The purpose of this paper is to compare early prediction of survival in patients following treatment for metastatic non-small cell lung cancer (NSCLC) using a support vector machine (SVM) paradigm, compared with standard logistic regression (LR). Retrospective blinded independent review by two board certified radiologist body imagers (observers 1 and 2) of the baseline and first post treatment CT scan in 26 patients with stages IIIB and IV NSCLC, consecutively identified from a phase 2 drug trial evaluating gemcitabine in combination with irinotecan, was performed. We show the Support Vector Machine (SVM) paradigm obtained results which consistently outperformed the standard Logistics Regression (LR) approach currently in clinical use. That is because of the highly non-linear nature of the problem, which led to the LR processing resulting in AUCs which were comparable to random guessing (AUC < 0.50) for the majority of experiments individually or as a combination of both board certified radiologists. Finally, the impact is that the proposed research and resultant algorithms will provide a computer aided risk assessment methodology and software packages which could be eventually used in the clinical environment to minimize the errors defined above. Consequently, cancer treatment could be administered more cost effectively by the eventual widespread use of these complex adaptive systems.