Methods

Details of REGARDS have been published previously. Briefly, REGARDS was designed to identify causes of regional and racial disparities in stroke mortality. The study population over sampled blacks and persons residing in the stroke belt (North Carolina, South Carolina, Georgia, Alabama, Mississippi, Tennessee, Arkansas, and Louisiana) from January 2003 to October 2007. A total of 30,239 participants were recruited from a commercially available list of residents using postal mailings and telephone data. Demographic information and medical histories were obtained using a computer-assisted telephone interview (CATI) system that was conducted by trained interviewers. Additionally, a brief in-home physical examination was performed 3 to 4 weeks after the telephone interview. During the in-home visit, trained staff collected information regarding medications, blood and urine samples, and a resting electrocardiogram.

For the purpose of this analysis, participants were excluded with data anomalies (n = 56), missing follow-up data (n = 490), missing AF data (n = 691), and missing baseline characteristics (n = 4,885). A total of 24,117 (mean age 65 ± 9.3 years; 55% women; 38% black) participants were included in the final analysis.

Fall history was self-reported during the CATI surveys. Consistent with recent guidelines, subjects were classified as having a positive fall history if they reported ≥2 falls within the year before the CATI survey. AF was identified in study participants from baseline electrocardiogram data and also by self-reported history of a physician diagnosis during the CATI surveys. The electrocardiograms were read and coded at a central reading center (Epidemiological Cardiology Research Center, Wake Forest School of Medicine, Winston-Salem, North Carolina) by electrocardiographers who were blind to other REGARDS data. Self-reported AF was defined as an affirmative response to the following question: “Has a physician or a health professional ever told you that you had atrial fibrillation?”

All-cause mortality was assessed by semiannual telephone follow-up and contact with proxies provided by the participant on recruitment. Subsequently, the date of death was confirmed through linkage with the Social Security and National Death Index or by death certificates. Mortality data were complete through March 31, 2014.

Age, gender, race, income, education, exercise habits, alcohol use, and smoking status were self-reported. Annual household income was dichotomized at $20,000. Education was categorized into “high school or less” or “some college or more.” Cognition was assessed over the telephone using the 6-item screener, which evaluates global cognitive function. Scores range from 0 to 6, with lower scores indicating worse cognition, and cognitive impairment was defined as a score ≤4. The presence of depressive symptoms was defined as a score of ≥4 on the 4-item Center for Epidemiologic Studies Depression Scale. Impaired mobility was assessed using the physical functioning scale of the Short-Form 12-Item Health Survey. Low scores are typical of someone who experiences many limitations in physical activities, including bathing or dressing, whereas high scores represent someone who is able to perform these types of activities without limitations. Scores below the age- and sex-specific twenty-fifth percentile were used to define impaired mobility. Exercise was dichotomized at ≥4 times per week and <4 times per week. Smoking was defined as ever (e.g., current and former) or never smoker. Alcohol use was classified by the number of drinks per week using the following criteria: none, moderate (1 to 2 drinks/day for men and 1 drink/day for women), and heavy (>2 drinks/day for men and >1 drink/day for women). Fasting blood samples were obtained and assayed for serum glucose, total cholesterol, and high-density lipoprotein (HDL) cholesterol. Diabetes was defined as a fasting glucose level ≥126 mg/dl (or a nonfasting glucose ≥200 mg/dl in those failing to fast) or self-reported diabetes medication use. The current use of aspirin and antihypertensive medications was self-reported. The use of warfarin and benzodiazepines was ascertained during the in-home visit by pill bottle review. After the participant rested for 5 minutes in a seated position, blood pressure was measured using a sphygmomanometer. Two values were obtained following a standardized protocol and averaged. Using baseline electrocardiogram data, left ventricular hypertrophy was defined by the Sokolow–Lyon Criteria. Coronary heart disease was ascertained by self-reported history of myocardial infarction, coronary artery bypass grafting, coronary angioplasty or stenting, or if evidence of previous myocardial infarction was present on the baseline electrocardiogram. Baseline stroke was ascertained by participant self-reported history. Cardiovascular disease was the composite of baseline coronary heart disease and stroke.

Categorical variables were reported as frequency and percentage, whereas continuous variables were reported as mean ± standard deviation. Statistical significance for categorical variables was tested using the chi-square method and the Wilcoxon rank sum procedure for continuous variables. Logistic regression was used to compute odds ratios (OR) and 95% confidence intervals (CI) for the association between AF and fall history at baseline. Multivariate models were adjusted for factors known to influence falls: Model 1 included age, sex, and race; Model 2 included Model 1 covariates plus body mass index, cognitive impairment, mobility impairment, alcohol consumption, exercise habits, diabetes, antihypertensive medications, and benzodiazepine use. Subgroup analyses were performed by age (dichotomized at 65 years), sex (male vs female), race (black vs white), cognitive impairment, mobility impairment, and benzodiazepine use using a stratification technique and comparing models with and without interaction terms. We also examined the associations between falls, AF, and all-cause mortality using the following groups: no AF + no falls (reference group), no AF + falls, AF + no falls, and AF + falls. Incidence rates per 1,000 person-years were calculated for each group. Kaplan–Meier estimates were used to compute the survival probability for each category and the differences in estimates were compared using the log-rank procedure. Follow-up time was defined as the time from the in-home visit until death, loss to follow-up, or the end of follow-up (March 31, 2014). Cox regression was used to compute hazard ratios (HR) and 95% CI for the association between the aforementioned categories and all-cause mortality. Multivariate models were adjusted using the following models: Model 1 adjusted for age, sex, race, education, income, and geographic region; Model 2 included covariates in Model 1 with the addition of systolic blood pressure, HDL cholesterol, total cholesterol, body mass index, diabetes, antihypertensive medications, warfarin, lipid-lowering therapies, left ventricular hypertrophy, and cardiovascular disease. A sensitivity analysis was performed to determine if participants with AF and falls have an increased risk of mortality compared with those who have AF alone and also in those with falls alone. Statistical significance for all comparisons including interactions was defined as p <0.05. SAS version 9.3 (Cary, North Carolina) was used for all analyses.