Chapter 1 42 Statistical analysis Distributions of all variables were examined. For continuous variables, the means and standard deviations were calculated and histograms were used to check normality. For categorical variables, proportions were calculated.We drew a directed acyclic graph to reduce the required sample size and prevent power issues without missing factors related to the outcome measure and without missing factors required to reduce bias. This method aims to assist in the selection of appropriate variables for the regression analysis, as is recommended by Greenland et al. (1999). Afterwards, multiple linear regression was performed. Directed acyclic graph We visualized our hypothesized relationships between the factors and their associationwith the primary outcomeNFR and secondary outcome LE. To simplify the graph, we examined the correlations between the factors in the graph and removed all negligible associations, defined as correlation coefficients between − 0.3 and + 0.3 (Hinkle et al., 2003). Pearson correlation coefficients were used to examine the correlations between continuous variables, Phi correlation coefficients for dichotomous variables, and Bi-serial correlation coefficients to determine the correlation between a dichotomous and a continuous variable (Akoglu, 2018; Kraemer, 2014). Further simplification was accomplished by following the method of Shrier and Platt (2008), including removal of all factors that were not directly or indirectly related to neither the primary nor the secondary outcome. Multiple imputation Multiple imputation was used to impute factors directly or indirectly related to the primary or secondary outcome (Pedersen et al., 2017).The number of imputations was ten, thus ten imputed datasets were created. The imputation model consisted of all variables included in the conceptual model (Table 1). Linear regression analysis Linear regressionwith a forward stepwise selectionmethod (α = 0.05) was manually performed with all variables directly related to NFR. As a result of the strategy used to select factors for the analysis, the model was unadjusted for other factors. We
RkJQdWJsaXNoZXIy MTk4NDMw