify opportunities to improve prescribing practices from the ED.

Thus, we sought to examine stroke prophylaxis actions upon and shortly following ED discharge of stroke-prone AF patients and to identify factors associated with delays in prescribing OAC following an ED visit. This retrospective study lays the groundwork for identifying longer-term outcomes (namely stroke and bleeding risk) and whether stroke risk stratification scores apply to the ED patient population.

Methods

Study design

We performed a retrospective cohort study of all adult health plan members with continuous prescription benefits and a primary diagnosis of non-valvular AF who were OAC-naive on ED arrival.

Study setting

The study included 21 community medical centers in Northern California over an 8-year period (January 1, 2010 through December 31, 2017). Northern California geographically comprises the state’s northernmost 48 counties (socioeconomic characteristics available in Appendix 1). Each site had cardiology available around the clock for consultation, outpatient pharmacy, and anticoagulation management service (AMS). AMS is a pharmacist-led, phone-based service that provides patient education and close monitoring of patients on warfarin and DOACs (Appendix 1). They also serve as consultants for clinicians.

Study population

To select patients not on OACs, we excluded patients with a prior OAC fill three months prior to the index visit and those with an OAC on their current medication list. Only patients discharged home from the ED were included. We included AF patients with indications for stroke prophylaxis with high CHA2DS2-VASc scores ≥ 2 and low/intermediate bleeding risk by HAS-BLED (score < 4), as an OAC prescription for this cohort would follow American Heart Association guidelines and minimize the risk of bleeding events for the risk-averse ED clinician [23]. The Kaiser Permanente Northern California Institutional Review Board approved this study and waived the requirement for informed consent.

Outcomes

The primary outcome was stroke prevention action (OAC Action) within 14 days of the index ED visit. An OAC Action was defined a priori as an OAC prescription or AMS consultation (i) by an ED clinician at the time of discharge or (ii) by any clinician within 14 days following ED discharge. We used the filling of the prescription as a proxy for prescribing. OACs prescribed and filled with prescription drug coverage included warfarin, a DOAC (dabigatran was placed on the system’s formulary starting in 2014, followed by apixaban in 2016), warfarin and enoxaparin bridge, and others (i.e., dalteparin, fondaparinux, or enoxaparin only).

Variables

For our cohort, we identified patient demographics (sex, age, race, Hispanic ethnicity) and stroke risk stratification scores by (1) CHA2DS2-VASc: congestive heart failure, hypertension, age, diabetes, prior stroke (including ischemic versus hemorrhagic), vascular disease, and sex, and (2) bleeding risk (defined by HAS-BLED score); and by year of index encounter [24,25,26,27,28]. Additionally, physician characteristics (sex, years in practice as a proxy for experience, race/ethnicity, language, and race concordance with the patient—as self-documented by the physician and patient) and hospital characteristics (number of ED beds and annual ED census, and availability of an outpatient observation unit) were also analyzed. Detailed information of data abstraction method is described in Appendix 2.

Statistical analysis

We generated descriptive statistics and bivariate analyses to characterize the sample and to identify candidate predictors of OAC Action. When comparing categorical measures, we used chi-square tests. For continuous independent variables, we used t-tests and the pooled or Satterthwaite method to determine differences in means between those with and those without appropriate OAC Action. Annual rates of OAC initiation were calculated for eligible discharges.

We used a hierarchical, multivariable risk-adjustment model to evaluate patient, physician, and hospital factors associated with OAC Action. On the patient level, we used multivariable logistic regression to determine the influence of stroke risk classification, CHA2DS2-VASc, on OAC Action, and the influence of other patient, physician, and hospital factors. We used a generalized estimating equation design with a Poisson distribution, identity link, and an independent correlation structure to account for clustering by physician and hospital, as well as repeated measures by a patient via an alternating logistic regression model that allowed for three-level clustering. This generated population-level rates of OAC Action within 14 days. Database management and all analyses were conducted using SAS (v 9.4, SAS Institute, Cary, NC).

Results

Of the 319,961 adult health plan members with at least one ED diagnosis of AF during 2010–2017, 9603 met our eligibility criteria and formed the primary sample for analysis (Fig. 1). The majority of patients (such as those with transient secondary AF) were excluded as AF was not the primary diagnosis, or the patient had gaps in insurance coverage or evidence of current or recent OAC use. Transient secondary AF is new onset of AF occurring secondary to any acute medical condition (ie. peri-procedural, hyperthyroidism, alcohol intoxication, or infection), and typically self-resolves.

Fig. 1
figure 1

Cohort assembly diagram. AF, atrial fibrillation or flutter; ED, emergency department; KP, Kaiser Permanentee

Full size image

The mean age of our cohort was 73.1 ± 11.4 years, of whom 5986 (62.3%) were female, and 7322 (76.3%) were non-Hispanic white. We compared characteristics between those patients who had any OAC Action within 14 days of diagnosis (N = 2577, 26.8%) versus those without OAC Action (N = 7026, 73.2%; Table 1). The mean ages, stroke risk, and bleeding risk scores, were similar between the two groups.

