Statistical Analysis

Objective response rates with corresponding 95% CIs were calculated using the Clopper-Pearson method. Duration of response, PFS, and OS were assessed by the Kaplan-Meier method, with 95% CI for median PFS and OS estimated using the Brookmeyer-Crowley method. Patients who survived at least 6 weeks were analyzed for OS to avoid immortal bias; 6 weeks was assumed to be sufficient for patients to respond to atezolizumab therapy.

The log-rank (Mantel-Cox) test was used to evaluate associations between biomarker subgroups and clinical activity (OS and PFS). Multivariate analysis was used to assess the effect of ICs and CD8-positive T cells on clinical outcome against prognostic factors as covariates. Pharmacodynamic changes of immune biomarkers were evaluated using linear mixed-effects models. P values were adjusted using Bonferroni correction to account for multiple time points. Due to the exploratory nature of the analysis, only nominal P values were reported.