Table 3.
Results of 10 independent replicates of 10-fold cross-validation of all combinations of the selected five metabolites for prediction of functional recovery after stroke.
| NO | Proline | Threonine | Leucine-Isoleucine | Glutamate | Arginine | Mean AUC in training set | Mean AUC in test set | SD of AUC in test set | 95% CI of mean AUC in test set |
|---|---|---|---|---|---|---|---|---|---|
| 1 | 1 | 0 | 0 | 1 | 1 | 0.952 | 0.835 | 0.109 | (0.756, 0.913) |
| 2 | 1 | 0 | 1 | 1 | 1 | 0.956 | 0.817 | 0.113 | (0.735, 0.897) |
| 3 | 1 | 0 | 0 | 0 | 1 | 0.920 | 0.808 | 0.099 | (0.737, 0.879) |
| 4 | 0 | 1 | 0 | 1 | 1 | 0.947 | 0.799 | 0.115 | (0.717, 0.881) |
| 5 | 0 | 1 | 1 | 1 | 1 | 0.991 | 0.795 | 0.094 | (0.727, 0.861) |
| 6 | 0 | 0 | 1 | 1 | 1 | 0.924 | 0.793 | 0.106 | (0.716, 0.868) |
| 7 | 1 | 1 | 0 | 1 | 1 | 0.972 | 0.792 | 0.130 | (0.699, 0.884) |
| 8 | 1 | 0 | 1 | 0 | 1 | 0.929 | 0.788 | 0.093 | (0.721, 0.854) |
| 9 | 1 | 1 | 0 | 0 | 1 | 0.929 | 0.773 | 0.114 | (0.691, 0.853) |
| 10 | 1 | 1 | 1 | 0 | 1 | 0.935 | 0.771 | 0.114 | (0.689, 0.852) |
| 11 | 1 | 0 | 0 | 1 | 0 | 0.870 | 0.767 | 0.115 | (0.684, 0.849) |
| 12 | 0 | 0 | 1 | 0 | 1 | 0.857 | 0.766 | 0.094 | (0.698, 0.833) |
| 13 | 1 | 0 | 1 | 1 | 0 | 0.871 | 0.758 | 0.103 | (0.684, 0.831) |
| 14 | 1 | 1 | 1 | 1 | 1 | 0.994 | 0.758 | 0.119 | (0.672, 0.843) |
| 15 | 0 | 1 | 1 | 0 | 1 | 0.895 | 0.744 | 0.120 | (0.658, 0.830) |
| 16 | 1 | 1 | 0 | 1 | 0 | 0.909 | 0.744 | 0.136 | (0.647, 0.842) |
| 17 | 1 | 1 | 1 | 1 | 0 | 0.908 | 0.728 | 0.121 | (0.642, 0.815) |
| 18 | 0 | 0 | 0 | 1 | 1 | 0.846 | 0.723 | 0.096 | (0.655, 0.792) |
| 19 | 0 | 1 | 0 | 0 | 1 | 0.852 | 0.713 | 0.102 | (0.640, 0.787) |
| 20 | 0 | 1 | 1 | 1 | 0 | 0.877 | 0.698 | 0.134 | (0.602, 0.794) |
| 21 | 0 | 0 | 0 | 0 | 1 | 0.758 | 0.698 | 0.084 | (0.637, 0.758) |
| 22 | 0 | 1 | 0 | 1 | 0 | 0.850 | 0.694 | 0.128 | (0.603, 0.785) |
| 23 | 1 | 0 | 0 | 0 | 0 | 0.745 | 0.654 | 0.115 | (0.572,0.737) |
| 24 | 0 | 1 | 0 | 0 | 0 | 0.750 | 0.653 | 0.130 | (0.561, 0.746) |
| 25 | 1 | 1 | 0 | 0 | 0 | 0.748 | 0.628 | 0.129 | (0.536,0.721) |
| 26 | 0 | 0 | 1 | 1 | 0 | 0.764 | 0.628 | 0.092 | (0.561,0.694) |
| 27 | 0 | 0 | 0 | 1 | 0 | 0.680 | 0.626 | 0.097 | (0.557,0.695) |
| 28 | 1 | 0 | 1 | 0 | 0 | 0.743 | 0.625 | 0.111 | (0.546, 0.704) |
| 29 | 1 | 1 | 1 | 0 | 0 | 0.751 | 0.604 | 0.119 | (0.519, 0.689) |
| 30 | 0 | 1 | 1 | 0 | 0 | 0.730 | 0.573 | 0.126 | (0.483, 0.664) |
| 31 | 0 | 0 | 1 | 0 | 0 | 0.645 | 0.536 | 0.102 | (0.463, 0.609) |
1 refers to containing this metabolite when performing 10-fold cross-validation, and 0 means that the metabolite is not contained.
According to the metabolites combinations, there were 31 combinations for prediction of functional recovery after stroke that were analyzed by cross-validation. The standard deviations (SD) of the AUC in the testing set are between 0.09 and 0.13, which indicates the model of this study is steady. The values of mean AUC in the test set are used as criteria for evaluation. The panel including proline, glutamate and arginine had the highest values of mean AUC in the test set [0.835, 95% CI (confidence interval): (0.756, 0.913)] [mean AUC in the training set:0.952, 95% CI: (0.951, 0.953)] in the 31 combinations. As a result, this panel was the best combination of predictive biomarkers of the 31 combinations.