Development and validation of a diagnostic model for early differentiation of sepsis and non-infectious SIRS in critically ill children - a data-driven approach using machine-learning algorithms

Table 2 Variables selected for the diagnostic model in the training dataset and their importance

Variable	Variable importance measure^a
Length of PICU stay until onset of non-infectious SIRS/sepsis	0.031
Interleukin-6	0.017
Platelet count	0.010
Procalcitonin	0.008
Cumulative sepsis or non-infectious SIRS episodes (n)	0.007
Core temperature	0.005
C-reactive protein	0.005
Central venous catheter	0.004

^aVariable importance measures are a proxy for the importance of the variable for correct outcome prediction, but have not the same meaning as classic influence measures based on distributional statistics (like effect sizes (e.g. Odds Ratios) or p values)

ISSN: 1471-2431