Traumatic Brain Injuries-Characteristics And Outcome Prediction

Objectives: Traumatic brain injury (TBI) is the leading cause of disability andmortality, and motor vehicle crashes (MCVs) is the leading reason of TBI. TBIsresulted from MVCs tend to be more severe, less probability of good outcome, andlonger hospital stay. To know the main characteristics of paediatric TBIs in US, andthe difference between China and US, is important for TBI prevention in China. Based on simple diagnosis and injury history at admission, predicting the outcomeand length of stay(LOS), are important for patients, his/her family, and for their doctor.This study were to explore the number, rate, key epidemiological distribution of TBI;how to construct, evaluate, and use outcome prediction model and length of stayprediction model.Methods: (1) Part one: descriptive methods and survey data analysis methods wereused. Data from the 2006 Healthcare Cost and Utilization Project Kids'InpatientDatabase (KID) were used to report the national estimates, rates, and characteristics ofTBI-associated hospitalizations caused by motor vehicle crashes and falls among U.S.children≤20 years of age. Compare of these characteristics between China and USwere explored. (2) Part two: Data were collected from those medical records of 1294hospitalized patients of TBI resulted from road traffic accidents. Multiple Logisticregression model was used to screening those independent variables statisticallysignificant to prognosis. The fitted model were evaluated as respect to global nullhypothesis, and goodness-of-fit, with likelihood ratio test and Hosmer andLemeshow Goodness-of-Fit test, and AUC respectively, finally, by computing areduced-bias estimate of the predicted probability, the predicting power of the fittedmodel was checked. (3) Part three: The LOS of 1162 nonfatal TBI was analyzed withKM and Cox proportional hazard regression model.Results: (1)In 2006, there were an estimated 43,691 TBI-associated hospitalizationsamong children in the U.S., accounting for $1.75 billion in total hospital charges.Motor vehicle crashes and falls were the top two external causes of TBI: causing34.4% and 25.8% of TBI injuries respectively. For TBIs caused by falls, rates werehighest among infants＜1 year of age. Rates of hospitalized TBI from motor vehiclecrashes increased significantly after 15 years of age in both males and females.Overall, TBIs resulting from motor vehicle crashes were more severe than TBIscaused by falls. There were several difference between chinese children and theircounterpart of US chindren on TBI characteristics resulted by traffic accidents,such as Chinese Children TBI tended to be pedestrain,and stayed longer in hospital. (2) Aftercontrolling for injury severity, age and coma history were the most influencialpredictors of bad outcomes. The probability of bad outcome (death or not improved)for elder patients (age＞=55) was 3.7 folds of those younger patients (age＜25). ThoseTBI with coma history longer than 1 hour, the OR of bad outcome would be 6.7 timeshigher than those no coma. We did not find significant effect of gender, general healthlevel before injury, the type of the accidents, the time between accident and admission,and who will pay the hospital bill. Hosmer and Lemeshow Goodness-of-Fit test(Chi-square 0.1553, P＞0.9253), show the model fit the data well. The area under ROCwas 0.83, so the power of prediction of the model was good. (3) The distribution ofLOS was positively skewed and censored values accounting for 61%, indicatingtraditional multiple linear regression models are not applicable in this case. Theestimated median LOS was 29 days. In addition to the injury severity, age and the roleof injured in the accidents had significant influence on LOS.Conclusions: (1)Findings regarding top TBI causes suggest the need to focus on fallprevention among infants＜1 year of age and motor vehicle safety among childrenafter 15 years of age. These findings have important implications for childhood TBIresearch and prevention programs, especially in developing countries whereintervention programs are lacking.(2) Age and injury severity were the strongestpredictors. The model with age and injury severity as independent variables showedmoderate predicting power. It has the potential to be clinically useful. (3) Forpredicting LOS, distribution and censored values should be considered, and survivalmethods may applicable in most cases.