【Objective】 To construct a prediction model of severe obstructive sleep apnea (OSA) risk in the general population by using nomogram in order to explore the independent risk factors of severe OSA and guide the early diagnosis and treatment. 【Methods】 We retrospectively enrolled patients who had been diagnosed by polysomnography and divided them into training and validation sets at the ratio of 7∶3. Patients were divided into severe OSA group and non-severe OSA group according to apnea hypopnea index (AHI)>30. Variables entering the model were identified by least absolute shrinkage and selection operator regression model (Lasso), and logistic regression (LR) method. Then, multivariable logistic regression analysis was used to establish the nomogram, and the area under the receiver operating characteristic curve (AUC) was used to evaluate the discriminative properties of the nomogram model. Finally, we conducted decision curve analysis (DCA) of nomogram model, STOP-Bang questionnaire and Berlin questionnaire to assess clinical utility. 【Results】 Through single factor and multiple factor logistic regression analyses, the independent risk factors for severe OSA were screened out, including moderate and severe sleepiness, family history of hypertension, history of smoking, drinking, snoring, history of suffocation, sedentary lifestyle, male, age, body mass index (BMI), waist and neck circumference. Lasso logistic regression identified smoke, suffocation time, snoring time, waistline, Epworth sleepiness scale (ESS) and BMI as predictive factors for inclusion in the nomogram. The AUC of the model was 0.795 [95% confidence interval (CI): 0.769-0.820] . Hosmer-Lemeshow test indicated that the model was well calibrated (χ²=3.942, P=0.862). The DCA results on the visual basis confirmed that the nomogram had superior overall net benefits within a wide, practical threshold probability range which displayed the nomogram was higher than that of STOP-Bang questionnaire and Berlin questionnaire, which is clinically useful. The Clinical Impact Curve (CIC) analysis showed the clinical effectiveness of the prediction model when the threshold probability was greater than 82% of the predicted score probability value. The prediction model determined that the high-risk population with severe OSA was highly matched with the actual population with severe OSA, which confirmed the high clinical effectiveness of the prediction model. 【Conclusion】 The model performed better than STOP-Bang questionnaire and Berlin questionnaire in predicting severe OSA and can be applied to screening. And it can be helpful to the early diagnosis and treatment of OSA in order to reduce social burden.