Comparisons of prognosis prediction accuracy between modified and unmodified versions of 8(th) edition ypTNM

BACKGROUND: The prognostic value of the existing 8(th) edition post-neoadjuvant treatment (ypTNM) appears to be limited, and necessary reassessment and modification should be carried out as needed. This study aimed to compare the prognosis prediction accuracy of modified and unmodified versions of the 8(th) edition ypTNM. METHODS: Esophageal cancer patients who had received neoadjuvant therapy from the Surveillance, Epidemiology, and End Results (SEER) database were included in this observational longitudinal study. The median follow-up time was 26 months. All-cause mortality was the outcome variable. Demographic and clinical variables were collected as covariates. Kaplan-Meier (log-rank test) and Cox proportional hazards models were conducted for developing modified ypTNM staging. The concordance index (C-index) was calculated to analyze the discriminative ability of modified ypTNM staging. RESULTS: Overall, 3,595 patients met inclusion criteria. The 8(th) edition staging was not able to significantly discriminate between patients with ypT1- and ypT2-, ypT3- and ypT4-, ypN2- and ypN3- disease, respectively. Using the modified staging, we found that patients with ypT0–2 [hazard ratio (HR) =1.232; 95% confidence interval (CI): 1.053–1.441] and ypT3–4 (HR =1.257; 95% CI: 1.136–1.390) with grade III + IV had a significant risk of death compared to those with grade I + II. As was the case for the ypN0 (HR =1.295; 95% CI: 1.073–1.562) group with middle and upper tumor locations compared to those with low tumor location. The modified staging possessed better homogeneity in terms of the chi-square likelihood ratio (143.443 vs. 102.044), Akaike information criterion (AIC) (32,683.716 vs. 32,719.115), and Schwarz’s Bayesian criterion (SBC) (32,723.496 vs. 32,741.847), as well as better discriminatory ability (C-index of 0.577 vs. 0.560, P=0.045) compared to the 8(th) edition staging. CONCLUSIONS: Although the modified ypTNM staging system we created by incorporating tumor grade and location to the original T and N displayed certain prognosis prediction accuracy compared with the 8(th) edition ypTNM staging, a larger sample size and prospective studies are needed to explore.