The GIRAF trial was a 24-month, randomized, parallel-group, controlled, open-label, hypothesis generating trial, to compare dabigatran with warfarin in patients with AF or atrial flutter that was conducted in Sao Paulo, Brazil. The study protocol was approved by the local ethics committee and complies with ethics principles from the Declaration of Helsinki and International Conference on Harmonization Good Clinical.
GIRAF trial is an investigator-initiated research, partially funded by Boehringer Ingelheim do Brasil, which also provided dabigatran. The sponsor had no role in study design, trial execution, data analysis, writing/reviewing the manuscript, or in the submission for publication. Clinical Trial Registration: NCT01994265 (URL: www.clinical.trials.gov)
Patients who were being followed at six centers in Sao Paulo (including a geriatric care unit, secondary and tertiary care cardiology hospitals), were invited to participate in the trial, but all the study procedures, including the final screening process, randomization, clinical and neurologic follow-up, and endpoints assessment, were performed at one site (Instituto do Coracao, HCFMUSP, Sao Paulo). Eligible patients were 70 years or older, had a history of AF or atrial flutter documented by a conventional 12-lead electrocardiogram (ECG) or by an ECG strip with duration of 30 seconds or longer, and had a CHA2DS2-VASc score of 2 or higher. Key exclusion criteria were illiteracy or less than 4 years of education, severe valvular heart disease (defined as any of the following anatomically severe valvular heart disease, per echocardiogram with compatible physical findings and cardiac auscultation: aortic stenosis/regurgitation, mitral regurgitation/stenosis, pulmonary regurgitation/stenosis or tricuspid regurgitation/stenosis), diagnosis of dementia (based on clinical judgment by the neurologist and on MMSE scores below education-adjusted norms for the Brazilian population), previous stroke or transient ischemic attack, severe liver disease, chronic kidney disease grade KDIGO 4 or worse (estimated glomerular filtration rate < 30 ml/min/1.73 m2), and major contraindications to oral anticoagulation. Full details of inclusion and exclusion criteria are available in the supplementary appendix.
After eligible patients provided informed consent, they were randomized 1:1 via a randomization program using the Research Electronic Data Capture (REDCap) system, to receive either open label dabigatran 150 mg or 110 mg twice daily (110 mg dose for patients ≥ 80 years or with an eGFR between 30 and 50 mL/min/1.73m2) or warfarin once daily titrated to achieve an international normalized ratio (INR) of 2.0 to 3.0.
Up to 15 days after randomization, all patients went through baseline cognitive evaluation. For patients who were using an oral anticoagulant before randomization other than its group assignment, switching to dabigatran was performed according to current guidelines . For switching from dabigatran (or other non-vitamin K oral anticoagulant), warfarin was started according to the creatinine clearance: if ≥ 50 mL/min, 3 days before discontinuing non-vitamin K oral anticoagulant C, and 2 days before discontinuing it if the creatinine clearance was between 30 and 50 mL/min.
A pre-specified, comprehensive, and thorough cognitive evaluation for different cognitive domains was performed at baseline and at 24 months, based on the recommendations of the National Institute of Neurological Disorders and Stroke-Canadian Stroke Network Vascular Cognitive Impairment Harmonization Standards . The Mini-Mental State Examination (MMSE) and the Montreal Cognitive Assessment (MoCA) were administered as brief measures of global cognitive functioning. In addition, participants were submitted to a neuropsychological test battery (NTB), including the following tests: Trail-Making tests A and B, short form (15 items) of the Boston naming test (BNT), clock drawing test (CDT), digit symbol substitution test (DSST), phonemic verbal fluency test (FAS), semantic verbal fluency test (SVF; animals/minute), and the Figure Memory Test (including immediate, learning, and delayed recall). Participants also underwent computer-generated neuropsychological tests (CGNT), which evaluated simple reaction time and sustained, selective and divided visual attention, with measures of accuracy (i.e., percentage of correct responses) and reaction time (in milliseconds). A detailed explanation of the CGNT can be found elsewhere . All the tests have been used previously in Brazilian Portuguese versions. Cognitive evaluations lasted approximately 90 min and were performed by two neurologists, blinded to group assignments, in separate visit days from clinical consultations. Details regarding the through cognitive evaluation are provided in the supplementary appendix.
