Validity of death certificates for injury-related causes of death

Exploration of the validity of death certificate information for classifying underlying causes of death has historically focused on "natural" or disease-related causes of death. Current interest in injury-related deaths has emphasized the need for proper certification and coding of these deaths. In this study, the authors compared agreement of the underlying cause of death as determined from death certificate information with that determined from an independent review of all relevant medical and legal documents of death by a panel of physicians. The study sample included all deaths (n = 446) occurring over an approximately 18-year follow-up period (1965-1983) in a randomly selected cohort of 18,313 US Army Veterans of the Vietnam era. Using the physician panel as the "gold standard," sensitivity and specificity were 90% or greater for broad groupings of motor vehicle crash deaths (International Classification of Diseases, Ninth Revision (ICD-9), codes E810-E825), suicides (codes E950-E959), and homicides (codes E960-E969). Agreement for deaths from unintentional poisonings (codes E850-E869), mostly drug- and alcohol-related, was poor (sensitivity, 50 percent); in general, the ICD-9 drug- and alcohol-specific nomenclature is difficult to apply. The specificity and sensitivity for the individual three-digit suicide and homicide codes were all greater than 90%, and although the specificity for three-digit motor vehicle crash deaths was also above 90%, the sensitivity was lower, from 29% to 83%. Agreement on the fourth digit of ICD-9--for example, the role of the decedent in a motor vehicle crash death--was generally poor. The lack of descriptive information on death certificates to allow detailed coding was chiefly responsible for the poor agreement.