Table 1 Characteristics of atrial fibrillation/flutter ED patients, n = 9603
Full size table

The following patient groups received increased Action compared with others within their category: males more than females, 29.0% vs 25.5% (p < 0.001), ages 75–84 years (n = 871, 32.7%), Asian (n = 200, 30.8%), and CHA2DS2-VASc score 4–5 (n = 953, 30.3%).

Physician and hospital characteristics were also compared. There were more male than female physicians (67.5% vs 32.4%), but they had similar OAC Action. Physicians had similar mean years of experience: 14.8 ± 8.5 years. Language concordance and discordance between physicians and patients for OAC Action were similar. There was a higher proportion of OAC Action among discordant physician–patient race/ethnicity than concordance (29.9% vs 26.7%).

In Table 2, the rate of OAC Action increased over time, from 21.0% in 2010 to 43.6% in 2017 (hospital-level data available in Appendix 3). Breaking down OAC Action, we found that the overall rate of ED prescribing remained low, 2.4% in 2010 to 6.0% in 2017 (Fig. 2).

Table 2 Oral anticoagulation action in the 14 days following an ED encounter
Full size table
Fig. 2
figure 2

Stroke prevention action within 14 days of ED visit

Full size image

After adjusting for patient demographic and clinical variables, patient and physician race/ethnicity and language concordance, clinical years of experience, and hospital-related characteristics were no longer significant as they were in the univariate model. In our parsimonious model (Table 3), females and older patients (> 85 years old) were approximately 4% less likely to receive an OAC Action compared with males and those less than 64 years of age. Patients of Hispanic White ethnicity had higher rates of OAC Action (5.5% higher, all other factors being equal); Black/African American and Asian differences, while still higher than referent non-Hispanic Whites, were non-significant or borderline significant. Patients with a CHA2DS2-VASc risk score of 4–5 were prescribed anticoagulants at rates 4.5% higher than a risk score of 2–3; and there was no difference in the highest risk CHA2DS2-VASc category (risk score of 6–9) compared with the lowest category in this cohort (risk score of 2–3). Asian physicians acted at 4.5% higher rates than other physician races/ethnicities (Table 3).

Table 3 Predictors of OAC action within 14 days of emergency department discharge diagnosis of atrial fibrillation/flutter
Full size table

Discussion

In this 21-center study of ED patients with low/intermediate bleeding risk, high stroke risk AF not anticoagulated on ED arrival, we found that rates of appropriate stroke prevention action increased over the 8-year study period. However, signficant gaps associated with patient sex and age were identified.

Our results demonstrate a significant disparity among women, who were 3.5% less likely to receive appropriate stroke action compared with men. These results are similar to the PINNACLE study that found women at all levels of risk had persistently lower rates of OAC use compared with men [29, 30]. With increased discussions on implicit bias, clinicians are increasingly aware of their own biases. This disparity highlights the need for increased interventions, namely gender bias training aimed at closing the gap to address gender and sex disparity.

The net clinical benefit of OAC may decrease with advancing age, where the quality of life-adjusted years of being on an OAC decreased compared with no treatment after age 92, where hemorrhage, stroke, or death were competing risks [31]. The risk of death from these other causes may be an important influence in declining net clinical benefit from OAC use. Thus, although physicians were less likely to take appropriate action according to current guidelines, physician intuition regarding those of advanced age may be in accordance with newer evidence.

Based on these recent studies, it appears that patients in the highest risk category (CHA2DS2-VASc score 6–9) would be more likely to possess characteristics that bias clinicians against OAC Action. For example, this higher risk stratification could account for age (1 point for age ≥ 65, 2 points for age ≥ 75), hypertension, and/or history of stroke, which potentially represents a greater co-morbidity burden given the significant overlap of these specific factors among HAS-BLED and CHA2DS2-VASc scores. However, the HAS-BLED score serves as a dynamic tool for assessing bleeding risk in patients rather than as a means to exclude high-risk patients from anticoagulation therapy. Its primary function is to highlight that modifiable bleeding risk factors are addressed, such as high blood pressure, and appropriately managed by outpatient clinicians. This approach allows for a more nuanced and patient-centered decision-making process regarding anticoagulation therapy, balancing the benefits of stroke prevention against the potential risks of bleeding complications [32]. Nevertheless, the ED is an acute care setting where close monitoring of reversible risk factors is not often a primary concern. Interesting results stem from the correlation of CHA2DS2-VASc stroke risk score with physician prescribing. CHA2DS2-VASc scores of 4–5 were associated with 4.5% higher rates of OAC prescribing than a reference of CHA2DS2-VASc scores of 2–3; however, the rate of OAC prescribing for patients with a CHA2DS2-VASc score of 6–9 was not different from the lower risk group (score of 2–3). There was no difference in OAC prescribing for the highest risk category CHA2DS2-VASc score of 6–9 compared with the lowest category of CHA2DS2-VASc score. We hypothesize that while ED physicians recognize that higher scores equal a higher risk of stroke, risk–benefit ratios might also be considered. Furthermore, in the setting of age-related frailty and patient-specific goals of care, it is certainly possible that patients in the highest risk category received advanced care involving cardiology consultation, palliative care consultation, and/or subsequent cardiology follow-up, which could explain a lower rate of OAC prescribing by physicians.

