A prediction model for xerostomia in locoregionally advanced nasopharyngeal carcinoma patients receiving radical radiotherapy

Table 3 The predictive performance of models for Grade 3 xerostomia

	Sensitivity (95%CI)	Specificity (95%CI)	PPV (95%CI)	NPV (95%CI)	AUC (95%CI)	Accuracy (95%CI)
Training set
RF	0.955 (0.868–1.000)	0.961 (0.937–0.986)	0.700 (0.536–0.864)	0.996 (0.987–1.000)	0.986 (0.972–1.000)	0.961 (0.937–0.985)
XGB	0.864 (0.720–1.000)	0.858 (0.814–0.903)	0.365 (0.235–0.496)	0.985 (0.969–1.000)	0.914 (0.844–0.984)	0.859 (0.816–0.902)
DTC	0.500 (0.291–0.709)	0.991 (0.980–1.000)	0.846 (0.650–1.000)	0.955 (0.928–0.981)	0.746 (0.639–0.853)	0.949 (0.922–0.976)
Testing set
RF	0.333 (0.025–0.641)	0.851 (0.782–0.921)	0.167 (0.000–0.339)	0.935 (0.884–0.985)	0.766 (0.626–0.905)	0.809 (0.736–0.883)
XGB	0.444 (0.120–0.769)	0.792 (0.713–0.871)	0.160 (0.016–0.304)	0.941 (0.891–0.991)	0.661 (0.478–0.843)	0.764 (0.684–0.843)
DTC	0.222 (0.000–0.494)	0.980 (0.953–1.000)	0.500 (0.010–0.990)	0.934 (0.887–0.981)	0.601 (0.457–0.746)	0.918 (0.867–0.969)

RF random forest, DTC decision tree classifier, XGB extreme-gradient boosting, PPV positive predictive value, NPV negative predictive value, AUC area under the curve

ISSN: 1472-6831