Table 2.
C-index for each prognostic model for survival prediction.
| Models for survival prediction | Predictors in each model | Training cohort | Validation cohort | |||||||
|---|---|---|---|---|---|---|---|---|---|---|
| C-index | 95% CI | C-index | 95% CI | P values | ||||||
| Pretreatment clinical prognostic models | cTN | cT + cN | 0.578 | (0.522-0.634) | 0.611 | (0.531-0.691) | .429 | .503 | .039* | <.001* |
| PreM | CEA+GLO | 0.627 | (0.572-0.682) | 0.552 | (0.482-0.622) | .307 | .356 | .007* | <.001* | |
| Posttreatment clinical prognostic models | ypTN | ypT+ypN | 0.675 | (0.615-0.735) | 0.532 | (0.441-0.623) | .156 | .636 | .033* | <.001* |
| PostM1 | CEA+GLO+ypT | 0.737 | (0.679-0.795) | 0.603 | (0.534-0.672) | ref | .647 | .027* | <.001* | |
| PostM2 | TRG+CEA+GLO | 0.664 | (0.603-0.725) | 0.609 | (0.542-0.676) | .647 | ref | .077 | .009* | |
| Radiomics signature | Radscore | Radscore | 0.937 | (0.917-0.957) | 0.730 | (0.651-0.809) | .010* | .077 | ref | .014* |
| Integrated prognostic model | iPostM | Radscore+TRG | 0.942 | (0.922-0.962) | 0.752 | (0.684-0.820) | <.001* | .009* | .014* | ref |
P values were calculated by comparing with the corresponding reference prognostic model in each column in the validation cohort (ref represents the reference model). A P value < 0.05 indicates a significant difference.
*Represent P < 0.05. In the validation cohort, five clinical prognostic models showed similar PFS predictive power. Notably, compared with PreM without TRG, the constructed PostM2 achieved better predictive performance (P = 0.356). The developed radiomics signature appeared to be more accurate than clinical prognostic models (P = 0.007 to 0.077). The integrated model (iPostM) combining radiomics signature and TRG gained the highest C-index in the validation cohort (0.752), outperforming the radiomics signature and all other clinical prognostic models in term of evaluating 3-year PFS (all P < 0.05).
cTN, the clinical stage prognostic model; PreM, the pre-treatment clinical prognostic model; ypTN, the pathologic stage prognostic model; PostM1, the post-treatment clinical prognostic model; PostM2, the post-treatment clinical prognostic model without pathologic stage; iPostM, the integrated prognostic model combining TRG and radiomics signature; CI, confidence interval; GLO, globulin; TRG, tumor regression grade; cT/N: clinical T/N stage; ypT/N, the pathologic classification after nCRT; CEA, carcinoembryonic antigen; CA19-9, carbohydrate antigen 19-9.