In recent years we have witnessed the increasing adoption of clinical practice guidelines (CPGs) as decision support tools that guide medical treatment. As CPGs gain popularity, it has become evident that physicians frequently deviate from CPG recommendations, both erroneously and due to sound medical rationale. In this study we developed a methodology to computationally identify these deviation cases and understand their movitation. This was achieved using an integrated approach consisting of natural language processing, data modeling, and comparison methods to characterize deviations from CPG recommendations for 1431 adult soft tissue sarcoma patients. The results show that 48.9% of patient treatment programs deviate from CPG recommendations, with the largest deviation type being overtreatment, followed by differences in drug treatments. Interestingly, we identified over a dozen potential reasons for these deviations, with those directly related to the patients' cancer status being most abundant. These findings can be used to modify CPGs, increase adherence to CPG recommendations, reduce treatment cost, and potentially impact sarcoma care. Our approach can be applied to additional diseases that are subject to high deviation levels from CPGs.