Abstract
Lung squamous cell carcinoma (LUSC) is a major challenge in oncology due to its high recurrence rate, highlighting the need to identify recurrence drivers. Based on the data provided by Zhang Zemin’s team in their research, we selected survival metadata from 119 patients with lung squamous cell carcinoma and data on the proportions of 51 immune cell types. We then combined machine learning methods to analyze the intrinsic tumor factors and tumor microenvironment (TME) in the recurrence of lung squamous cell carcinoma (LUSC). Chi-square tests identified immune cell subsets enriched in recurrent patients, such as high-MKI67 Tregs, low-FGFBP2 NK cells, low-FOXP3 Tregs, and low-KLRB1 CD8 + T cells. A random forest algorithm further pinpointed pathological response rate (PRR) as the primary predictive factor, with high-MKI67 Tregs as a key secondary contributor. Univariate and multivariate Cox regression analyses and Kaplan-Meier survival curves confirmed that pathological complete response (pCR) and low Treg_MKI67 expression were associated with better survival (p = 0.0055 and p = 0.011, respectively). Patients achieving pCR and with low Treg_MKI67 expression had superior recurrence-free survival (RFS; 2-year RFS: 96.7% vs. 57.9% for non-pCR and high Treg_MKI67, p = 0.00066). These findings underscore the prognostic value of PRR and TME markers. They also highlight the potential of using single-cell RNA sequencing and machine learning to guide personalized therapies and reduce recurrence risk by targeting Treg activity, laying the groundwork for integrating immune markers into clinical practice to improve LUSC prognosis. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12967-025-07193-9.