Mass spectrometry protein expression profiles in colorectal cancer tissue associated with clinico-pathological features of disease

Table 2 Performance of predictive models for classification of clinico-pathological characteristics in tumour tissue

CHARACTERISTICS	¹Advanced Dukes' stage	Poorly differentiated	Lymph node involvement	Invasiveness	²Disease recurrence
Number of features	5	2	4	9	10
Positive prediction rate	6/12	10/13	5/13	3/7	5/6
Sensitivity	0.500	0.769	0.385	0.429	0.833
³CI	0.223-0.777	0.460-0.938	0.151-0.677	0.118-0.798	0.364-0.991
Positive predictive value	0.750	0.833	0.625	0.750	0.833
CI	0.356-0.955	0.509-0.971	0.259-0.898	0.219-0.986	0.364-0.991
Negative prediction rate	17/19	16/18	15/18	23/24	22/23
Specificity	0.894	0.889	0.833	0.958	0.957
CI	0.654-0.981	0.639-0.981	0.577-0.956	0.768-0.998	0.760-0.998
Negative predictive value	0.739	0.842	0.652	0.852	0.957
CI	0.513-0.889	0.585-0.958	0.428-0.828	0.654-0.951	0.760-0.998
Absolute error	0.258	0.161	0.355	0.161	0.069
⁴ROC error	0.302	0.171	0.391	0.307	0.105
Fisher's exact test	P = 0.020	P = < 0.001	P = 0.133	P = 0.027	P = < 0.001

¹Includes Dukes' C1 and C2; ² Median follow-up time for recurrent disease patients: 33 months; median follow-up time for disease-free patient: 27 months (analysis excludes patients who died through surgical complications - see Table 1); ³CI = 95% confidence interval; ⁴ROC = receiver-operator characteristics
The KNN algorithm [29] was used in 'leave-one-out' cross-validation prediction with the number of features (marker peaks) specified. Marker peaks were selected using a t-test statistic except for lymph node involvement and invasiveness characteristics of tumour tissue where the SNR test statistic was used.

ISSN: 1471-2407