Data Mining Validation of Fluconazole Breakpoints Established by the European Committee on Antimicrobial Susceptibility Testing

DOI: 10.1128/AAC.00081-09

Article Figures & Data

Open in new tab
Download powerpoint
FIG. 1.
(A) CART showing values for dose/MIC versus outcome for patients with OPC; (B) CART for both datasets (OPC and candidemia).

TABLE 1.

Sensitivity, specificity, false-positive rate, area under the ROC curve, and MCC for MIC values as a predictor of failure

Disease(s)	Classifier	MIC for failures (mg/liter)	Sensitivity (%)	Specificity (%)	False-positive rate (%)	Area under ROC curve	MCC
OPC	J48	>2	97	72	28	0.81	0.74
	CART	>4	96	83	17	0.89	0.81
	OneR	>4	97	72	28	0.84	0.74
	Naïve Bayes	>2	99	74	26	0.94	0.79
	Simple Logistic	>4	95	77	23	0.95	0.75
Candidemia	J48	NC^a
	CART	>4	31	96	4	0.63	0.31
	OneR	NC
	Naïve Bayes	>2	31	96	4	0.48	0.31
	Simple Logistic	NC
All	J48	>4	87	92	8	0.86	0.80
	CART	>4	87	92	8	0.89	0.80
	OneR	>4	87	92	8	0.89	0.80
	Naïve Bayes	>2	91	87	13	0.91	0.77
	Simple Logistic	>2	91	87	13	0.91	0.77

↵ a NC, not calculated (the classifier did not find a value for splitting the populations of successes and failures).

TABLE 2.

Sensitivity, specificity, false-positive rate, area under the ROC curve, and MCC for dose/MIC values as a predictor of treatment success

Disease(s)	Classifier	Dose/MIC for success	Sensitivity (%)	Specificity (%)	False-positive rate (%)	Area under ROC curve	MCC
OPC	J48	>50.0^a	93	72	28	0.95	0.68
	CART	>75.0^a	84	100	0	0.98	0.80
	OneR	>37.5	97	81	19	0.89	0.81
	Naïve Bayes	>41.7	97	81	19	0.85	0.81
	Simple Logistic	>38.8	98	72	28	0.97	0.76
Candidemia	J48	NC^b
	CART	>75.0	31	96	4	0.63	0.34
	OneR	NC
	Naïve Bayes	>1,578.0	0	100	0	0.56	0.0
	Simple Logistic	NC
All	J48	>50.0	87	90	10	0.87	0.77
	CART	>75.0	91	90	10	0.90	0.80
	OneR	>37.5	86	93	7	0.90	0.80
	Naïve Bayes	>674.0	91	59	41	0.86	0.50
	Simple Logistic	NC

↵ a The J48 and CART classifiers produced complex trees for dose/MIC versus outcome (Fig. 1). The initial values splitting the tree for successes and failures were 12.5 and 18.75, respectively. Therefore, the table shows the dose/MIC values predicting the absence of failures.
↵ b NC, not calculated (the classifier did not find a value for splitting the populations of successes and failures).