Statistical Analysis
Categorical data is summarized as number (percentage), and continuous
data as mean (standard deviation). Significance was determined by
Chi-square test (Fisher exact test) for categorical variables and either
Kruskal-Wallis (non-parametric) or 1-way ANOVA (parametric) testing for
continuous variables. Normality for all continuous variables was
determined using skewness and kurtosis. Initial regression analysis and
ROC curve generation for predictors of blood culture positivity were
performed using stepwise binary logistic regression (ɑ to enter model =
0.05 and ɑ to exit model = 0.1). Machine learning via a classification
and regression tree (CART) analysis was performed to model non-linear
relationships between predictor variables of interest and bacteremia and
to generate an optimal decision tree diagram. The CART model was first
constructed using a training data set (70% of study population data)
and then its efficacy tested via a testing data set (30% of study
population data). Statistical significance was defined as P ≤ 0.05. All
statistical analysis and figures were obtained using
Minitab® Statistical Software version 21.1.0.