Factors associated with early treatment discontinuation in the AACR Project GENIE Biopharma Collaborative (BPC) non-small cell lung cancer (NSCLC) cohort

Background: Early treatment discontinuation in NSCLC is frequent and is associated with worse outcomes. We sought to identify clinical and genomic factors associated with early discontinuation among patients with NSCLC treated with immune checkpoint inhibitors (ICIs) or tyrosine kinase inhibitors (TKIs), now central to most NSCLC treatment strategies. We used the AACR Project GENIE BPC registry, which captures granular drug exposures along with other clinical and genomic variables. Methods: Data were obtained from the BPC NSCLC v2.0 cohort on May 19, 2025. Patients receiving ICIs or TKIs were included if the identified drugs had different start and end dates. Early discontinuation thresholds were first clinically estimated, then empirically defined using 1) Gaussian Mixture Models and 2) Classification and Regression Tree analysis, with consensus cutoffs of 3.0 and 4.3 months for TKIs and ICIs, respectively. Multivariable logistic regression (MLR) was applied to evaluate associations between early discontinuation and: sex; stage; drug class; oncogenic driver mutations in EGFR , ALK , KRAS , and BRAF ; and line of therapy (first-line [1L] vs later-line). Because the cohort predates wider first-line ICI adoption, most ICI exposures in BPC occur in later lines. Results: 910 of 1846 (49%) treatment episodes were included. By our definition, early discontinuation occurred in 420 (46%) of these. In MLR, TKI exposure (OR 0.17, 95% CI 0.12-0.24) and receiving TKI or ICI as 1L treatment (OR 0.56, 95% CI 0.39-0.78) were inversely associated with early discontinuation. Stage IV was associated with a trend towards early discontinuation (OR 1.56, 95% 0.94-2.65). Conversely, the presence of an EGFR mutation trended away from early discontinuation (OR 0.72, 95% CI 0.48-1.07). The final model demonstrated good discrimination with an AUC of 0.74. Calibration analysis showed excellent agreement between predicted and observed probabilities. Conclusions: Receiving TKIs in any line and ICI treatment in 1L were inversely associated with early discontinuation, suggesting differential tolerability. EGFR mutation status demonstrated a trend toward lower discontinuation risk, independent of TKI exposure or line of therapy. Stage IV trended towards a higher early discontinuation risk, which could reflect early progression of disease while on treatment. Limitations include: 1) the exclusion of episodes with a single dose or a clinical trial medication (both are recorded with the same start/end date in BPC); 2) a lack of ground truth for discontinuation reasons; 3) unknown cross-episode interactions in patients with multiple treatment episodes; and 4) unmeasured confounders, such as social determinants of health (SDOH). Despite these limitations, our findings highlight distinct factors that may inform individualized strategies and warrant further investigation.