Background – Studies have shown that early changes in quantitative measures of circulating tumor DNA (ctDNA) in response to immunotherapy can predict clinical outcomes and may be useful for identifying molecular responders (MRs) and molecular non-responders (nMRs) who may benefit from a change in therapy. Yet few studies have evaluated treatment with tyrosine kinase inhibitors (TKis).

Methods – The cohort consisted of deidentified patients from the Tempus clinicogenomic database with stage IV solid tumors who underwent ctDNA NGS. Patients who had a pre-treatment baseline liquid biopsy (T0) and a longitudinal sample 21-180 days after initiating TKi therapy (T1) were included. TKis were classified by the Tempus Medical Ontology, which was validated using the NCI Metathesaurus TKi concept (NCI Code C1967). xM for treatment response monitoring estimates tumor fraction (TF) via an ensemble algorithm that incorporates pathogenic variant allele frequencies, copy number information, and germline information. Evaluable patients had at least one blood sample with a TF ≥ limit of blank of 0.09%. MR was quantified as a ≥ 50% decrease in TF from T0 to T1. Patients with consistently low TF < 1% at both T0 and T1 were also classified as MRs given previous data demonstrating that baseline low TF is correlated with prolonged clinical outcomes. All other patients were classified as nMRs.

Results – Real-world overall survival (rwOS) was defined from T1 to death or last known clinical record in event-free patients. Stratified Cox proportional hazards models were used to estimate the hazard ratio (HR) for MR status (MR vs. nMR), stratified by line of therapy (LOT, first line, 1L vs. ≥ second-line, 2L+). A 1-sided Wald test was conducted using a 5% significance level. Predicted rwOS estimates were obtained from the Cox models. The evaluable cohort consisted of 109 patients (1L=62, 57%, 2L+=47, 43%). Cancer types included non-small cell lung cancer (NSCLC, n=52, 48%), breast cancer (BC, n=35, 32%), and 10 other cancer types each comprising < 5% of the patient cohort. The most common therapies were osimertinib in NSCLC patients (n=34) and alpelisib + fulvestrant in BC patients (n=11). Median time from the start of TKi therapy to T1 was 93 days and median follow-up was 10.3 months. Sixty patients had MR and 49 patients were classified as nMRs. MR was significantly associated with longer rwOS than nMRs (HR = 0.45; P = 0.006). Median predicted rwOS among MRs receiving TKis in 1L (n=43) was 28.4 months vs. 16.5 months for nMRs (n = 19). The HR for death between MR vs nMR was consistent across different cancer subgroups and LOT: HR=0.38 for NSCLC patients, HR=0.57 for BC patients, HR=0.46 for 1L patients and HR=0.44 for 2L+ patients.

Conclusions – These data demonstrate that longitudinal molecular treatment response can stratify rw-outcomes to TKi therapy in a heterogeneous, real-world pan cancer cohort.

VIEW THE PUBLICATION

VIEW THE POSTER