Study of AZD9291 in NSCLC Patients Harboring T790M Mutation Who Failed EGFR TKI and With Brain and/or LMS

1. Primary end points * Overall response rate (ORR) in CNS -brain metastasis cohort * Overall survival - Leptomeningeal with or without brain metastasis cohort 2. Secondary end points * Whole body disease control rate (DCR) * Time to brain progression * Progression free survival (PFS) in BM cohort * Overall survival (OS) * Adverse events (AEs) * Exploratory analysis of EGFR mutation/T790M in tissue, plasma or cerebrospinal fluid (CSF) 3. Treatment AZD9291 160mg po daily (1 cycle of 28 days) 4. Total patient sample size Eligible patients will be divided into two cohorts, brain metastasis (BM) cohort and leptomeningeal metastasis (LM) with or without BM cohort. The two patient cohorts use different primary endpoints. 5. BM cohort For the BM cohort, the primary outcome is overall response (CR+PR). Let P denote the overall response rate of AZD9291. From some preliminary data of response in the brain or leptomeningeal seeding with AZD 9291 in phase II AURA and BLOOM study, we hypothesized that H0:P= 10% and H1:P =30%. The BM cohort uses Simon's optimal two-stage design as follows. Stage 1: Treat n1 = 18 patients. If r1 = 2 or fewer patients respond, stop the BM cohort concluding that the study therapy is inefficacious. Otherwise, proceed to Stage 2. Stage 2: Treat additional n2 = 17 patients. If r = 6 or fewer patients respond among the cumulative n1 + n2 = 35 patients, then conclude that the study therapy is inefficacious. Otherwise, accept the study therapy for further investigation. This design has 1-sided alpha = 5% for P0 = 10% and power = 90% for P1 = 30%. Considering about 10%, of attrition due to ineligibility and dropout, a total of 40 patients will be enrolled into the BM cohort. 6. LM ± BM cohort The primary outcome of the LM ± BM cohort is overall survival (OS). Let μ denote the median OS of AZD9291 in this cohort. Based on some preliminary results (we need a reference), we hypothesize that H0: μ=3 months vs. H1: μ=5 months. Assuming an exponential OS distribution under H0, the investigators will conduct the 1-sample log-rank test. In this cohort, the investigators expect an accrual rate of 30 patients per year and about 10% of attrition, and will have an additional follow-up of 6 months after completion of accrual. Under these assumptions, we need N=40 patients (D=29 deaths) for 90% of power by the 1-sample log-rank test with 1-sided alpha=5% (refer to Kwak and Jung 2013). The trial of this cohort is expected to take about 22 months (=16 months for patient accrual + 6 months for additional follow-up). 7. Statistical analysis plan The analysis is conducted separately for each cohort. OS, PFS, and time to brain progression will be summarized by Kaplan-Meier method. DCR and AEs will be summarized using contingency tables. Cohort specific analyses are conducted as follows. BM cohort Two-stage testing on ORR will be conducted as described in "Target number" section. If the final sample size is different from n1=18 or n1 + n2=35, then a 1-sided p-value will be calculated by Jung et al. (2006) and accept the experimental therapy if it is smaller than alpha = 5%. The ORR will be unbiasedly estimated by Jung and Kim (2004), and its confidence interval will be calculated by Jennison and Turnbull (1983). LM ± BM cohort The final analysis of this cohort will be conducted when D29 deaths are observed, which is expected to be about 22 months from the study open. The investigators will conduct a 1-sample log-rank test assuming the exponential OS distribution with a median of 3 months under H0.