Fig. 1.

A Study overview. a Consort diagram of patients with metastatic cancer treated with ICI. b The pipeline for analysis of patient samples. B In the ICI cohort, of the 65 patients who achieved SD at the first scan, approximately 40% (n = 26) did not ultimately reach DCB. C The five common model indices, including AIC (Akaike’s Information Criterion), BIC (Bayesian Information Criterion), R2 (r-squared value), R2_adj (adjusted r-squared), and RMSE (root-mean-squared error) are normalized, and the Lasso model indicates better model performance than the ASR and Cox regression models. The weighted residuals of Cox Proportional Hazards (CoxPH) cannot be calculated, so we couldn't obtain the residual standard deviation (SIGMA) in CoxPH. (D) Time-dependent ROC curve at 1-year OS for the ASR, Lasso, univariate Cox regression in the training cohort (ASR AUC = 0.763, Lasso AUC = 0.796, and univariate Cox regression AUC = 0.792). (E) Time-dependent ROC curve at 2-year OS for the ASR, Lasso, and univariate Cox regression in the training cohort (ASR AUC = 0.681, Lasso AUC = 0.783, and univariate Cox regression AUC = 0.755). ICIs: immune checkpoint inhibitors; BICS: peripheral-blood-immune-cell-based signature; OS: overall survival; DCB: durable clinical outcomes; NDB: no durable benefit; NDB_LS: long-term survival with no durable clinical benefit; NDB_SS: short-term survival with no durable clinical benefit; PD: progressive disease; PR: partial response; SD: stable disease; LASSO: least absolute shrinkage and selection operator; ASR: all-subsets regression; AUC: area under the curve; ROC: receiver operating characteristic