Table 1.
Results from 10-fold cross validation.
| Split1 | Split2 | Split3 | Split4 | Split5 | Split6 | Split7 | Split8 | Split9 | Split10 | Median | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Round 1 | 100% | 84% | 73% | 37% | 75% | 70% | 50% | 75% | 62% | 23% | 71% |
| Round 2 | 53% | 88% | 42% | 80% | 53% | 100% | 55% | 88% | 88% | 80% | 80% |
| Round 3 | 83% | 49% | 62% | 71% | 50% | 100% | 53% | 100% | 75% | 100% | 73% |
| Round 4 | 100% | 60% | 48% | 80% | 75% | 80% | 88% | 100% | 50% | 93% | 80% |
| Round 5 | 80% | 96% | 100% | 35% | 71% | 68% | 88% | 100% | 92% | 70% | 84% |
| Random 1 | — | — | — | — | — | — | — | — | — | — | — |
| Random 2 | 80% | — | — | — | — | — | — | — | — | — | — |
| Random 3 | — | — | — | — | — | — | — | — | — | — | — |
| Random 4 | — | — | — | — | — | — | — | — | — | — | — |
| Random 5 | — | — | — | — | — | — | — | — | — | — | — |
We used the sTRA-only, CA19-9-only, and dual-labeled tissue markers to seek panels that distinguish 30 short-TTP samples from 15 long-TTP samples (Fig. 6). Each value is the average accuracy of the panels from the training set applied to the set-aside test samples for each split (see Materials and Methods for details). The average of the medians from the five independent rounds was 78%. For the rounds marked ‘Random’, we randomly assigned a case or control status to each of the 45 samples and repeated the process. Except for Split1 of the second round, the program did not find any panels that met the minimum performance in the training sets.