Development and validation of a serum inflammatory biomarker-driven machine-learning model for prognostic stratification in surgical limited-stage small cell lung cancer.

Background: Robust prognostic markers for small cell lung cancer (SCLC) are currently lacking, underscoring the need for novel prediction models to optimize individualized treatment and improve patient outcomes. Inflammatory/nutritional indexes have been extensively employed in prognostic investigations of malignant tumors. The study aimed to precisely ascertain the prognosis of SCLC patients undergoing surgery by preoperative serological indexes. Methods: We included patients with SCLC who underwent surgery at The Affiliated Hospital of Qingdao University. Potential predictors included basic clinical characteristics and preoperative serum inflammatory/nutritional indexes. We employed 10 machine learning algorithms and their 101 combinations to select the superior model and establish a novel nomogram. Follow-up involved regular clinic visits or telephone contact, with imaging and laboratory tests conducted at defined intervals to assess overall survival (OS) and progression-free survival (PFS). The cohort was randomly split into training and validation cohorts in a 7:3 ratio. Harrell's C-index, Kaplan-Meier curves, log-rank tests, and Cox regression analyses were used for model evaluation and prognostic assessment. Results: A total of 219 patients were included in this study. Prognostic nutritional index (PNI), lymphocyte-to-monocyte ratio (LMR), platelet-to-neutrophil ratio (PNR), neutrophil-to-lymphocyte ratio (NLR), systemic immune-inflammatory index (SII), pan-immune-inflammation value (PIV), and systemic inflammatory response index (SIRI) were correlated with the prognosis of SCLC patients. Smoking status and the tumor-node-metastasis (TNM) stage were independent prognostic indicators of OS. The Random Forest model achieved the highest mean concordance index (C-index) (0.784). Patients classified as high-risk based on this model exhibited a higher prevalence of smoking and more advanced pathological N stage and TNM stage. No significant differences were observed between risk groups regarding age, gender, body mass index (BMI), alcohol history, tumor site, pathological T stage, Ki-67 index, or visceral pleural invasion (VPI). Nomograms based on risk grouping, smoking status, and TNM stage demonstrated high precision and considerable clinical value. Multivariate Cox analysis identified PNI and NLR as the most valuable prognostic markers, with optimal cut-off values of 50.6 and 1.99, respectively. Conclusions: A machine learning model based on serological inflammatory/nutritional indexes can reasonably estimate the long-term prognosis of SCLC patients and is anticipated to serve as a practical instrument for identifying the ideal candidates for thoracic surgery.