Identification and validation of an explainable predictive model for early diagnosis of non-small cell lung cancer metastasis: A peripheral immune score based on integrative machine learning

Background: Metastasis remains the leading cause of high mortality in non-small cell lung cancer (NSCLC), but early detection is challenging due to the limited sensitivity and specificity of current imaging methods. Peripheral immune markers offer predictive potential, yet their clinical use is limited by a lack of interpretable models. This study developed and validated an interpretable peripheral immune score (PIS) using machine learning (ML) to aid early diagnosis of NSCLC metastasis. Methods: We conducted a multicenter cross-sectional study of NSCLC patients in China. A derivation cohort of 309 patients from three campuses of Shanghai Hospital of Traditional Chinese Medicine (March 2023-May 2025) was split 8:2 for training and validation. Baseline data and 37 peripheral immune markers were collected. Causal inference screened predictive markers, and eight ML algorithms were applied. Model performance was assessed using AUC, decision curve analysis, and calibration. The best model was interpreted using SHAP and deployed as an online PIS Calculator. Results: The Random Forest (RF) model showed the highest performance. After feature reduction, a final interpretable RF model with 19 features accurately predicted metastasis in the validation set (AUC = 0.942). This model was translated into the Intelligent Peripheral Immunity Score (PIS), identifying high-risk patients even without radiographic evidence. Conclusion: The PIS system integrates peripheral immune markers with ML to provide an accurate, interpretable tool for early detection of NSCLC metastasis, overcoming limitations of conventional imaging and complex models, and offering a clinically actionable solution for improved patient management.