The Accuracy of Cause-of-Death Data
Errors in ascertaining true cause of death may arise from inaccurate diagnosis by the attending physician (if any), failure to perform an autopsy, inaccurate autopsy results, including inability of the pathologist to specify multiple causes of death, and failure of the death certifier to take into account the results of autopsy.
In the last case, problems arise from changing vital statistics conventions and the inevitable resulting misassignments of correctly diagnosed deaths. There is a large volume of postwar research on the accuracy of reported cause of death (Preston et al. 1972; Manton and Stallard 1984). This literature is useful not only for isolating principal themes in the misdiagnosis of cause of death, but also in setting upper bounds on the accuracy of prewar statistics.Prewar populations present special problems. Specific clinical problems of diagnosis and pathology are outside the scope of this essay, but several general points are important. First, because of intracategory misassignment (e.g., strokes due to cerebral hemorrhage being misassigned to strokes due to cerebral thrombosis), there are significant gains in accuracy from aggregating individual causes of death into broader categories (Preston et al. 1972). The more fine-grained the category, the higher the likely degree of error.
Second, any death wrongly excluded from the correct category, if it does not escape the registration system entirely, represents an incorrect addition to another category; problems of overattribution, in other words, may be as serious as problems of underattribution.
Third, as death has increasingly occurred at older ages from chronic and degenerative diseases, researchers have been forced to replace the traditional underlying-cause model, in which one disease or condition is assumed to have resulted in death, with more complex multiple cause-of-death approaches (Manton and Stallard 1984).
In the case Ofhistorical populations of low life expectancy, however, use of the simple underlying-cause model is more defensible than in the case of modem populations. Of course, it is frequently inappropriate for the case of aged persons, and the contributory role of chronic conditions such as nutritional deficiency and low- level diarrheal disease may be significant.Finally, because of the improving state of medical knowledge, increasing presence of an attending physician, and increasing frequency of competent autopsy, the overriding problem of the proportion of deaths attributed to unknown, ill-defined, and bizarre or nonsensical causes in cause-of-death statistics has been in steady decline (Preston et al. 1972). Among ill-defined causes, we include old age, senility, debility, fever, teething, dropsy, and so on. However, even in compilations as early as the 1850 U.S. Census of Mortality, the proportion of deaths assigned to bizarre causes is tiny, and attributions to unknown and ill-defined causes are confined largely to deaths of persons over 65.
Moreover, we are not totally ignorant of the underlying nature of many deaths assigned to unknown and ill-defined causes. There is an a priori reason to believe that historical cause-of-death data systematically underreport causes of death most common in old age because of the inherent difficulty of correctly diagnosing chronic and degenerative conditions. The most likely scenario is that these deaths ended up in the unknown and ill-defined categories, a hypothesis in support of which Preston (1976) has adduced strong statistical evidence. In particular, most deaths attributed to old age and senility can, with some confidence, be attributed to cardiovascular disease and cancer.
Thus, there are further compositional advantages, in addition to the enhanced validity of the underlying-cause mortality model, to dealing with low-life-expectancy historical populations. Calculations indicate that in a population with a life expectancy of 30, of the men 52.9 percent eventually die from causes falling into fairly robust categories: respiratory tuberculosis, other infectious and parasitic diseases, diarrheal disease, the influenzapneumonia-bronchitis complex, and violence. In a population with a life expectancy of 70, by contrast, the corresponding figure is only 14.6 percent (Preston et al. 1972; Preston 1976). Partially vitiating this compositional advantage of low-lifeexpectancy populations is the relatively large proportion of deaths from diseases of infancy and early childhood that are difficult to diagnose. Nevertheless, it appears that nineteenth- and early- twentieth-century mortality statistics give a fairly accurate picture (excluding problems related to chronic and degenerative disease) of the actual cause-of-death structure.
F. Landis MacKellar