Construction and verification of risk prediction model for early death in severe brain injury patients based on ultrasound combined with biomarkers

Wu Qingsong; Deng Qingjun

doi:10.3969/j.issn.1002-1949.2024.10.003

Abstract：Objective To construct a risk prediction model for early death(within 30 days of onset)of patients with severe brain injury.Methods This study was a prospective observational study.The relevant data of 139 patients with severe brain injury admitted to Chongqing Red Cross Hospital(Jiangbei District People's Hospital)from June 2022 to May 2023 were collected as a training set,a prediction model for predicting the mortality was established with Logistic regression,and the model's discrimination,calibration and clinical applicability were verified.Finally,the model results were presented by a nomogram.31 patients with severe brain injury from June 2023 to October 2023 were collected as a verification set for the external verification of prediction model.Results A total of 50 out of 139 patients with severe brain injury died during hospitalization with mortality rate of 35.97％.Univariate analysis showed that tracheal intubation status,light reflex status,cerebral hernia formation,coagulation disorders,Glasgow coma scale(GCS),mean arterial pressure,the acute physiology and chronic health evaluation Ⅱ(APACHE Ⅱ),platelet count,neuron-specific enolase(NSE),pulsatility index(PI)and optic nerve sheath diameter(ONSD)at admission were risk factors for early death in the patients with severe brain injury.Multivariate Logistic regression analysis showed that the formula for predicting the probability of death was:Logit(P)=-14.266+0.14 ×(APACHE Ⅱscore)+0.047×(NSE)+1.57×(PI)+0.908×(ONSD)+2.375×(coagulation status).The area under curve(AUC)of risk prediction model predicting early death in the patients with severe brain injury was 0.886(P＜0.01),and its sensitivity and specificity were 76.00％and 91.00％.The external verification results showed the AUC of prediction model was 0.810 with the sensitivity of 75.00％and the specificity of 87.00％.Conclusions This risk prediction model based on APACHE Ⅱ,NSE,PI,ONSD and coagulation disorders has high effectiveness in evaluating early death of the patients with severe brain injury.