Purpose/Significance A prediction model is constructed based on real-world data to achieve prediction and early screening of type 2 diabetic microvascular complications. Method/Process Based on the real world data of Nanjing Drum Tower Hospital in the past 10 years, a particle swarm optimization based deep belief network (PSO-DBN) prediction model for microvascular complications in type 2 diabetes mellitus is constructed by taking test results and medical record documents into consideration. Result/Conclusion The PSO-DBN model can predict diabetic microvascular complications, and the performance is better than that of random forest and support vector machine (SVM) benchmark models, it provides references for the research of disease prediction model of real-world data.