Our study also attempted to determine if patient-physician race/ethnicity and language concordance influenced prescribing or referrals; however, no significant correlation was identified. We found that patients of Hispanic ethnicity received the highest rates of OAC Action (6.1% higher rates, all other factors including physician race, being equal). Although patients of Black/African American and Asian race had higher rates of OAC Action than referent non-Hispanic Whites, the differences were non-significant or borderline significant. Specific patient-physician racial/ethnic and/or language concordance could explain this, as some studies have indicated that racial concordance may improve medication adherence and account for higher-quality patient-physician communication [33, 34]. Reassuringly, we found no bias of physicians, regardless of race/ethnicity, towards any racial/ethnic minorities. Additionally, physician years of experience did not influence prescribing. Interestingly, for physician race/ethnicity, Asian physicians, specifically, were 4.8% more likely to take OAC Action compared with any other race/ethnicity, which may be related to culture-specific factors.

Our study also found that the rates of OAC prescribing increased starting in 2013 when dabigatran became available on the market. The trend of increased OAC prescribing correlates well with adding dabigatran to the system’s formulary in 2014, followed by apixaban in 2016. This suggests that ease of prescribing influences rates of prescribing. These findings are important in the US setting, where multiple insurers with varied coverages and high medication co-pays can make it challenging for physicians to identify cost-effective medications to encourage patient adherence. Improved insurer coverage nationwide, as well as approval and availability of generic OAC medications, could significantly increase prescribing and affordability and, thereby, increase patient access to medications and subsequent patient adherence.

Potential strategies to improve OAC prescribing for high-risk patients with AF include reducing patient costs of DOACs, enlisting assistance from AMS, and providing a clinical decision support (CDS) tool to help minimize bias in prescribing. The use of a CDS tool that includes a risk calculator (e.g., for CHA2DS2-VASc) could improve prescribing, as a calculator operates independent of bias (such as ageism, sexism, racism, etc.). Additionally, an integrated healthcare system further reduces the variability often seen in individual hospitals with regard to insurance, follow-up, formulary medications, etc. Thus, the gap in care that we observed is likely to be amplified in other settings and may require further strategies to improve important long-term clinical outcomes.

In 2021, we designed an ED-based CDS tool with an auto-populating CHA2DS2-VASc risk calculator to help physicians identify ED patients with AF who were eligible for OAC initiation at the time of ED discharge [35]. We evaluated the effectiveness of the CDS and its promotion to improve OAC initiation in a pragmatic, stepped-wedge, cluster-randomized trial [36]. The use of CDS during the intervention phase worked as we had hypothesized: no gender difference was observed.

While ED-based CDS tools can assist with disparities, uptake can be challenging for those who remain uncertain whether CHA2DS2-VASc applies to ED populations, especially when professional guidelines for AF care changed in 2023 to include duration of AF as an indicator for risk and treatment. Future research should determine whether CHA2DS2-VASc and HAS-BLED risk scores can be used in acute care populations [7, 37].

Limitations

First, our results are limited to health plan members with prescription coverage benefits; however, our results are concordant with those of other settings. As an integrated health system, however, timely access to primary care physicians is more often available and could result in more prescriptions. Second, although stroke risk categories changed for women during this time period, namely Class I recommendation for OAC consideration was for a score of ≥ 2 in males and ≥ 3 in females, the Class IIa recommendation was for a score of ≥ 1 in males and ≥ 2 in females, our data remain in line with guideline recommendations and minimizes data reclassification. While guidelines state that bleeding risk scores should not be used at the patient level to determine eligibility for OACs, we are limited to aggregate data and were unable to consider benefits and harms on an individual level; we used HAS-BLED as a gross measurement of bleeding risk as an exclusion. Third, our results coincide with the introduction of DOACs and end in 2017 and hence may not reflect the current uptake of these medications. Fourth, we did not examine the impact of ED prescribing on subsequent clinical outcomes, which we intend to evaluate in future research. Fifth, we may have inadvertently included patients with a prior diagnosis of AF by using patients identified by ICD-9 initially and ICD-10 codes as of October 2015. However, even if the diagnosis had been made previously, the indication for OAC is still present, and the low OAC prescribing rate is still an important and relevant finding. Sixth, due to the nature of our KP databases, we used the filling of a prescription as a proxy for prescribing. Thus, physicians may have appropriately prescribed OAC without patients filling their prescriptions. This could lead to an underestimation of prescriptions. However, we would not expect the effect to invalidate the primary finding of a very low OAC treatment rate. Finally, we used structured data to capture co-morbidities. We did not have access to unstructured patient-level clinical data that may have influenced OAC prescribing (e.g., fall risk, ability to adhere to medications, etc.), which may have dissuaded OAC initiation for some patients.

Conclusion

Rates of AF thromboprophylaxis can be improved further by addressing sex and age disparities, as females and those age ≥ 75 (versus 65) were less likely to receive appropriate OAC Action.