In patients randomized to open-label warfarin, INR was measured weekly until the INR goal, then bi-weekly and monthly if the drug dosing was stable and the INR remained within target range (2.0 to 3.0). The time that the INR was within the therapeutic range during the trial was calculated with the use of the method of Rosendaal et al. . Clinical consultations were performed every 3 months for both groups.
Primary outcomes were changes in cognitive performance at 24 months from baseline, measured with MoCA, MMSE, NTB, and CGNT scores as each test analyzes specific cognitive domains. Importantly, despite analyzing different domains of cerebral performance, all tests analyze cognitive function, that represents the main outcome of our study. The NTB and CGNT (accuracy and reaction time measures) scores were calculated as composites Z-scores, by averaging individual tests’ Z-scores weighted according to the number of available tests per patient. Prior to calculation, all components (tests) were standardized to indicate a better performance with higher scores (e.g., by using the negative of reaction time for the CGNT components). The minimum necessary number of components for calculating a patient’s composite score was set to six for the NTB score and seven for the CGNT. The respective value of the composite score was considered missing if the minimum number components condition was not met. Exploratory outcomes, based on post hoc analyses, were changes in cognitive domain scores for executive functioning, attention, language, and memory at 24 months in comparison to baseline. The executive functioning domain included the CDT, trails A and B. The attention domain included DSST and all CGNT tests. The language domain included the BNT, FAS, and SVF tests. The memory domain included the Figure Memory Test. The minimum necessary number of components for calculating the composite score was two for the executive functioning, language, and memory domains, and seven for the attention domain; a missing value was assigned otherwise. For all tests, cognitive decline was defined as any decline in Z-scores over time. Additional methods that were performed for neuroimaging for the diagnosis of silent stroke and biomarker assessments are described in the supplementary appendix.
On the basis of a post-hoc analysis of two randomized controlled trials  and on clinical practice expertise of the authors, we estimated a mean drop of 2 points in the MMSE score after 24 months with a standard deviation of 2 points. Assuming a 10% dropout rate and similar between group differences at 24 months for all primary outcomes, we calculated that a sample size of 200 patients would provide our study a 80% power to detect a 50% difference of change in cognitive scores (measured by any of the primary outcomes) in patients treated with dabigatran compared to warfarin. These estimate were later further supported by a study  that estimated a mean 0.2 drop in the NTB Z score with standard deviation of 0.5.
Primary analysis was conducted according to the modified intention-to-treat (mITT) population, including all patients who underwent both baseline and 24-month cognitive evaluations, censoring for patients who had stroke or other cerebrovascular events throughout the study. Additional sensitivity analyses were also performed to test the consistency of our findings: first using a per-protocol analysis on the mITT population (excluding patients that switched or stopped their oral anticoagulation during the 24-month period and including all randomized patients that underwent the first cognitive evaluation) and second using regression-based multiple imputation to estimate missing values for at 24-month evaluations. A linear regression was carried out for each primary outcome with treatment (D or W, 0/1 coded), age (years), education (log of years), and baseline raw score as covariates, with no interaction factors. After individual analyses of the relationship of covariates and dependent variables, we found only weak linear relationships. Additional analyses of the residuals of the linear regressions disclosed no major discrepancies to the standard assumptions. We report the results as least-square mean changes from baseline for each group (higher scores indicate better cognitive performance) and as the difference between-groups, at baseline and 24 months, with 95% confidence intervals. Cohen’s d standardized size effects are also reported based on the mean treatment difference between groups and residual standard deviation. The confidence intervals and P-values reported refer to a two-sided alpha of 0.05 with no correction for multiple hypothesis testing. To account for the increased risk of a type 1 error in the multiple comparisons of primary endpoints, adjusted P-values were also computed using Hommel’s method and reported in the “Results” section. All statistical analysis were performed using the R software, version 4.1.2 (R Foundation for Statistical Computing), and graphics were elaborated using GraphPad Prism version 9.3.0 for Windows, GraphPad Software, San Diego, California USA, www.graphpad.com.