The associations between fish intake and ECG parameters were evaluated using linear regression for continuous measures and logistic regression for binary measures. Fish intakes were evaluated as indicator categorical variables. Tests for trend were performed evaluating intake categories as ordinal variables. Associations between dietary n-3 fatty acid intake and ECG measures were evaluated by fitting restricted natural cubic splines (27- 28) and plotting the fitted smooth curve. This method uses piecewise third-order polynomials at equally spaced quantiles to model the best non-linear fit of a relationship, constrained to be linear beyond the extreme knots and to be everywhere twice continuously differentiable. Potential threshold effects were evaluated by adding a multiplicative interaction term (threshold indicator variable times continuous variable), the significance of which was evaluated using a likelihood-ratio test. Not all ECG measures were obtainable on every participant due to underlying rhythm, structural, or artifactual irregularities. Therefore, for each measure, individuals with missing values (1.3% to 5.9% of participants, depending on the measure) were excluded from that analysis. To minimize potential confounding, covariates were included based on clinical relevance as factors that may influence both exposures and outcomes, previously published associations, or associations with exposures/outcomes in the current data set. The final model was adjusted for age, gender, race, education, smoking, body mass index, diabetes mellitus, CHD, physical activity, and intakes of beef or pork, fruits, vegetables, fried fish, alcohol, and total calories. For parsimony in model construction, other covariates that did not materially alter the relations between fish consumption and the outcome measures were excluded from the final model, including enrollment site, annual income, treated hypertension, exercise intensity, use of aspirin, beta-blockers, calcium-channel blockers, class 1A or class III antiarrhythmics, tricyclic antidepressants, lipid-lowering medication, and estrogen, and estimated intake of total fat, saturated fat, linolenic acid, carbohydrates, protein, fiber, and wine. Few participants (3.7%) were taking fish oil supplements, and adjustment for fish oil use or exclusion of these participants had little effect on results. Missing covariate values (<4% for dietary covariates; <1% for other covariates) were imputed using age, race, gender, diabetes, and prevalent cardiovascular disease; analyses using the population median or excluding missing data were not appreciably different. Potential effect modification was assessed in pre-specified subgroups for treated hypertension and prevalent CHD using stratified analyses, the significance of which was evaluated using likelihood-ratio testing with multiplicative interaction terms (exposure times covariate). All p values were 2-tailed (alpha = 0.05). Analyses were performed using Stata 8.2 (Stata Corp., College Station, Texas).