Gene expression profile of peripheral blood in colorectal cancer

Table 1 Characteristics of the training and testing sets[1,2] n (%)

	Training set (n = 162)			Testing set (n = 176)			P value
	CRC (n = 55)	Non-CRC (n = 107)	P value	CRC (n = 56)	Non-CRC (n = 120)	P value	Cases	Controls
Age, yr (S.E.)	66.47 (1.50)	68.31 (1.12)	0.335	67.38 (1.83)	69.99 (1.03)	0.216	0.704	0.270
Gender			0.630			0.176	0.387	0.313
Male	32 (58.2)	58 (54.2)		28 (50.0)	73 (60.8)
Female	23 (41.8)	49 (45.8)		28 (50.0)	47 (39.2)
Stage			-			-	0.447	-
I	21 (38.2)	-		15 (26.8)	-
II	10 (18.2)	-		9 (16.1)	-
III	14 (25.5)	-		21 (37.5)	-
IV	10 (18.2)	-		11 (19.6)	-
Tumor site		-	-		-	-	0.286	-
Colon	28 (50.9)			30 (53.6)
Rectum	22 (40.0)			16 (28.6)
Cecum	4 (7.3)			5 (8.9)
Colon+Rectum	1 (1.8)			5 (8.9)

¹Data are given as means (SE) or as the number of cases (%); ²P-values were estimated using the t-test. CRC: Colorectal cancer.

Table 2 Multivariate analysis of colorectal cancer-related molecular markers and the discrimination model based on age, sex, and 15 genes, using the logistic regression model on the training set

			95%CI of OR
	B	OR	Upper	Lower	P value
Sex	0.577	1.780	7.582	0.418	0.435
Age	0.028	1.028	1.083	0.976	0.293
MCM4	0.142	1.152	4.504	0.295	0.838
ZNF264	1.450	4.265	18.208	0.999	0.050
RNF4	-0.550	0.577	5.146	0.065	0.622
GRB2	2.009	7.456	37.131	1.497	0.014
MDM2	1.359	3.892	15.166	0.999	0.050
STAT2	-1.178	0.308	1.466	0.065	0.139
WEE1	1.264	3.540	14.784	0.848	0.083
DUSP6	2.465	11.769	40.330	3.435	1.33E-11
CPEB4	2.045	7.725	27.695	2.155	0.002
MMD	-1.067	0.344	0.865	0.137	0.023
NF1	-1.417	0.243	1.517	0.039	0.130
IRF4	0.057	1.059	3.350	0.335	0.923
EIF2S3	-2.105	0.122	0.718	0.021	0.020
EXT2	-1.933	0.145	1.235	0.017	0.077
POLDIP2	-1.294	0.274	1.515	0.050	0.138

B: Coefficient of logistic regression; OR: Odds ratio; CI: Confidence interval.

Table 3 Discrimination power and receiver operating characteristic analysis of different combinations of colorectal cancer-associated genes in the training set

				95%CI
Genes used for models	AUC	SE	P value	Lower	Upper
DUSP6	0.804	0.038	< 0.001	0.73	0.879
DUSP6, CPEB4	0.855	0.032	< 0.001	0.791	0.919
DUSP6, CPEB4, EIF2S3	0.882	0.032	< 0.001	0.820	0.945
DUSP6, CPEB4, EIF2S3, MDM2	0.895	0.030	< 0.001	0.838	0.953
DUSP6, CPEB4, EIF2S3, MDM2, MMD	0.905	0.028	< 0.001	0.849	0.960

P-values for AUC were estimated using the Z test. ROC: Receiver operating characteristic; AUC: Area under the ROC curve; SE: Standard Error; CI: Confidence interval.

