Objective Response Rate (ORR) Defined as the Proportion of Patients Who Have Achieved Complete Response (CR) or Partial Response (PR) According to Response Evaluation Criteria in Solid Tumors (RECIST) Version 1.1

Treatment (Pembrolizumab)

Patients receive pembrolizumab IV over 30 minutes on day 1. Treatment repeats every 21 days for up to 24 months in the absence of disease progression\* or unacceptable toxicity. \* NOTE: Patients with confirmed disease progression may continue to receive treatment if they are otherwise clinically stable until there is an increase in tumor burden of 25% or more following initial confirmation of progression. Under exceptional circumstances, and with protocol P.I. and CITN P.I. approval, patients may receive treatment beyond 2 years. Laboratory Biomarker Analysis: Ancillary studies Pembrolizumab: Given IV

Analysis denominator · Participants

1. group Description

   The protocol was designed w/ a standard Simon 2 stage design. Null hypothesis that true response rate is 5% was tested against a 1-sided alternative. In stage 1, 9 patients were accrued. If no responses in these 9 patients, study was to stop. Otherwise,15 more patients were to be accrued for a total of 24 patients. Study was amended to increase number of patients to 50, following discussion w/ pharmaceutical collaborator and FDA.

   group Ids

   1. OG000

   non Inferiority Comment

   The null hypothesis will be rejected if 3 or more responses are observed in 24 patients. This design yields a type I error rate of 0.10 and power of 0.90 when the true response rate is 25%. Progression-free survival (PFS) at 16 months will be estimated by Kaplan-Meier method. Unless otherwise stated, all statistical tests will be conducted at the α=0.05 (1-sided) level.

   non Inferiority Type

   OTHER

   other Analysis Description

   ORR will be estimated as the # of responders as a % of the # of eligible participants who received at least 1 dose of treatment. If a substantial amount of primary endpoint data are missing (at least 1 value missing from more than 20% of participants), using nonparametric estimation to estimate the ORR requires the missing completely at random assumption may give misleading results. In this case, analyses of the primary endpoint will be performed using parametric generalized linear models fit by maximum likelihood. These methods provide unbiased estimation and inferences under the parametric modeling assumptions and the assumption that the missing data are missing at random (MAR). MAR assumes that the probability of an observation being missing may depend upon the observed responses and upon observed covariates. A generalized linear model for the ORR will use a binomial error distribution. The model will include as covariates all available baseline predictors of the missing outcomes.