OBJECTIVES: To evaluate multivariate predictors of CIMAvax-EGF therapeutic success using the causal inference approach.

METHODS: Univariate and multivariate causal inference analyses were performed retrospectively for patients with advanced NSCLC treated with CIMAvax-EGF or the best supportive care to evaluate the relationship between survival time and pretreatment potential predictive biomarkers, including basal serum EGF concentration, peripheral blood parameters and inmunocenescence biomarkers (The proportion of CD8 + CD28− T cells, CD4+ and CD8+ T cells, CD4/CD8 ratio and CD19+ B cells). The expected causal effect was evaluated via regression analysis testing the interaction between the possible predictors and treatment. A Likelihood Ratio Test was conducted to evaluate whether the interaction between biomarkers and treatment was significant.. The predictive causal association (PCA) was calculated for all possible models and the model with the highest value was selected. The probability of treatment success for individual patients was computed. All analysis were performed in R using EffectTreat library.

RESULTS: The mean of PCA was increase from 0.486, when only one predictor is considered, to 0.855 using the multivariate approach with the 10 predictors. However, already with 6 predictors the minimum values of PCA (min=0.727) exceeds the maximum value obtained for the best predictor in the univariate case (max=0.694). Using the basal EGF concentration, the CD4+/CD8+ ratio, the proportion of CD4+ T cells, CD19+ B cell, Neutrophils-to-Lymphocyte ratio and the absolute monocyte count as predictors, the PCA values were >0.72 in all realities".

CONCLUSIONS: The use of several predictors can help to improve our predicting capacity. The methodology, based on information theory and causal inference was useful for the evaluation of multivariate pretreatment predictors.