Background – Only a fraction of non-small cell lung cancer patients who receive immune checkpoint inhibitor (ICI) treatment will respond, largely because current FDA-approved biomarkers still leave room for improvement to identify more responders. We hypothesized that modeling the time-dependent interaction between treatment and tumor genetics may reveal novel associations. We further hypothesized that real-world data (RWD), as opposed to clinical trial data, may contain these associations, as RWD possesses a potentially more diverse patient population and time-to-event treatment history across the diverse regimens used in routine clinical practice.

Methods – We examined the relation between tumor genetics, treatment, and real-world progression-free survival (PFS) in two RWD cohorts (Tempus discovery cohort, n = 1458; Dana Farber Cancer Institute Profile independent validation cohort, n = 466) by utilizing a Cox proportional hazards model with a time-dependent covariate for whether a patient was receiving an ICI-containing treatment regimen. Specifically, we included an indicator variable that denoted whether an individual was receiving an ICI regimen (=1) or not (=0), and enumerated the time intervals associated with each treatment status. We performed a univariate analysis for each mutated gene from tumor sequencing, and included a gene-treatment interaction term to permit identification of genes that were predictive (i.e. specific to a treatment) rather than prognostic. Our model included additional covariates to control for known predictive (tumor mutation burden [TMB]) and prognostic factors (age, sex).

Results – In the discovery cohort, we identified a significant association between ROS1 single nucleotide variants (SNV; prevalence = 5%) and favorable PFS on ICI therapy (interaction HR = 0.34, 95% CI = 0.18-0.64, p = 7.8e-4, q = 0.08) versus non-ICI treatment. This association was not explained by ROS1 rearrangement or ROS1 TKI therapy. ROS1 SNV associated with worse PFS when not on ICI treatment (marginal HR = 2.3, 95% CI = 1.4-3.6, p = 4e-4, q = 0.044). This ROS1 SNV association was replicated in the validation cohort, which exhibited a favorable interaction with ICI treatment (interaction HR = 0.11, 95% CI = 0.04-0.33, p = 7.46e-5), and poor PFS on non-ICI treatment (marginal HR = 5.39, 95% CI = 2.5-11.6, p = 1.96e-5). ROS1 SNV treatment interaction effect was also not explained by TMB or PD-L1 status. Finally, we observed the ROS1 SNV ICI-specific protective effect was specific to patients ICI-treated in the first line setting.

Conclusions – We identified an association between ROS1 SNVs and favorable PFS specific to ICI treatment, independent of common pre-existing predictive biomarkers for ICIs. Because this association was specific to the first line ICI setting, our finding may clarify ICI usage where clinical practice for ICI treatment is still evolving.

VIEW THE PUBLICATION

VIEW THE POSTER