Presentation (CTI)

OBJECTIVES: To outline methodological strategies for applying artificial intelligence (AI) and machine learning (ML) to real‑world data (RWD) in non-small‑cell lung cancer (NSCLC) to improve stratification based on molecular biomarkers and clinical characteristics for treatment selection while meeting credibility standards for real‑world evidence (RWE), including prespecified protocols, transparent reporting, and reproducibility requirements outlined by FDA, NICE, and ISPOR/ISPE frameworks.
METHODS: We outline a structured framework for applying AI/ML to NSCLC real-world data, guided by regulatory and methodological standards. Key components include:

1. Prespecified analysis protocols defining cohort selection, covariates, and endpoints;
2. Documentation of data provenance and quality checks;
3. Bias mitigation strategies (e.g., confounding control, handling missingness);
4. Model development principles emphasizing internal validation and interpretability;
5. Transparent reporting to support reproducibility.

Clinical context is based on NSCLC molecular testing guidance (e.g., EGFR, ALK, KRAS, PD‑L1)¹,² to ensure variable selection and endpoints align with standard-of-care decision points. Evidence planning and documentation follow frameworks from FDA³, NICE⁴, and ISPOR/ISPE⁵.
RESULTS: With rigorous design and validation, AI/ML models such as gradient boosting and regularized regression applied to curated NSCLC RWD (e.g., EHR and registry data) can achieve discrimination and calibration suitable for decision support. The framework highlights performance metrics (e.g., area under the curve, calibration curves) and visualizations showing how clinical and biomarker data inform treatment decisions. Published evaluations of AI/ML models in oncology show that open, well-documented workflows, predefined analysis plans, and model diagnostics facilitate credibility and clinical interpretation without reliance on proprietary tools.
CONCLUSIONS: AI‑enabled analyses of NSCLC RWD can complement evidence from randomized trials by providing timely, context‑specific insights to support personalized treatment decisions. Adherence to recognized frameworks and reporting standards strengthens validity, reproducibility, and confidence, supporting responsible translation of AI into clinical decision‑making.