Feature Deconstruction Strategy Based Identification of Acute Features Associated With Impairment of Activities of Daily Living in Patients With Traumatic Brain Injury / 中山大学学报(医学科学版)

ObjectiveTo identify acute phase features associated with the prognosis of traumatic brain injury （TBI）. MethodsThrough two traditional strategies， correlation analysis and prediction model， and one innovative research strategy based on feature deconstruction， a retrospective analysis was conducted using demographic， acute phase and chronic phase features of 354 TBI patients to identify acute phase features associated with activities of daily living （ADL） in chronic phase of TBI. For feature deconstruction strategy， the LASSO （Least Absolute Shrinkage and Selection Operator） algorithm was used to build a prediction model that could effectively predict ADL based on non-ADL chronic phase features. The model could indicate the key chronic phase dimensions determining the ADL in TBI patients. We then identified demographic and acute phase variables that were significantly associated with these key chronic phase features. ResultsThe feature deconstruction strategy revealed that ADL could be deconstructed into chronic phase dimensions such as weak limbs in TBI population. Importantly， to the best of our knowledge， this strategy revealed for the first time the association of these important acute phase features with specific chronic phase impairment features. For example， TBI patients had a higher risk for chronic phase recent memory impairment if they had a prolonged coma time and low GCS scores at acute phase ［scaled coma time OR95%CI = 94.288 （35.095， 273.231）； scaled GCS OR95%CI = 0.068 （0.030， 0.147）］； the patients had a higher risk for insight impairment and disorientation at chronic phase if they had hydrocephalus at acute phase ［insight impairment OR95%CI = 6.760 （3.653，12.855） ； disorientation OR95%CI = 6.538 （3.530， 12.490）］. All strategies showed that the strongest risk factors for ADL damage in the chronic phase included prolonged coma time and low GCS scores as well as hydrocephalus. ConclusionThis study provides an innovative research strategy to establish the association between acute injury features and chronic recovery features， and to identify demographic and acute phase features associated with the prognosis of TBI.