Bivariate linear regression analyses assessed the relationship between each predictor and each dimension of fatigue, with general fatigue as the primary outcome. Given our interest in identifying predictors of different dimensions of fatigue (rather than characterizing fatigue groups or cases), fatigue was treated as a continuous variable in all analyses. Variables that were associated with an MFSI-SF scale with bivariate p<0.10 were included in a multivariable model for that scale. Multivariable linear regression models were fit using multiple imputation (20 imputations generated using chained equations)43 to handle missing values, and coefficient estimates and standard errors were obtained using Rubin’s rules. Multivariable models controlled for time since diagnosis. The percent of variance explained, as measured by R2, was obtained by averaging the R2 values from the 20 imputation analyses.44 Analyses were conducted in Stata version 13.1.
The target sample size was based on the prevalence of key predictor variables and the magnitude of the hypothesized association with fatigue. Power analyses determined that 240–280 patients would be required to detect effect sizes of 0.35–0.45.
The target sample size was based on the prevalence of key predictor variables and the magnitude of the hypothesized association with fatigue. Power analyses determined that 240–280 patients would be required to detect effect sizes of 0.35–0.45.