Table 1.

Sensitivity, specificity, accuracy, and AUC of NSE, ProGRP, and the proposed algorithm for SCLC

Test		SCLC (N)			Sn (%)	Sp (%)	Ac (%)	Cut-off	Mean (SD)	AUC (95% CI)	P
		+	–	Total
NSE	+	23	50	73	95.8	34.2	49.0	17.30 ng/mL	38.27 (37.84) ng/mL	0.847 (0.760-0.935)	< 0.001
	–	1	26	27
	Total	24	76	100
ProGRP	+	18	4	22	75.0	94.7	90.0	85.70 pg/mL	374.59 (1,279.93) pg/mL	0.867 (0.766-0.969)	< 0.001
	–	6	72	78
	Total	24	76	100
Algorithm	+	21	4	25	87.5	94.7	93.0	14.24 ng/mL	-	-	-
	–	3	72	75				83.97 ng/mL
	Total	24	76	100

The percentage of positive results for NSE was 68.06% (49/72) in NSCLC and 31.94% (23/72) in SCLC. The mean±SD of NSE level in SCLC was significantly greater than that in NSCLC (76.1±54.7 vs. 26.3±19.2; P<0.001); this explains the low specificity for NSE. When the 2-by-2 table for estimating the diagnostic accuracy of NSE, ProGRP, and algorithm test results was prepared, the positive test results were categorized into true positives and false positives, and negative results into true negatives and false negatives. Diagnostic accuracy was defined as the fraction of true positives and true negatives derived from all classifications.

Abbreviations: SCLC, small cell lung cancer; NSCLC, non-small cell lung cancer; NSE, neuron-specific enolase; ProGRP, pro-gastrin-releasing peptide; Sn, sensitivity; Sp, specificity; Ac, diagnostic accuracy; SD, standard deviation; AUC, the area under the receiver operating characteristic curve; CI, confidence interval.