A random forest model to predict the efficacy of anti-PD-1/PD-L1 monoclonal antibody in lung adenocarcinoma.

Abstract

e20558 Background: Not all patients with lung adenocarcinoma (LUAD) could benefit from anti-PD-1/PD-L1 monoclonal antibody therapy, and existing biomarkers cannot explain all clinical benefits. Therefore, an effective model to predict the efficacy of anti-PD-1/PD-L1 monoclonal antibody is necessary. Methods: DNA methylation sequencing was used in LUAD patients’ FFPE samples, who received anti-PD-1/PD-L1 monoclonal antibody. The DNA methylation level of each gene was transformed into binary format according to its median value. The progression-free survival (PFS) was defined as the duration from the beginning of anti-PD-1/PD-L1 monoclonal antibody therapy to disease progression or death. Patients were randomly divided into training cohort (70%) and test cohort (30%). In the training cohort, uni-Cox regression was used to estimate the value in predicting PFS, and LASSO-Cox regression was performed to screened genes to build random forest model. The accuracy of this random forest model was validated in the test cohort. Finally, DNA methylation data were used to calculated the risk score of patients in the TCGA-LUAD cohort, and GSEA and GSVA were employed in RNA data to explore the mechanism and explain the efficacy of the prediction model. Results: In this study, 40 patients with LUAD were enrolled, and the median PFS of them was 7.57 months. A total of 29 genes with P < 0.05 and C-index > 0.7 were selected to perform LASSO-Cox regression, and 13 genes ( BTBD7, CAD, FBXL20, FKBP7, HBP1, IGFBP3, RAB5B, RASGRP1, RNF8, SOHLH1, TMEM223, UHRF1 and WWC2) were used to build prediction model. The C-index of risk score from this model was 0.96 and 0.93 in the training and test cohort, respectively. Patients were divided into the high and low risk groups according to the median risk score, and the median PFS of the high risk group was shorter than that of the low risk group (training cohort: 3.05 vs. 10.89 months, P < 0.01, HR = 2.51, 95% CI: 1.50-4.22; test cohort: 1.64 vs. 11.31 months, P = 0.03, HR = 1.80, 95% CI: 1.04-3.14). In the low and high risk group of the TCGA-LUAD cohort, there were 92.2% (388/421) and 0.5% (2/421) significant enriched ( P < 0.05 and FDR < 0.25) immunity-related pathways from results of GSEA, respectively, meanwhile 94.3% (397/421) and 1.9% (8/421) significant enriched immunity-related pathways from results of GSVA, respectively. Therefore, the lower risk score of this random forest model accompanied with the higher activation of immunity-related pathways which might help anti-PD-1/PD-L1 monoclonal antibody to express better efficacy. Conclusions: A random forest model based on DNA methylation data was constructed and showed a potential value in predicting the PFS of anti-PD-1/PD-L1 monoclonal antibody therapy in LUADs, and the risk score of this model is negatively correlated to the activation of immunity-related pathways, which could help to identify LUAD patients that would benefit from immunotherapy.