PMC

SLIM performance is shown when the models are constrained to a false positive rate (FPR) of either 20% (A) or 40% (B). The true positive rate (TPR) and solid squares, and the FPR (open squares) are given for extractable models of size 5 or size 10, or symptom based features (size 10), as shown on the X axis. The upper and lower 95% confidence intervals are given as error bars in each case. The legend in panel A applies to panel B as well.

The mean values for 10-fold cross validation (10-CV) true positive rate (TPR; sensitivity) and false positive rate (FPR; 1-specificity) are shown for models that were trained with all features (gray), the subset of “extractable” features (dashed), or the subset of “symptom” features (solid). The diagonal dotted line is the reference for chance performance. Student's t-test revealed p < 0.0001 comparing the symptom feature model with either the extractable model or the full feature set; the full set did not differ from the extractable set (p > 0.5).