Background: Microsatellite stability (MSS) tumors hardly benefit from immunotherapies and are more probable to occur postoperative recurrence. However, some studies have revealed that a subset of MSS patients harbor “hot immune microenvironment” tumors, indicating high heterogeneity in such wide range of patient population. On the other hand, researches of mechanism of MSI formation found potential similarities in endometrial and gastrointestinal tumors. We hypothesized the transcriptomic features in these cancers correlated with immune-related signatures and patients’ prognosis. Methods: Early stage (I-III stage) MSS tumors, including endometrial, colorectal, and gastric from TCGA project were analyzed as training cohort (n=170). A combined cohort consisting of 604 colorectal and stomach cancers from GEO datasets (GSE39582，GSE62254) was validation cohort. The RNA-Seq profiling data and disease-free survival (DFS) data of patients were collected. Cibersort tool was used to evaluate twenty-two immune cells’ enrichment. The prediction model was developed by three steps: Univariate cox regression of DFS was conducted to select 9 immune cells. Then the train cohort was divided into two groups based on non-negative matrix factorization (NMF) method using this 9 immune cell features. Differentially expressed genes of these two groups were identified and screened further by lasso regression. Log-rank test was used to evaluate the difference of DFS. Results: A six-gene lasso-cox model was developed. The genes were LYZ, WFDC2, CAPS, RHPN1, TFF2 and TGFBR2. Based on the score evaluated by this model, patients in training cohort were divided into high-risk and low-risk groups. Low-risk population had much longer DFS (HR 0.07, 95%CI 0.03-0.18, p<0.001). In validation cohort, lower risk score was also verified to be associated with a lower likelihood of recurrence (HR 0.66, 95%CI 0.5-0.88, p=0.0047). Conclusions: We developed a model of six-genes predicting disease-free survival based on RNA-Seq data in early stage MSS patients. Further validation was needed to implement in larger clinical cohorts.