Serum Monocyte Chemoattractant Protein-1 in Pancreatic Cancer

Figure 2

Recursive partitioning analysis was used to demonstrate the overall differences between demographic, clinical, and log MCP-1 characteristics of patients with PDA and patients with benign IPMN. Sensitivity and specificity of the resulting algorithm to determine cancer versus benign patients was completed using the standard definitions of these measures. Word at end of node demonstrates that the algorithm predicts either Benign or Cancer for the subgroup and graph then represents how many Benign/Cancer patients are in each subgroup. Standard logistic regression analysis utilizing log MCP-1 could not attain this level of sensitivity and specificity combined, rather it could achieve high values of one of these metrics at the expense of the other.