Data were collected from an electronic medical record of all hospitalized patients diagnosed with AF and/or atrial flutter at a private tertiary institution in Salvador, Bahia, from May 23, 2011 (date of approval of dabigatran in Brazil) to June 30, 2016. Patients were screened from ICDs I48 and R00, and only those with documented AF or atrial flutter were recruited.

Retrospective observational study based on electronic chart review, with annual cross-sections for five consecutive years (trends study).

Inclusion criteria were: age ≥ 18 years, AF diagnosis and/or atrial flutter confirmed by ECG and/or Holter. Patients without electronic prescription of discharge were excluded, as this was the source of the data regarding the use of anticoagulant. On readmissions, the most recent admission was chosen for analysis.

Demographic and anthropometric data were collected, in addition to the cardiovascular risk factors traditionally related to AF (systemic arterial hypertension [SAH], diabetes, valve disease, myocardial infarction, heart failure, history of bleeding, medications). The risk scores for stroke and bleeding were CHA₂DS₂VASc and HASBLED, validated in international studies.12^-14 AF was classified according to the II Brazilian Guidelines for Atrial Fibrillation of the Brazilian Society of Cardiology.15 When there was insufficient information for classification, AF was considered of indefinite duration. Prior AF was defined as the identification of episodes of this arrhythmia prior to the reference admission during review of medical records.

Echocardiographic data were collected from the most recent test, respecting the period of up to 1 year before admission. Valvular heart disease was defined as the presence of any moderate or severe mitral or aortic lesion. The presence of a valve prosthesis was defined by echocardiogram or clinical history.

Electronic prescription of discharge was used to collect information on anticoagulants, antiplatelets and their doses, and other drugs with potential for drug interaction.

All information regarding hemorrhagic events was collected, but only major bleeding - intracranial hemorrhage, need for blood products or corrective surgical treatment - were considered for analysis.

The following were considered absolute contraindications for oral anticoagulants: active bleeding, severe hemorrhagic diathesis, thrombocytopenia < 50,000, invasive surgery or procedure to be done, major trauma, hemorrhagic stroke, intracranial or spinal tumor, spinal anesthesia, uncontrolled SAH. Relative contraindications included: end-stage neoplasia, active peptic ulcer, advanced dementia (without a caregiver), alcoholism and frequent falls (more than three per year). Patients with zero CHA₂DS₂-VASc were not considered candidates for anticoagulation.

Inappropriate use of DOAC16 was defined as follows: prescriptions containing drugs with strong drug interaction potential; dose inconsistent with the dosage recommended by the manufacturer, considering patient’s age, weight and renal function; presence of absolute contraindications (mechanical cardiac valve prosthesis or moderate to severe mitral stenosis).

Continuous variables were summarized by mean and standard deviation or median and interquartile distance, as indicated by the frequency distribution. Comparisons of quantitative variables between 2 groups were done using Student’s t test for independent samples and Mann-Whitney. Categorical variables were summarized using simple and relative frequencies, compared between groups using the chi-square test. For analysis of correlation between the HASBLED and CHA₂DS₂-VASc scores, the Spearman’s technique was used.

To identify predictors of anticoagulation and anticoagulant type, we used binary logistic regression. Firstly, we made a selection of variables using univariate logistic regression. The variables associated with anticoagulation and use of DOAC with p value < 0.05 in the univariate analyses were later included in multivariate logistic regression models using the backward stepwise method. The variables that reached p < 0.05 in the final analyses were considered statistically significant.

All analyses were conducted in the Statistical Package for the Social Sciences (SPSS) version 23.

This study was approved in a plenary session at the Ethics Research Committee Prof. Celso Figuerôa, Hospital Santa Izabel on 11-24-2014, according to Resolutions 466/12 and 251/97, protocol number 917.116.

Patients were separated into two groups: patients with and without anticoagulants. The group on anticoagulants was subsequently divided into patients using VKA or DOAC (Table 1). The mean age of the patients was 70 ± 15 years, and 52% were male with mean body mass index (BMI) of 27 ± 6. Regarding the type of AF, 42% had paroxysmal AF, 12% had persistent AF, 27% had permanent AF, 19% with indefinite duration; 10% of the patients had atrial flutter. There was a high prevalence of comorbidities, with 71% of hypertensive patients, 23% of diabetics, 26% with a history of heart failure, 7.4% with a history of myocardial infarction and 18% had ischemic stroke. Of the study population, 25% had moderate or severe valvulopathy, 3.9% had biological valve prosthesis and 1.8% had mechanical valve prosthesis (Table 1).

Mean CHA₂DS₂-VASc risk score was 3.4 ± 2.0 and the HASBLED score was 1.2 ± 1.0, with 11.2% of these having HASBLED 3 and 79.8% having CHA₂DS₂-VASc ≥ 2 (Chart 1).