Table 4 Mean expression levels, standard error and statistical power of selected genes between case and control groups in the training and testing sets

	Training set			Testing set
Selected genes	Case (n = 55)	Control (n = 107)	Power	Case (n = 56)	Control (n = 120)	Power
MDM2	-0.4225 (0.08945)	-0.8913 (0.04572)	1	-0.3270 (0.09063)	-0.9209 (0.03618)	1
DUSP6	2.5483 (0.13248)	1.5458 (0.06415)	1	2.0335 (0.12041)	1.7462 (0.06135)	1
CPEB4	1.3413 (0.11016)	0.3932 (0.09799)	1	1.4595 (0.11851)	0.4014 (0.06980)	1
MMD	2.0567 (0.15441)	1.3178 (0.09799)	1	1.7029 (0.15958)	1.4320 (0.07806)	1
EIF2S3	3.4489 (0.07883)	3.6158 (0.05331)	1	3.4311 (0.05937)	3.5620 (0.03815)	1

Values in cells: Mean expression levels (standard error); α-level is 0.05.

Table 5 Performance of the statistical model based on the five-gene profile logistic probabilities for the training set

Logit(P)	Sensitivity	Specificity	PPV	NPV	Accuracy
0.020	99%	16%	2.3%	99.9%	44.2%
0.051	95%	63%	12.1%	99.6%	73.9%
0.178	90%	72%	41.1%	97.1%	78.1%
0.500	78%	92%	82.7%	89.1%	87.0%
0.475	80%	90%	87.8%	83.3%	86.6%
0.685	61%	95%	96.4%	52.9%	83.5%
0.901	25%	99%	99.6%	12.6%	73.9%

Logit(P): Logistic probabilities; PPV: Positive predictive value; NPV: Negative predictive value.

Table 6 Performance of the statistical model on the training, testing sets and external validation dataset from 12 public microarray studies with Logit(P) = 0.5

	Training set	Testing set	External validation
Non-Cancers	107	120	88
True negative	98	110	72
False positive	9	10	16
Colorectal Cancers	55	56	519
False negative	12	19	15
True positive	43	37	504
Total	162	176	607
Sensitivity	78.2%	66.1%	97.1%
Specificity	91.5%	91.7%	81.8%
PPV	82.7%	78.7%	96.9%
NPV	89.1%	85.3%	82.8%
Accuracy	87.0%	83.5%	94.9%

Logit(P): Logistic probabilities; PPV: Positive predictive value; NPV: Negative predictive value.

Table 7 Logistic regression models for 12 pooled microarray data sets as the external validation of colorectal cancer -associated genes from three studies

	Model 1			Model 2			Model 3
	B	S.E.	P value	B	S.E.	P value	B	S.E.	P value
Five selected genes of this study:
MDM2	6.069	1.461	< 0.001
DUSP6	1.360	0.235	< 0.001
CPEB4	-3.177	0.383	< 0.001
MMD	0.335	0.442	0.448
EIF2S3	1.462	0.244	< 0.001
Seven selected genes of Marshall et al[14]
ANXA3				0.559	0.212	0.008
CLEC4D				46.259	9.918	< 0.001
LMNB1				1.883	0.330	< 0.001
PRRG4				-1.284	0.371	0.001
TNFAIP6				1.787	0.377	< 0.001
VNN1				0.207	0.159	0.194
IL2RB				0.269	0.216	0.213
Five selected genes of Han et al[15]
CDA							-0.496	0.090	< 0.001
MGC20553							-1.386	0.197	< 0.001
BANK1							0.565	0.373	0.129
BCNP1							-0.944	1.148	0.411
MS4A1							-1.483	0.457	0.001
Constant	-32.758	6.001	< 0.001	-124.678	25.437	< 0.001	16.601	2.995	< 0.001
H-L		0.460			0.044			0.194
R2		0.853			0.841			0.693
AUC		0.978			0.985			0.957
Accuracy		0.949			0.974			0.939
Specificity		0.818			0.886			0.716
Sensitivity		0.971			0.988			0.977

Model 1: Five selected genes of this study; Model 2: Seven selected genes of Marshall et al[14]; Model 3: Five selected genes of Han et al[15]; B: Logistic regression coefficient beta; SE: Standard error of B; P: P value with statistical significance; H-L: Hosmer and Lemeshow test P value R2: Nagelkerke R Square; AUC: Area under ROC.