Table 3. Comparison of Performance of the Different RPA Staging Systems and TNM Staging System Under 5 Evaluation Criteria.
| Criteria | SYSUCC training set | SYSUCC internal validation set | Foshan external validation set | Bootstrap validation set | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| RPA staging | RPA staging 118 | RPA staging 217 | TNM staging | RPA staging | RPA staging 118 | RPA staging 217 | TNM staging | RPA staging | RPA staging 118 | RPA staging 217 | TNM staging | RPA staging | RPA staging 118 | RPA staging 217 | TNM staging | |
| Hazard consistency (similarity of survival rate for subgroups within each stage group) | ||||||||||||||||
| Standardized score | 0.00 | 0.23 | 1.00 | 0.85 | 0.00 | 0.12 | 0.66 | 1.00 | 0.00 | 0.52 | 1.00a | 0.67 | 0.00 | 0.25 | 1.00 | 0.75 |
| Rank | 1 | 2 | 4 | 3 | 1 | 2 | 3 | 4 | 1 | 2 | 4a | 3 | 1 | 2 | 4 | 3 |
| Hazard discrimination (differences in survival rate across stage groups) | ||||||||||||||||
| Standardized score | 1.00 | 0.44 | 0.92 | 0.00 | 0.92 | 0.14 | 1.00 | 0.00 | 1.00 | 0.00 | 0.28a | 0.08 | 1.00 | 0.28 | 0.64 | 0.00 |
| Rank | 4 | 2 | 3 | 1 | 3 | 2 | 4 | 1 | 4 | 1 | 3a | 2 | 4 | 2 | 3 | 1 |
| Likelihood difference (difference in goodness-of-fit between models) | ||||||||||||||||
| Standardized score | 0.00 | 0.83 | 1.00 | 0.93 | 0.00 | 0.81 | 0.84 | 1.00 | 0.00 | 0.92 | 1.00a | 0.83 | 0.00 | 0.80 | 1.00 | 0.88 |
| Rank | 1 | 2 | 4 | 3 | 1 | 2 | 3 | 4 | 1 | 3 | 4a | 2 | 1 | 2 | 4 | 3 |
| Explained variance (percentage of variation in survival rate accounted for by stage groupings) | ||||||||||||||||
| Standardized score | 0.00 | 0.41 | 1.00 | 0.80 | 0.00 | 0.14 | 0.89 | 1.00 | 0.00 | 0.55 | 1.00a | 0.60 | 0.00 | 0.38 | 1.00 | 0.74 |
| Rank | 1 | 2 | 4 | 3 | 1 | 2 | 3 | 4 | 1 | 2 | 4a | 3 | 1 | 2 | 4 | 3 |
| Balance (difference in sample sizes across stage groups) | ||||||||||||||||
| Standardized score | 0.00 | 0.37 | 0.96 | 1.00 | 0.00 | 0.32 | 0.81 | 1.00 | 0.09 | 0.00 | 1.00a | 0.81 | 0.00 | 0.37 | 0.94 | 1.00 |
| Rank | 2 | 1 | 3 | 4 | 1 | 2 | 3 | 4 | 2 | 1 | 4a | 3 | 2 | 1 | 3 | 4 |
| Overall | ||||||||||||||||
| Score | 1.00 | 2.29 | 4.87 | 3.58 | 0.92 | 1.53 | 4.20 | 4.00 | 1.09 | 1.99 | 4.28a | 2.99 | 0.71 | 2.15 | 4.52 | 3.54 |
| Rank | 1 | 2 | 4 | 3 | 1 | 2 | 4 | 3 | 1 | 2 | 4a | 3 | 1 | 2 | 4 | 3 |
Abbreviations: RPA, recursive partitioning analysis; SYSUCC, Sun Yat-Sen University Cancer Center; TNM, tumor-node-metastasis.
a
According to the laboratory practice standards used in the First Hospital of Foshan, an EBV DNA titer less than 1000 was reported as negative and not given an actual value. As a compromise, we adopted the nearest integer (1000) as a substitute in the calculation procedure.