Estimating the distribution of external causes in hospital data from injury diagnosis.

Hospital discharge datasets are a key source for estimating the incidence of non-fatal injuries. While hospital records usually document injury diagnosis (e.g. traumatic brain injury, femur fracture, etc.) accurately, they often contain poor quality information on external causes (e.g. road traffic crashes, falls, fires, etc.), if such data is recorded at all. However, estimating incidence by external causes is essential for designing effective prevention strategies. Thus, we developed a method for estimating the number of hospital admissions due to each external cause based on injury diagnosis. We start with a prior probability distribution of external causes for each case (based on victim age and sex) and use Bayesian inference to update the probabilities based on the victim's injury diagnoses. We validate the method on a trial dataset in which both external causes and injury diagnoses are known and demonstrate application to two problems: redistribution of cases classified to ill-defined external causes in one hospital data system; and, estimation of external causes in another hospital data system that only records nature of injuries. In comparison with age-sex proportional distribution (the method usually employed), we found the Bayesian method to be a significant improvement for generating estimates of incidence for many external causes (e.g. fires, drownings, poisonings). But the method, performed poorly in distinguishing between falls and road traffic injuries, both of which are characterized by similar injury codes in our datasets. While such stop gap methods can help derive additional information, hospitals need to incorporate accurate external cause coding in routine record keeping.