Malignant Potential of Gastrointestinal Cancers Assessed by Structural Equation Modeling

Parameters reported in pathologic reviews have been failing to assess exactly the malignant potential of gastrointestinal cancers. We hypothesized that malignant potential could be defined by common latent variables (hypothesis I), but there are substantial differences in the associations between malignant potential and pathologic parameters according to the origin of gastrointestinal cancers (hypothesis II). We shed light on these issues by structural equation modeling.

We conducted a cross-sectional survey of 217 esophageal, 192 gastric, and 175 colorectal cancer patients who consecutively underwent curative surgery for their pathologic stage I cancers at Keiyukai Sapporo Hospital. Latent variables identified by factor analysis and seven conventional pathologic parameters were introduced in the structural equation modeling analysis.

Because latent variables were disparate except for their number, 'three' in the examined gastrointestinal cancers, the first hypothesis was rejected. Because configural invariance across gastrointestinal cancers was not approved, the second hypothesis was verified. We could trace the three significant paths on the causal graph from latent variables to lymph node metastasis, which were mediated through depth, lymphatic invasion, and matrilysin expression in esophageal cancer, whereas only one significant path could be traced in both gastric and colorectal cancer. Two of the three latent variables were exogenous in esophageal cancer, whereas one factor was exogenous in the other gastrointestinal cancers. Cancer stemness promoted viability in esophageal cancer, but it was suppressed in others.

These results reflect the malignant potential of esophageal cancer is higher than that of the other gastrointestinal cancers. Such information might contribute to refining clinical treatments for gastrointestinal cancers.