Skip to main content
. 2019 Feb 11;9:83. doi: 10.1038/s41398-019-0419-4

Table 2.

A summary of predictive performance in the training and independent test sets of the prediction models with averaged coefficients

Training set (Cologne) Independent test set (Rotterdam)
Schizophrenia patients 60 9
Controls 77 12
(a) Training set Independent test set
Number of features AUC AUC
Model with >0.80 inclusion fraction 6 0.807
Males only 0.851
Females only 0.751
Males only 0.880
Model with relative feature importance >0.90 8 0.821
Males only 0.863
Females only 0.773
Males only 0.917
(b)
Model with >0.80 inclusion fraction 17 with 5 first-order interactions with sex 0.858
Males only 0.890
Females only 0.819
Males only 0.889
Model with relative feature importance >0.90 13 with 3 first-order interactions with sex 0.854
Males only 0.887
Females only 0.818
Males only 0.815

Model selection (a) did not consider first-order interactions with sex and (b) did allow for first-order interactions with sex