| Background:The incidence of traumatic brain injury(TBI)is increasing all over the world,which is the main cause of traumatic-associated death.In addition,neurological impairment in patients also leads to a lifelong disability,which is also a major contributor to worldwide social problems and family burden.In the context of mass lesion and/or increasing of intracranial pressure due to brain swelling and/or even the present of herniation during an early phase of injury,the absent of intracranial space-occupying lesion and a primary decompressive craniectomy(DC)appears to be the only life-saving surgical intervention.Although primary DC has saved many patients’ lives,the high rate of mortality and disability after operation is the dilemma of current treatment.Therefore,early prevention and prediction of poor prognosis for primary DC appears to be extremely critical.To date,there is still a lack of early predictive tools for the prognosis of TBI patients with primary DC.Objective:The purpose of this retrospective study was to investigate the clinical data of TBI patients undergoing primary DC and build two predictive models: 1)early predictive model for30-day mortality in patients after primary DC;2)early predictive model for overall prognosis in 6 months in patients after primary DC.The models provide a scientific screening tool for the high-risk population with poor prognosis of TBI patients undergoing primary DC,so that the predictive model can be easily and directly used in clinical practice.The models are expected to help doctors and family members to make clinical decisions,apply medical resources reasonably and reduce mortality rate.Method:In the first part of the study,we retrospectively reviewed the medical records of patients in one hospital who were performed primary DC from 2012 to 2019.In the current study,we defined 30-day mortality as the observed outcome in patients undergoing primary DC after TBI.Independent predictors of 30-day mortality were analyzed using multivariate logistic regression models.After developing multi-level predictive models on the basis of different predictors,the quality of the models,such as discrimination,calibration and clinical effectiveness were multi-dimensionally evaluated using computer programs which were written in R language.The predictive nomograms were further constructed.The internal validation data was invoked by R code to validate the repeatable of predictive models such as discrimination,calibration and clinical effectiveness.We enrolled the clinical data of patients in another hospital from 2016 to 2019 as external validation data sets,which was implemented into the models to observe the transportability of nomograms such as discrimination,calibration and clinical effectiveness.In the second part,we defined GOS1-2 as poor prognosis within 6 months after operation.Independent predictors of6-months poor prognosis were analyzed using multivariate logistic regression models.Furthermore,the early predictive models for the 6-months poor prognosis of the TBI patients after primary DC were constructed using R software.After evaluation the models,internal and external validation was performed to verify the efficiency of the models.Finally,the computer programming technology was used to develop a web page simulation and prediction calculator.Result:(1)Development,validation of early predictive models for 30-day mortality in TBI patients undergoing primary DC: The 30-day mortality rate was 30.8% in modeling data set.Pupillary(P=0.011),subdural hematoma(P=0.026),cistern(P=0.003),preoperative hypoxia(P=0.005),intra-operative hypotension(P=0.003),APTT(P=0.005),ISS(P=0.019)were the early independent predictors of 30-day mortality in TBI patients undergoing primary DC.We classify and combine these prediction factors and design three models using R language.The results show that the predictive model based on clinical features combined with imaging features,laboratory tests and operative data had the strongest predictive ability,which had the highest C-statistic[0.926(95% CI: 0.889-0.953)],calibration and clinical effectiveness.Based on these models,three predictive nomograms were constructed according to different situations.Internal and external validation of these nomograms was performed through R language,which showed a good discrimination ability,calibration and clinical effectiveness.(2)Development,validation of early predictive models for 6-month poor prognosis in TBI patients undergoing primary DC: The overall poor prognosis of patients in modeling data set within 6-month after primary DC was 45.6%.We observed that age(P=0.001),GCS(P<0.001),blood loss(P=0.045),cisterna(P<0.001),intraoperative hypotension(P=0.001),APTT(P=0.012)were the early independent predictors of the overall 6-month poor prognosis in TBI patients after primary DC.R software was used to build three predictive models based on clinical features,image features,laboratory tests,and operation-related indicators.Three predictive nomograms were built.After internal and external validation,these models showed a good discrimination,calibration and clinical effective ability.Conclusion:We constructed early predictive models based on the independent predictors of 30-day mortality and 6-month overall poor prognosis.The internal and external validation of the models was excellent.The predictive models can be visualized and operated in the way of nomograms.We also published a popularized web page predictive tool,which can be directly,simply applied to clinical practice.The early predictive models for the poor prognosis in TBI patients undergoing DC could assistant doctors and patients’ relatives to make clinical decisions.It could also make medical resources distribution more reasonable,that be used as a reference for individualized treatment. |
PDF Full Text Request