9045 Background: CRC patients undergoing CT are likely to experience multiple concurrent toxicities. Rather than appearing singularly, the hypothesis that certain toxicities occur in clusters may suggest a common pathobiology. We used Markov networks (MN), a probabilistic graphical born at the confluence of statistics and artificial intelligence describing the dependency among set of variables, to identify clusters of CT-induced toxicities, to examine how clusters are connected to each other, and how single toxicities are related to a specific cluster. Methods: Using a standardized data collection tool, we retrospectively reviewed electronic medical charts of 300 consecutive CRC pts receiving FOLFOX, FOLFIRI or 5-FU to record baseline demographic and clinical information. Toxicities were recorded using NCI-CTC criteria during the first cycle of CT. Following the standard Bayesian approach, the MN clustering the CT-induced toxicities was learned from the data as the network with the highest posterior probability given the data. Results: The network, in which associations between toxicities are represented as links, identified five strongly-related symptom clusters: a constitutional cluster involving fatigue, anorexia, and weight loss; a gastrointestinal cluster where dehydration was the connector between diarrhea, constipation and bloating on a side and taste nausea and vomiting, taste alteration, fever and chills on the flipside; a dermatological cluster composed by dry skin, HFS, rash and itching and connected with hemorrhage/bleeding and wound complication toxicity. Furthermore, we noticed strong connections between cough, dyspnea and infection, with palpitation and pain and we detected another cluster where depression and anxiety where connected with cystitis. Conclusions: The application of network analyses to define CT-induced toxicity clusters is new. The technique was effective in defining the relationships between individual toxicities associated with cycle 1 therapy. The lack of randomness between the relationships defined by the network provides a strong suggestion that each cluster shares a common pathobiological basis, which may provide an opportunity for intervention. [Table: see text]