Development of a risk-stratified follow-up strategy based on early and late recurrence risk factors in locally advanced esophageal squamous cell carcinoma after definitive chemoradiotherapy

Abstract

The high recurrence rate following definitive chemoradiotherapy (dCRT) for locally advanced esophageal squamous cell carcinoma (ESCC) remains a paramount clinical challenge. However, there is currently a lack of research to explore the different effects of inflammation-related indicators on early recurrence (ER) and late recurrence (LR). We retrospectively analyzed 905 locally advanced ESCC patients. Six machine learning algorithms were used for consensus feature selection, followed by multivariate Cox regression to identify independent prognostic factors for ER and LR. A prognostic nomogram was developed, and restricted mean survival time (RMST) quantified survival differences, guiding follow-up strategies. Machine learning combined with Cox analysis identified RT dose, tumor thickness, and N stage as independent risk factors for ER, while concurrent chemotherapy, N stage, pan-immune-inflammation value (PIV), and platelet-to-albumin ratio (PAR) were identified for LR. Risk models effectively stratified patients into high-, moderate-, and low-risk groups with significantly distinct PFS and OS, and high-risk groups were enriched with adverse features including elevated N stage, PAR, PIV, and tumor burden. The developed nomogram demonstrated good calibration, discriminative ability (with 5-year OS and PFS AUC values of 0.738 and 0.771), and provided higher net clinical benefit than single variables. The C-index of the model was 0.652 and 0.677 for OS and PFS. Based on RMST analysis revealing a significant survival gradient across risk subgroups (5-year OS-RMST: 21.0 vs. 32.5 vs. 44.7 months for high-, moderate-, and low-risk groups, respectively), a risk-stratified follow-up strategy was developed, recommending intensive (every 1–3 months), step-down, and annual surveillance schedules for high-, moderate-, and low-risk patients. We developed and internally validated a prognostic nomogram based on inflammatory indicators that can effectively quantify risks and stratify them. Moreover, we formulated a novel risk-adaptive follow-up strategy framework based on RMST, providing potential decision-making tools for the individualized and precise management of ESCC. However, this strategy remains hypothesis-generating and requires external prospective validation before widespread clinical adoption.