Background: The use of large national databases for clinical research has increased recently in the field of trauma care as they allow study of rare events without the logistical difficulties of a prospective study. However, many of these databases use administrative billing codes, such as International Classification of Disease-9th Rev. (ICD-9) codes, to identify preexisting patient comorbidities. While the accuracy of these billing codes for research purposes has previously been called into question, this has not been studied in a trauma population.
Methods: All patients with proximal tibia fractures in the 2011 and 2012 American College of Surgeons' National Trauma Data Bank were reviewed. Rates of 12 individual comorbidities in this population were computed using both ICD-9 diagnosis codes and also National Trauma Data Bank chart-abstracted variables. The sensitivity was computed for ICD-9 coding of each comorbidity taking chart-abstracted data elements as criterion standard. With the use of multivariate logistic regression, controlling for age and Injury Severity Score (ISS), the odds ratio for mortality was computed for each comorbidity, using both ICD-9 diagnoses and chart-abstracted diagnoses.
Results: A total of 32,441 patients with proximal tibia fractures were identified. The sensitivities of ICD-9 billing codes for the comorbidities analyzed ranged from 18.8% for previous myocardial infarction to 2.4% for alcoholism. In individual multivariate analyses of each comorbidity, there was a change in the statistical significance of the odds ratio for mortality for 6 of 12 comorbidities analyzed.
Conclusion: Researchers and those evaluating research in the field of trauma should carefully consider the accuracy of data elements in future studies, especially ICD-9-coded comorbidity diagnoses.