Table 3.

Performance metrics of machine learning models and physician’s prediction.

	Accuracy	Sensitivity	Specificity	PPV	NPV
Logistic regression	0.542 (0.400–0.683)	0.417 (0.277–0.556)	0.667 (0.533–0.800)	0.556 (0.415–0.696)	0.533 (0.392–0.674)
Random forest	0.667 (0.533–0.800)	0.542 (0.400–0.683)	0.792 (0.677–0.907)	0.722 (0.595–0.849)	0.633 (0.497–0.770)
Gradient boosting	0.708* (0.580–0.837)	0.708 (0.580–0.837)	0.708 (0.580–0.837)	0.708 (0.580–0.837)	0.708 (0.580–0.837)
Support vector machine	0.667 (0.533–0.800)	0.708 (0.580–0.837)	0.625 (0.488–0.762)	0.654 (0.519–0.788)	0.682 (0.550–0.814)
Physician’s prediction	0.522 (0.380–0.663)	0.238 (0.117–0.358)	0.795 (0.681–0.909)	0.528 (0.387–0.669)	0.520 (0.378–0.661)

PPV positive predictive value, NPV negative prediction value.

*p < 0.05 compared with logistic regression or physician’s prediction.

The values were presented as mean (95% confidence interval).