The analysis objective was to assess if limited tumor assessments can predict long-term overall survival (OS) in patients (pts) with locally advanced or metastatic (Stage IIIB-IV) non-small cell lung cancer (NSCLC) treated with durvalumab.
We used data from a Phase II, non-comparative, open-label multicenter study of durvalumab in NSCLC pts with ≥2 prior systemic treatment regimens, including 1 platinum-based (ATLANTIC). Per exploratory analysis, the first 2 post-baseline assessments were used to develop the model. Using an elastic net statistical method, combined with cross validation, we identified important baseline variables, built a scoring system (defined as 0.28*sex + 0.188*histology group + 0.034*smoker group – 0.176*line of therapy – 0.041*tumor assessment) in which assessments are represented as a single variable (interpreted as a weighted average), and identified the optimal score thresholds to segment pts into 2 groups (‘good’ vs. ‘bad’) with significant differences in long-term OS.
As of June 3, 2016, 444 pts had received treatment; 191 from cohort 2 (EGFR/ALK wild-type pts) with sufficient assessments (baseline and ≥1 follow-up) were used to develop the model. Median age was 64.0 years, 61.8% had WHO PS 1, 18.8% had squamous histology, mean number of prior anticancer regimens was 4.0, and 83.7% were current/ex-smokers; PD-L1 expression was high (≥25% of tumor cells stained) in 57.1%, low/negative in 35.6%, and unknown in 7.3%. OS results are summarized in the table. The model was validated using data from a Phase I/II open-label trial of durvalumab (1108).Table:
|ATLANTIC (model building)||Study 1108 (validation)|
|Bad Group||Good Group||Bad Group||Good Group|
|(n = 157)||(n = 34)||(n = 117)||(n = 58)|
|Median OS (95% CI), days||340||NE||265||739|
|(292, 403)||(557, NE)||(194, 315)||(616, NE)|
|6-month OS rate (95% CI)||0.742||0.941||0.605||0.937|
|(0.665, 0.804)||(0.785, 0.985)||(0.517, 0.682)||(0.855, 0.973)|
|1-year OS rate (95% CI)||0.478||0.882||0.371||0.819|
|(0.397, 0.554)||(0.716, 0.954)||(0.282, 0.460)||(0.708, 0.891)|
|HR [Good vs. Bad] (95% CI)||0.2059 (0.0569, 0.4437)||0.2637 (0.1661, 0.4187)|
NE, not estimable
We developed an algorithm based on baseline characteristics and tumor assessments to segment NSCLC pts treated with durvalumab into 2 groups with distinct OS. The scoring system was independently validated. However, the predictive versus prognostic value of this algorithm needs further evaluation using data from randomized trials.
Clinical trial identification
NCT02087423 (release date: March 4, 2014)
Legal entity responsible for the study
X. Zhang: Full time employee of AstraZeneca. K. Park: Consulting: Astellas, AZ, Boehringer Ingelheim, Clovis, Lilly, Hanmi, Kyowa Hakko Kirin, Novartis, Ono Pharma. Speaker Bureau: Boehringer Ingelheim, Research funding: AZ. N.A. Rizvi: Advisory Board: Merck, AZ, Roche, BMS, Novartis, Pfizer, Lilly, Novartis, Abbvie Co-founder and shareholder: Gritstone Oncology Scientific Advisory Board: Nilogen Oncosystems. P.A. Dennis, Y. Huang, P. Mukhopadhyay: Employee and shareholder AstraZeneca. R. Narwal: Employee and shareholder MedImmune. R. Arani: Employee B&I, AstraZeneca, Shareholder AstraZeneca.