Background: Early changes in circulating tumor DNA (ctDNA) may predict clinical outcomes. Yet few studies have evaluated these changes in patients (pts) treated with targeted therapies. Furthermore, there is limited data on the clinical value of continual monitoring. Here, we evaluate ctDNA tumor fraction (TF) as a longitudinal molecular biomarker to predict clinical outcomes in an advanced real-world (rw) pan-cancer cohort treated with targeted therapies, including therapies targeting oncogenic alterations.

Methods: The cohort included deidentified pts from the Tempus clinicogenomic database with stage IIIB or IV solid tumors who underwent ctDNA NGS treated with targeted therapy. Pts had a pre-treatment baseline liquid biopsy (T0) and ≥1 on-treatment sample (Ti) within 21-180 days of targeted therapy. Tempus xM for treatment response monitoring estimates TF via an ensemble algorithm that incorporates variant and copy number information. Evaluable pts had ≥ 1 sample with a TF ≥0.09%. At each on-treatment time point, pts were classified as either a Molecular Responder(MR) if TF < 1% at both T0 and Ti or if TF decreased by ≥ 50% from T0 to Ti; otherwise pts were classified as a Molecular Non-Responder (nMR). In the early time point sub cohort, T1 < 84 days. In the longitudinal sub cohort, pts had ≥ 2 on-treatment samples and were classified as longitudinal MR/nMR based on the most frequent classification across all Ti.Real-world progression-free survival (rwPFS) and real-world overall survival (rwOS) were defined from T1 to progression or death (rwPFS), or death (rwOS). For rwPFS, pts with progression events between T0 and T1 were excluded.

Results: In the full 104 pt cohort, 50 pts had NSCLC (48.1%), 40 first-line (1L); 35 (33.7%) pts had breast cancer (BC), 15 1L. The most common alterations were EGFR in NSCLC (n = 34, 32 pts treated with osimertinib) and PIK3CA in BC, (n = 11, all pts treated with alpelisib + fulvestrant).rwPFS was longer for 43 MR pts vs. 26 nMR pts (HR = 0.67, P > 0.05, not statistically significant). rwOS was longer for 59 MR pts vs. 45 nMR pts (HR = 0.40, P = 0.003). Among 40 pts with 1L NSCLC, the rwOS was longer in 31 MR pts vs. 9 nMR pts (HR = 0.37, P = 0.03).In the early time point sub cohort of 43 pts, 24 (55.8%) pts had NSCLC and 9 (20.9%) had BC.MRs had longer rwPFS and rwOS than nMRs (22 MRs vs. 15 nMRs, HR = 0.40, P = 0.03 for rwPFS and 25 MRs vs. 18 nMRs, HR = 0.18; P = 0.001 for rwOS).In the longitudinal sub cohort of 47 pts, 29 (61.7%) pts had NSCLC, 7 (14.9%) had BC. rwOS was longer for 30 longitudinal MRs vs. 17 longitudinal nMRs (HR = 0.23, P = 0.04).

Conclusions: These data demonstrate that a molecular biomarker for targeted therapy treatment response monitoring both early on-treatment and in pts continually monitored may be a useful tool for treatment decision making. Larger datasets are needed to further validate these results in cancer-specific cohorts.

VIEW THE PUBLICATION