Figure 2.

(a) Forest plots of the meta-analysis showing the mean difference between responders and non-responders of the three pre-treatment inflammatory proteins (Olink data) that associate with ORR. For each of the two datasets (i.e., UK and NL), we report the number of responders (R) and non-responders (NR), the dataset weight in the random-effects meta-analysis, the mean protein difference (log₂-FC), along with its 95% CIs. Pooled mean effect sizes, with 95% CIs and p-values are also shown, as well as the statistics of Cochran tests for heterogeneity. (b) Survival curves in the 87 patients with advanced melanoma presenting with high (black line) and low (grey line) pre-treatment IL-6 levels, as measured using the Olink assay. Vertical lines indicate censored data points. Log-rank test hazard ratio (HR), along with the 95% CI and p-value, is shown. (c) Bar plot showing response rates in the 87 patients with advanced melanoma, stratified according to the number of inflammatory markers of response (i.e., IL-6, HGF, and MCP-2, as measured using the Luminex assay) being elevated (i.e., above the median) at baseline. (d) AUC-ROC curve showing ORR prediction accuracy, as estimated by training a LASSO logistic regression model using LOOCV in the UK sample and using the NL sample as validation set. Variables included in the LASSO model were age, sex, BMI, metastatic stage, IL-6, HGF, and MCP-2 (Olink data). AUC, area under the curve; LOOCV, leave one out cross-validation.