Background: Antiangiogenic agents, in particular monoclonal antibodies (mAbs) against VEGF, a major driver of tumor angiogenesis, are widely used in cancer therapy including metastatic colorectal cancer (mCRC). However, some patients do not profit from antiangiogenic treatments (AT), other patients benefit initially, but subsequently develop resistance not only to chemotherapy but also to AT. So far, no biomarkers are available to predict resistance to AT. Having an accurate assessment of imminent resistance to an AT may e.g. enable to respond by treating the patient with a more broadly acting antiangiogenic agent and thereby further delay resistance to the treatment and at the same time avoid employing a not anymore efficacious treatment. We hypothesized that repeated analysis of multiple cytokines related to angiogenesis together with machine learning approaches may enable an accurate prediction of anti-VEGF resistance during first-line treatment of mCRC patients with FOLFOX plus bevacizumab. The PERMAD trial aimed at establishing a CAF marker combination that enables the prediction of treatment resistance of patients with mCRC receiving Bevacizumab plus mFOLFOX6 in a palliative first-line setting about three months prior to radiological progress using an omics approach and bioinformatics. Methods: A phase I/II biomarker trial was conducted, including 15 centers in Germany and Austria. All mCRC patients included were treatment naïve and received FOLFOX plus Bevacizumab treatment. 102 different, preselected CAFs were prospectively collected and centrally analyzed in plasma samples (n = 647) obtained prior to treatment and biweekly until radiological progress determined by CT scan every 2 months. The values of CAFs affected in a similar fashion by both chemotherapy and disease progress were excluded. Using the remaining CAFs we employed a random forest predictor to define a combination of 5 CAF (CAF marker combination) whose change in values/pattern correlated with subsequent progress 3 months prior to radiological progress according to RECIST 1.1. Results: Using the samples described above and a random forest predictor we established a CAF marker combination comprising 5 CAF whose specific change in value/pattern over time indicated treatment resistance 3 months prior to radiological progress. The model allowed to differentiate timepoints without progress from timepoints predicting progress 100 days before radiological progress with an accuracy of 83%, a sensitivity of 76% and specificity of 88%. Conclusions: Using advanced bioinformatics, we identified a CAF marker combination that points out treatment resistance to FOLFOX plus Bevacizumab in patients with mCRC 3 months prior to radiological progress. Clinical trial information: NCT02331927