Objective: To investigate the relationship between immune indexes and the incidence of hepatocellular carcinoma in cirrhotic patients with abnormal a-fetoprotein, and to establish a new prediction model for hepatocellular carcinoma occurrence. Methods: 241 patients with cirrhosis with abnormal a-fetoprotein were retrospectively collected from Beijing Ditan Hospital Affiliated to Capital Medical University from January 2017 to January 2018. The clinical data of the patients were analyzed, and the 5-year hepatocellular carcinoma incidence rate of cirrhosis patients with abnormal a-fetoprotein was analyzed. A total of 241 patients were randomly divided into a training set and a validation set according to 3∶1. Independent variables were screened through univariate and multivariate analysis. Multivariate logistic regression was used to construct a prediction model, and ROC curve was drawn to evaluate the predictive value of the model for hepatocellular carcinoma incidence in patients with cirrhosis with abnormal a-fetoprotein. Results: The 5-year incidence of hepatocellular carcinoma in cirrhotic patients with abnormal a-fetoprotein was 41.9%. Age, monocyte count, alkaline phosphatase, prothrombin time, CD4⁺ T cell count and CD8⁺ T cell count were screened as clinical predictors with significant differences, and a multivariate logistic regression prediction model was constructed based on these 6 clinical characteristics. The ROC curve area under the 5-year prediction model was 0.89(95% CI: 0.94-0.84), and the prediction value was significantly better than that of the 1-year and 3-year prediction value. Conclusion: The 5-year incidence of hepatocellular carcinoma in cirrhotic patients with abnormal A-fetoprotein is higher, among which age increased monocyte count, increased alkaline phosphatase and prothrombin time are risk factors, and increased CD4⁺ T cell count and CD8⁺ T cell count are protective factors. The prediction model of hepatocellular carcinoma including CD4⁺ T cell count and CD8⁺ T cell count has good predictive value for hepatocellular carcinoma occurrence, and its long-term prediction efficiency is better than short-term prediction efficiency. In clinical practice, the changes of these 6 indexes in the prediction model can be dynamically monitored to provide a basis for clinical treatment.