Machine learning to refine prognostic and predictive nodal burden thresholds for post-operative radiotherapy in completely resected stage III-N2 non-small cell lung cancer

Abstract

BackgroundThe role of post-operative radiotherapy (PORT) for completely resected N2 non-small-cell lung cancer (NSCLC) is controversial in light of recent randomized data. We sought to utilize machine learning to identify a subset of patients who may still benefit from PORT based on extent of nodal involvement. Materials/MethodsPatients with completely resected N2 NSCLC were identified in the National Cancer Database. We trained a machine-learning based model of overall survival (OS). SHapley Additive exPlanation (SHAP) values were used to identify prognostic and predictive thresholds of number of positive lymph nodes (LNs) involved and lymph node ratio (LNR). Cox proportional hazards regression was used for confirmatory analysis. ResultsA total of 16,789 patients with completely resected N2 NSCLC were identified. Using the SHAP values, we identified thresholds of 3+ positive LNs and a LNR of 0.34+. On multivariate analysis, PORT was not significantly associated with OS (p = 0.111). However, on subset analysis of patients with 3+ positive LNs, PORT improved OS (HR: 0.91; 95% CI: 0.86–0.97; p = 0.002). On a separate subset analysis in patients with a LNR of 0.34+, PORT improved OS (HR: 0.90; 95% CI: 0.85–0.96; p = 0.001). Patients with 3+ positive lymph nodes had a 5-year OS of 38% with PORT compared to 31% without PORT. Patient with positive lymph node ratio 0.34+ had a 5-year OS of 38% with PORT compared to 29% without PORT. ConclusionsPatients with a high lymph node burden or lymph node ratio may present a subpopulation of patients who could benefit from PORT. To our knowledge, this is the first study to use machine learning algorithms to address this question with a large national dataset. These findings address an important question in the field of thoracic oncology and warrant further investigation in prospective studies.