Figure 1.
A decision tree to guide the manual assignment of the modified Bell’s staging criteria to the study subjects.
Table 1.
Demographics of NEC patients by Bell’s staging criteria.
Figure 2.
Automated NEC staging assignment results.
Left: modeling training. Right: blind testing. Bottom: manual versus automated NEC staging assignment comparative analysis. To gauge the impact of different training/testing cohort partition on the statistical learning, we performed a bootstrapping analysis that randomly partitioned the cohorts into 100 different training/testing sets. Results were summarized where median and interquartile range (IQR) values were calculated for each comparative category.
Figure 3.
Clinical variable’s contribution (LD1) to the NEC outcome LDA model.
LDA: Linear discriminant analysis. LD1: first discriminant variable.
Figure 4.
NEC outcome predictive results.
A. ROC AUC analysis. To gauge the impact of different training/testing cohort partition on the statistical learning, we performed a bootstrapping analysis that randomly partitioned the cohorts into 100 different training/testing sets. The distribution of 100 ROC curves, training and testing respectively, are illustrated. B. Use of the NEC outcome prediction metric to risk-stratify NEC subjects into low, intermediate and high risk groups.
Figure 5.
NEC outcome predictive LDA models with reduced number of variables (listed in descending order from right to left in Figure 3 by the absolute value of their weights).
The model performance was gauged by ROC analysis. Vertical dotted line: the model performance deteriorates when the model’s panel size is less than 7 parameters.