Prognostic impact of gastrointestinal bleeding and expression of PTEN and Ki-67 on primary gastrointestinal stromal tumors

BACKGROUND Prognostic indicators for gastrointestinal stromal tumors (GISTs) are under investigation. The latest risk classification criteria may still have room for improvement. This study aims to investigate prognostic factors for primary GISTs from three aspects, including clinicopathological parameters, immunohistochemical (IHC) expression of PTEN, and Ki-67 labeling index (LI), and attempts to find valuable predictors for the malignancy potential of primary GISTs. METHODS Tumor samples and clinicopathological data from 84 patients with primary GISTs after R0 resection were obtained. Immunohistochemical analysis was performed based on tissue microarray (TMA) to estimate expression of PTEN and Ki-67 in tumor cells. RESULTS The cut-off point of Ki-67 LI was determined as 1%, using a receiver operator characteristic test with a sensitivity of 71.7% and a specificity of 64.5%. Univariate analysis demonstrated the following factors as poor prognostic indicators for relapse-free survival (RFS) against a median follow-up of 40.25 months: gastrointestinal (GI) bleeding (P = 0.009), non-gastric tumor location (P = 0.001), large tumor size (P = 0.022), high mitotic index (P < 0.001), high cellularity (P = 0.012), tumor rupture (P = 0.013), absent or low expression of PTEN (P = 0.036), and Ki-67 LI >1% (P = 0.043). Gastrointestinal bleeding (hazard ratio, 3.85; 95% confidence interval, 1.63 to 9.10; P = 0.002) was a negative independent risk predictor in multivariate analysis, in addition to tumor size (P = 0.023), and mitotic index (P = 0.002). In addition, GI bleeding showed a good ability to predict recurrence potential, when included in our re-modified risk stratification criteria. CONCLUSIONS This study suggests that GI bleeding is an independent predictor of poor prognosis for RFS in primary GISTs. Expression of PTEN and Ki-67 are correlated with high risk potential and may predict early recurrence in univariate analysis.