Predictors of 1 year mortality in adult injured patients admitted to the trauma center

BACKGROUND: Traditional approach to predicting trauma-related mortality utilizes scores based on anatomical, physiological, or a combination of both types of criteria. However, several factors are reported in literature to predict mortality independent of severity scores. The objectives of the study were to identify predictors of 1 year mortality and determine their magnitude and significance of association in a resource constrained scenario. MATERIALS AND METHODS: Prospective observational study enrolled 572 patients. Information regarding factors known to affect mortality was recorded. Other factors which may be important in resource constrained settings were also included. This included referral from a peripheral hospital, number of surgeries performed on the patient, and his socioeconomic status (below poverty line (BPL) card). Patients were followed till death or upto a period of 1year. Logistic regression, actuarial survival analysis, and Cox proportionate hazard model were used to identify predictors of 1year mortality. Limited estimate of external validity of the study was obtained using bootstrapping. RESULTS: Age of patient, Injury Severity Score (ISS), abnormal activated partial thromboplastin time (APTT), Glasgow Coma Scale (GCS) score at admission, and systolic blood pressure (BP) at admission were found to significantly predict mortality on logistic regression and Cox proportionate hazard models. Abnormal respiratory rate at admission was found to significantly predict mortality in the logistic regression model, but no such association was seen in Cox proportionate hazard model. Bootstrapping of the logistic regression model and Cox proportionate hazard model provide us with a set of factors common to both the models. These were age, ISS, APTT, and GCS score at admission. CONCLUSION: Multivariate analysis (logistic and Cox proportionate hazard analysis) and subsequent bootstrapping provide us with a set of factors which may be considered as valid predictors universally. However, since bootstrapping only provides limited estimates of external validity, there is a need to test these factors against the well accepted requirements of external validity namely population, ecological, and temporal validity.