Chart 1

Considering the total population over a 5-year period, 75% of the patients were discharged on anticoagulants (20% received VKA and 55% received DOACs), 15% were on antiplatelets alone and 10% were not on antithrombotic therapy.

From 2011 to 2015, there was a progressive increase in the use of anticoagulants from 64% to 87%, largely at the expense of an increase in the use of DOACs, which rose from 29% to 70% (a relative increase of 144.8%), and reduced use of VKA and antiplatelets alone in the same period. Average annual incorporation of DOACs was 10.4% (Chart 2). About 25% of the patients were on antiplatelets alone or combined with anticoagulants.

Chart 2

*Patients from the year 2016 were excluded from this analysis, as only data from admissions taking place in the 6 first months of that year were collected.

Most patients who had anticoagulants prescribed on hospital discharge had a history of previous episodes of AF (p = 0.001) and TIA/stroke (p = 0.008), were older (p = 0.056), had lower mean HASBLED (p < 0.001), higher mean weight (p = 0.002) and higher prevalence of SAH (p = 0.052) than non-anticoagulated patients. In the multivariate logistic regression analysis, the following remained as independent predictors of anticoagulation: previous episodes of AF, SAH and HASBLED (p < 0.001) (Table 2).

To identify variables associated with the prescription of a given class of anticoagulants, we built a multiple logistic regression model using DOAC as a dependent variable and paroxysmal AF, CHF, serum creatinine, left ventricular ejection fraction, LA diameter, biological valve prosthesis, moderate valvular heart disease and electrical or chemical CV as independent variables. The variables serum creatinine (p < 0.002), LA diameter (p = 0.003) and biological valve prosthesis (p = 0.007) were inversely associated with the prescription of a DOAC, that is, were predictors of VKA prescription (Table 3).

Of the 93 patients with AF whose discharge prescriptions did not contain anticoagulants, 43 had legitimate reasons for doing so: 14 with zero CHA₂DS₂VASc, 16 with absolute contraindications and 13 with relative contraindications. In the other 50 patients (54%), no reasons were found for the non-prescription of anticoagulants (45 patients) or the reasons on record were inconsistent with the literature (5 patients) - “requested by the attending physician”; “already had an indication of using anticoagulants and the attending physician did not prescribe”; “has bronchiectasis and history of bleeding”; “due to the risks of anticoagulation”; “will have pacemaker implanted” (Table 4).

In our study, history of AF episodes and the presence of SAH individually increased by three-fold the chances of patients receiving anticoagulant prescription at discharge. On the other hand, at every 1 point where HASBLED increased, the chance of anticoagulation was reduced by 50%. These are logical and intuitive predictors of anticoagulation, since the first two increase the perception of thromboembolic risk and the last one increases the risk of bleeding. In the NCDR PINNACLE registry, there were more hypertensives among patients taking anticoagulants than among those who did not use anticoagulants (80% versus 74%; p < 0.001).21

In an analysis of participants from the 2nd and 5th cohorts of the GARFIELD-AF study in the UK, the two main reasons for the physician to avoid using anticoagulant in patients with CHA₂DS₂-VASc ≥ 2 were risk of fall and hemorrhagic event, respectively.22

Regarding the predictors of use of DOACs, we observed that for each 1.0 mg/dl increase in serum creatinine, there was 82% less chance of the patient being discharged on DOAC. For 1 year, Luger et al. analyzed AF patients who had stroke or TIA and concluded that the decision on the use of VKA or DOAC was mainly determined by the patient’s renal function and absence of previous anticoagulant therapy, both reducing the chance of using DOAC.23

In addition to renal function, the presence of a biological valve prosthesis also significantly reduced the chance of DOAC prescription. Although not all studies comparing DOACs with VKA excluded patients with biological prostheses, which do not represent contraindication for the use of DOACs, the unfavorable experience of dabigatran in patients with mechanical prostheses24 may cause some fear of using these anticoagulants in the context of any valve prosthesis.

Our study demonstrated that in 28% of DOAC prescriptions, the doses were inadequate for the patients’ clinical profile. However, as opposed to the findings of the cited studies, the most frequent dosing error was improper reduction of DOAC dose, rather than lack of adjustment. Both the lack of adjustment and improper dose reduction may compromise the efficacy and safety of anticoagulant therapy with DOACs. The randomized cluster study IMPACT-AF showed that multifaceted educational interventions were able to improve the frequency of anticoagulation and reduce the incidence of stroke.25

Limitations of this paper include its retrospective design based on electronic medical records, making it subject to registration bias due to lack of information in medical records. In some cases, for example, it was not possible to assess the suitability of the DOAC dose because we did not have serum creatinine, weight or height information. Some patients were excluded because they did not have electronic prescription for discharge. However, these events were infrequent and did not compromise the results found.

Regarding the possibility of generalizing our findings to other populations, although it was conducted in a single center, the characteristics of the study population are similar to those of major international registries of AF. However, because it was conducted in a private medical center, our data cannot be extrapolated to patients with AF from the public healthcare system.