Mid-term changes in cognitive functions in patients with atrial fibrillation: a longitudinal analysis of the Swiss-AF cohort

Background Longitudinal association studies of atrial fibrillation (AF) and cognitive functions have shown an unclear role of AF-type and often differ in methodological aspects. We therefore aim to investigate longitudinal changes in cognitive functions in association with AF-type (non-paroxysmal vs. paroxysmal) and comorbidities in the Swiss-AF cohort. Methods Seven cognitive measures were administered up to five times between 2014 and 2022. Age-education standardized scores were calculated and association between longitudinal change in scores and baseline AF-type investigated using linear mixed-effects models. Associations between AF-type and time to cognitive drop, an observed score of at least one standard deviation below individual's age-education standardized cognitive scores at baseline, were studied using Cox proportional hazard models of each cognitive test, censoring patients at their last measurement. Models were adjusted for baseline covariates. Results 2,415 AF patients (mean age 73.2 years; 1,080 paroxysmal, 1,335 non-paroxysmal AF) participated in this Swiss multicenter prospective cohort study. Mean cognitive scores increased longitudinally (median follow-up 3.97 years). Non-paroxysmal AF patients showed smaller longitudinal increases in Digit Symbol Substitution Test (DSST), Cognitive Construct Score (CoCo)and Trail Making Test part B (TMT-B) scores vs. paroxysmal AF patients. Diabetes, history of stroke/TIA and depression were associated with worse performance on all cognitive tests. No differences in time to cognitive drop were observed between AF-types in any cognitive test. Conclusion This study indicated preserved cognitive functioning in AF patients, best explained by practice effects. Smaller practice effects were found in non-paroxysmal AF patients in the DSST, TMT-B and the CoCo and could indicate a marker of subtle cognitive decline. As diabetes, history of stroke/TIA and depression—but not AF-type—were associated with cognitive drop, more attention should be given to risk factors and underlying mechanisms of AF.

. Description of the neurocognitive test battery and all 17 items included in the cognitive assessment in the Swiss-AF study. Test description and items are grouped by test (MoCA, Trail Making Test Part A and B (TMT-A, TMT-B), Semantic Fluency Test (SF), and Digit Symbol Substitution Test (DSST). Information on definition of scores and measurement properties is also provided. The table was adapted according to Springer et al. [19].

Item No
MoCA Items (scoring according to Manual; www.mocatest.org) The Test evaluates visuospatial and executive functions, confrontation naming, memory, attention, language and abstraction [22].

Trail Making Test Part A, (TMT-A) Item
The test measures visual attention and psychomotor speed [23]. Internal consistency has been reported with Cronbach's alpha = .86 to .88 [39].

Semantic Fluency, Animals (SF), Item
The test measures semantic fluency-a combination of semantic memory and executive functions, complementing phonemic fluency within the MoCA [30].

Digit Symbol Substitution Test (DSST), Item
The test assesses information processing speed, visuomotor coordination and attention [25]. DSST high test retest reliability has been reported. This test has high test-retest reliability [40]. Text S1. Description of the age-education standardized cognitive function score The Swiss-AF baseline data were used for standardization. A linear regression model was fit to the observed baseline data, for each subsequent observation the linear predictor was calculated. Finally, Z-scores were calculated by dividing the linear predictor by the residual standard error of the model. This model assumes a linear association between age and cognitive functioning, that was shown to be correct.
The standardization is performed via the following steps: 1. we fit a linear regression model to the observed values at baseline 2. for each subsequent observation (i.e. follow-up measurements) we calculate the linear predictor based on the model 3. to standardize (= calculate a Z-score) we divide the linear predictor by the residual standard error of the model (as fit using baseline values).
For purposes of the modeling, and to obtain meaningful expected values, we use as reference values (where relevant) the mean variable values for age and education level (years) at baseline based on the full Swiss-AF population. Thus, for example, we calculate the age and education adjusted Z-score for DSST using the following formula: where σ is the square-root of the residual variance from the linear model.                 Data are presented as mean (± SD) or counts (percentages). GFR: glomerular filtration rate; min.: minutes; ml: milliliter; TIA: transient ischemic attack; *Basic education: ≤6 years (less than compulsory education curriculum); middle education: 6 to ≤12 years (high school or similar); advanced education: ≥12 years (college or university degree).