The risk of dying : predicting trauma mortality in urban Indian hospitals

Sammanfattning: Introduction: With increased urbanisation and motorisation, trauma is emerging as one of the top threats to population health globally. Each year almost five million people die as a result of trauma, more than the total number of deaths from HIV/AIDS, tuberculosis, malaria, and maternal conditions combined. An overwhelming majority of these deaths occur in low- and middle-income countries, and almost 20% occur in India alone. In high income countries trauma mortality has been successfully reduced by extensive primary prevention and implementation of trauma systems. A systematic approach to prioritising patients according to needs is a crucial component of such systems. Several so called prediction models, i.e. often statistically derived algorithms to estimate the risk of mortality in an individual, have been developed to aid in this process. However, many available prediction models for trauma care have methodological limitations. Therefore, this research aimed to develop such a prediction model for an Indian trauma context. Methods: In study I [I] three cohorts of trauma patients admitted to a single centre in Mumbai were studied. Models were developed using multivariate logistic regression to assess the temporal trends in trauma mortality between 1998 and 2011. In study II [II] a prospective cohort study was conducted in three public university hospitals across urban India to derive vital signs based prediction models for early trauma mortality. Stepwise logistic regression was used to identify main effects. Then, in study III [III], the models derived in study II were validated temporally and compared with recently published prediction models. Validation was performed by applying the models in a temporally independent sample compared to study II. Finally, in study IV [IV] the transferability of vital signs-based prediction models was assessed by deriving models both in data from public university hospitals in India and data from the National Trauma Data Bank in the United States. Main findings: Analysis of 4189 trauma patients showed that early mortality significantly decreased between 1998 and 2011 [I]. Two models for predicting early mortality were derived using data from a prospective cohort of 1689 adult trauma patients admitted between October 2013 and January 2014. The first model included systolic blood pressure and Glasgow coma scale and the second model included systolic blood pressure, heart rate, and Glasgow coma scale [II]. These models were validated in a temporally independent sample of 2811 adult trauma patients. There was no evidence that comparatively more complex models had better predictive performance than the model with only systolic blood pressure and Glasgow coma scale [III]. Finally, when applied in data from another context, i.e. transferred, models from India overestimated the risk of early mortality in patients from the United States. In contrast, models from the United States underestimated the risk in patients from India. This miscalibration could be adjusted using updating methods in small samples [IV]. Conclusions: Between October 2013 and July 2014 early mortality was about 8% in adult trauma patients presenting to three public university hospitals in urban India. Substantial differences in systolic blood pressure and Glasgow coma scale between non-survivors and survivors indicate that haemorrhage and traumatic brain injury are major clinical issues. In prioritising trauma patients Indian clinicians and policy makers should consider to including vital-sign based decision support. The model based on systolic blood pressure and Glasgow coma scale presented here may help in this process. Finally, future validation studies of logistic prediction studies should explicitly include an updating component.