Background: No predictive models are currently available to predict poor prognosis in patients with severe heatstroke. We aimed to establish a predictive model to help clinicians identify the risk of death and customize individualized treatment. Methods: The medical records and data of 115 patients with severe heatstroke hospitalized in the intensive care unit of Changzhou No. 2 People's Hospital between June 2013 and September 2019 were retrospectively analyzed for modeling. Furthermore, data of 84 patients with severe heatstroke treated at Jintan No. 1 People's Hospital from June 2013 to 2021 were retrospectively analyzed for external verification of the model. We analyzed the hematological parameters of the patients recorded within 24 h of admission, which included routine blood tests, liver function, renal function, coagulation routine, and myocardial enzyme levels. Risk factors related to death in patients with severe heatstroke were screened using Least Absolute Shrinkage and Selection Operator regression. The independent variable risk ratio for death was investigated using the Cox univariate and multivariate regression analyses. The nomogram was subsequently used to establish a suitable prediction model. A receiver operating characteristic curve was drawn to evaluate the predictive power of the prediction model and the Acute Physiology and Chronic Health Evaluation (APACHE II) score. In addition, decision curve analysis was established to assess the clinical net benefit. The advantages and disadvantages of both models were evaluated using the integrated discrimination improvement and Net Reclassification Index. A calibration curve was constructed to assess predictive power and actual conditions. The external data sets were used to verify the predictive accuracy of the model. Results: All independent variables screened by Least Absolute Shrinkage and Selection Operator regression were independent risk factors for death in patients with severe heatstroke, which included neutrophil/lymphocyte ratio, platelet (PLT), troponin I, creatine kinase myocardial band, lactate dehydrogenase, human serum albumin, D-dimer, and APACHE-II scores. On days 10 and 30, the integrated discrimination improvement of the prediction model established was 0.311 and 0.364 times higher than that of the APACHE-II score, respectively; and the continuous Net Reclassification Index was 0.568 and 0.482 times higher than that of APACHE-II, respectively. Furthermore, we established that the area under the curve (AUC) of the prediction model was 0.905 and 0.918 on days 10 and 30, respectively. Decision curve analysis revealed that the AUC of this model was 7.67% and 10.67% on days 10 and 30, respectively. The calibration curve showed that the predicted conditions suitably fit the actual requirements. External data verification showed that the AUC on day 10 indicated by the prediction model was 0.908 (95% confidence interval, 82.2-99.4), and the AUC on day 30 was 0.930 (95% confidence interval, 0.860-0.999). Conclusion: The survival rate of patients with severe heatstroke within 24 h of admission on days 10 and 30 can be effectively predicted using a simple nomogram; additionally, this nomogram can be used to evaluate risks and make appropriate decisions in clinical settings.