cal documents for each case of death were used (medical files from hospitals and ambulatories and the reports from autopsy experts).
The results of these surveys form a valuable database. Unfortunately, the materials used are very poorly described. Not even the total numbers of deaths in each survey sample are available. For Bedniy's study, we do not know from what region of Russia the certificates were collected. There is no information about age or sex.
In spite of these deficiencies, one can make some important observations based on these surveys. Generally, two types of results are available-the percentage of errors in medical diagnoses and the percentage of errors in coding. The survey findings indicated that the total percentage of diagnosis errors ranged from 6.6 percent (Minsk, 1981-1982) to 12.7 percent (unknown places in Russia, 1979), and the percentage of coding errors varied from 4.1 percent (Minsk, 1981-1982) to 17.7 percent (Tula and other sites, 1960). For a majority of causes of death, the proportion of coding errors was larger than the proportion of diagnosis errors. The percentage of errors was lower for neoplasms and for accidents and violence, and was much higher for hypertensive diseases, cerebrovascular disorders, ischemic heart disease, and respiratory and digestive diseases.
Table 2-2 shows the differences between real and registered causes of death resulting from the combination of the above two types of error. In many cases, diagnosis and coding errors compensate each other. The table shows that there are tendencies toward overregistration of deaths from cerebrovascular disorders, atherosclerotic heart diseases, and respiratory diseases, and of underregistration of deaths from cancer, hypertensive diseases, and myocardial infarction. Yet without knowing the absolute numbers, we cannot evaluate the extent to which these opposite tendencies balance each other.
To check the coherence of the set of ratios shown in Table 2-2, we applied these ratios to the cause-specific death figures for Russia (1960 and 1979), Belarus (1981-1982), and Turkmenistan (1981-1982). We found no conflicts between the total number of deaths from all causes before and after correction, which gives a good indication of the reliability of the results of these surveys.
One of the most interesting features of the results is that they do not support the widespread opinion about overregistration of cardiovascular mortality. Indeed, rather large errors observed for different circulatory diseases compensate each other, and the percentage of error in the total of circulatory diseases is rather small. Only in the Bystrova study was a small overregistration of total cardiovascular mortality observed, while the other two studies indicated underregistration of 2 to 3 percent. This fact contradicts, among others, the point of view expressed by Belenkov et al. (1987), who state that there is a substantial overregistration of cardiovascular deaths, especially among the elderly. The results of the surveys do not allow us to analyze differences by age more precisely. But looking at the estimation of global coverage of cardiovascular diseases (Table 2-2), we can