Legends:
Figure 1: Binary Logistic Regression Receiver Operating
Characteristic (ROC) Curve. ROC curve constructed from binary logistic
regression analyzing temperature 1-hr post-acetaminophen’s ability to
predict bacteremia among this patient population. Area under the curve
(AUC) = 0.70.
Figure 2: CART Optimal Decision Tree Diagram. Characterization
and regression tree (CART) analysis utilizing temperature 1-hr after
acetaminophen, presence of pre-acetaminophen temperature ≥ 39.0°C,
evidence of focal infection and hypotension to generate an optimal tree
diagram for determination of bacteremia. Class 1 = Cohort (Bacteremia)
& Class 2 = Controls (Blood Culture Negative).
Figure 3: CART Receiver Operating Characteristic (ROC) Curve.ROC curve constructed from characterization and regression tree (CART)
analysis. Patient data was split into training (70%) and testing arms
(30%) for machine learning model construction and validation. AUC for
training data set was 0.86 vs 0.71 for the test set.
Figure 4: Relative Variable Importance. Relative variable
importance for the optimal tree diagram generated by characterization
and regression tree analysis. Variable importance measures model
improvement when splits are made on a predictor in the optimal tree
diagram. Relative importance is defined as % improvement with respect
to the top predictor.