Machine Learning Approaches to Personalized Therapy for Advanced Non-small Cell Lung Cancer With Real-World Data

Machine Learning Approaches to Personalized Therapy for Advanced Non-small Cell Lung Cancer With Real-World Data

This research will leverage machine learning (ML) and causal inference techniques applied to real-world data (RWD) to generate evidence that personalizes treatment strategies for patients with advanced non-small cell lung cancer (aNSCLC). Rather than influencing regulatory decisions or clinical guidelines, the goal of this trial is to refine treatment selection among existing therapeutic options, ensuring that care is tailored to individual patient characteristics. Additionally, by generating real-world evidence, these findings will inform the design and implementation of future clinical trials. Importantly, the methodological advancements will establish a pipeline that extends beyond aNSCLC, facilitating the identification of optimal dynamic treatment regimes (DTRs) for other complex diseases.

The proposed research will enhance patient-centered outcomes research (PCOR) and comparative effectiveness research (CER) methodologies by addressing two key challenges: (1) appropriate handling of missing EHR data and (2) rigorous causal inference techniques for sequential treatment strategies. By focusing on treatment strategies tailored to individual patients and incorporating patient-reported outcomes (PROs), this study is fundamentally patient-centered. Furthermore, the research is guided by practicing physicians, a patient advocate, and a former patient caregiver, ensuring that it remains aligned with the needs and priorities of those directly affected by aNSCLC.

This study will develop novel reinforcement learning algorithms by integrating multiply robust matching-based approaches. This study will tailor each component of DTR to optimize treatment sequences for aNSCLC patients, leveraging two large-scale, high-quality nationwide real-world electronic health record (EHR) databases: the Flatiron aNSCLC database and the CancerLinQ lung cancer database. These databases provide comprehensive clinicodemographic and longitudinal patient data.

Additionally, incorporating PRO data from two National Cancer Institute (NCI)-designated Comprehensive Cancer Centers -Huntsman Cancer Institute (HCI) and Moffitt Cancer Center (MCC) - will enable this trial to capture the patient perspective when personalizing aNSCLC care recommendations. Key outcomes will include overall survival, quality-adjusted life years (QALYs), time to second progression or death (PFS2), and time to worsening of selected PROs, all framed as time-to-event outcomes.

These methodological innovations will establish a reproducible pipeline for translating real-world evidence from large-scale EHR data into personalized DTR recommendations for aNSCLC patients and other complex disease populations.

Inclusion Criteria

Subjects must meet all of the following eligibility criteria:

• Diagnosed with advanced NSCLC between January 1, 2011, and June 30, 2024.
• Follow-up available until December 31, 2024, with a minimum potential follow-up period of at least six months.

Exclusion Criteria

Subjects meeting any of the following criteria at baseline will be excluded:

• Fewer than one day of follow-up post-initiation of first-line (1L) therapy.
• Presence of a targetable mutation, including ALK, BRAF, EGFR, KRAS, or ROS1.
• PD-L1 expression <50% at baseline (restricted to patients with PD-L1 ≥50%).
• First-line treatment limited to immunotherapy or chemoimmunotherapy (excluding other treatment regimens).
• Patients receiving second-line (2L) treatment, including those enrolled in a clinical study.