A novel tree-structured analysis for non-invasive diagnosis of gastric adenocarcinoma.

Non-invasive diagnosis of gastric adenocarcinoma (GAC) is usually difficult due to the low sensitivity and specificity of serologic markers,including pepsinogens and gastrin. For the improvement of the diagnostic values of these markers, a "recursive partitioning and amalgamation" algorithm was employed to construct a decision protocol. A total of 636 subjects including 161 healthy subjects, 163 patients with GAC, 196 with gastric ulcer and 116 with duodenal ulcer were enrolled. Serum levels of gastrin, pepsinogen I, pepsinogen II, and the ratio of pepsinogen I / pepsinogen II were determined for each of the subjects. The proposed "decision tree" classifies subjects into five subgroups with different risks of GAC and peptic ulcer, based on the information of age, serum pepsinogen and gastrin levels. Using this novel analysis system, an expected probability of GAC or ulcers could be obtained. Patients with an age > 62 years and a serum level of pepsinogen I < or = 33 ng/ml were strongly indicated for further confirmatory tests of GAC. This treestructured analysis is also helpful in clarifying the interactions between various serologic markers and demographic factors.