Underregistration and Alternative Definitions of Neonatal Deaths

The Soviet definition of infant mortality is not the same as the World Health Organization (WHO) standard (see also the chapters by Shkolnikov et al. and by Kingkade and Arriaga, in this volume). As a consequence, the number of neonatal deaths—deaths before age 1 month—in the former Soviet republics is seriously underreported. The result is an underestimate of infant mortality, which is the sum of neonatal and post-neonatal mortality. In the present analysis, we correct neonatal mortality rates (NMR) for the former Soviet republics by using the relationship between NMR and post-neonatal mortality rates (PNMR) observed in countries with good vital statistics. We expect the PNMR (deaths between ages 1 and 12 months) to be unaffected by the Soviet definition of an infant death, except for possible age heaping at 1 month of age.

We analyzed 1,327 pairs of NMR and PNMR available for 35 countries over a 40-year period.² Figure 6-1 shows the relationship between the two rates. When the PNMR is transformed to its logit form,³ the relationship is linear. An ordinary least squares regression equation was fitted:

Figure 6-1

Neonatal and post-neonatal mortality rates. (Data over a 40-year period from countries with good vital statistics registration.) These 1,327 data points are drawn from 35 countries over a period from 1945 to 1989. The R² from the regression of NMR vs. (more...)

NMR = .0555 + (-.0166 * logitPNMR)

The R² for the equation is 0.80, and the p-values for the slope and constant are each less than 0.001. The standard error of the constant is 0.0002, and the standard error for the slope is 0.0006. In addition, the residuals are homoskedastic.

We estimated the corrected neonatal mortality rates (NMRc) by applying the PNMR for each of the former Soviet republics (thought to be accurate) to the above regression equation. Adding this newly generated NMRc to the PNMR yields a new estimate of the infant mortality rate, IMRc. It should be noted that all of the correction made to the infant mortality rates is due to correction of the neonatal mortality rate (from NMR to NMRc). Table 6-1 shows the NMRc and IMRc for all the former Soviet republics and the estimated proportion of underregistered neonatal and infant deaths; for NMR, this percentage varies from 26.6 percent in Turkmenistan to about 53.5 percent in Latvia. The different definition of neonatal death may not account for all of this wide variation in the neonatal mortality rate; some variation may be due to higher rates of underreporting of neonatal deaths.

TABLE 6-1

Infant Mortality Reported and Corrected in the NIS, 1990.

Underregistration of Adult Deaths

Most authors presume that registration of mortality in the Russian Federation, the Baltic states, and the other former Northern republics is complete. Anderson and Silver (1990, 1991, and in this volume), however, have analyzed regional mortality patterns in the former Soviet Union and concluded that there is substantial underregistration of adult deaths in the former Central Asian republics. To date, judgments that there has been substantial underreporting of deaths in certain republics have been based solely on the fact that observed mortality rates appear to be too low. Such assessments presuppose that the determinants of relative levels of adult mortality within the former Soviet Union or among industrialized countries are known. For example, Anderson and Silver (1990, 1991) report lower age-specific mortality in Tajikistan than in the United States for males; in the age groups over 70 years, the differences are as high as 20 to 50 percent. The authors conclude that lower adult mortality in Central Asia than in the United States is "implausible," although they provide no epidemiological justification for this judgment.

Studies of adult mortality patterns (ages 15-59) in industrialized and developing countries have demonstrated wide variations in adult male and female mortality as measured by ₄₅q₁₅—the probability of death between ages 15 and 60. For example, in Japan, male ₄₅q₁₅ is 113 per 1,000, compared with 175 per 1,000 for all U.S. males, 300 for U.S. black males, and 187 for Finnish males (Murray et al., 1992). Given the wide range in adult mortality levels that is not easily explained by variables such as income per capita, it is not convincing to argue that there is significant underregistration in Central Asian states solely because their observed rates are lower than those of other states.

To define further the extent of underregistration in different states of the former Soviet Union, we use the growth balance method and the Bennett-Horiuchi technique (United Nations, 1983). We apply the growth balance method using registered deaths in 1989 and the census population for 1989 by age for each republic. Application of this method depends on having a population that approximates a stable population with a long-term constant birth rate and no net migration. The relationship between Nx/Nx+ and Dx+/Nx+, however, is not linear for almost all republics; Nx is the population at age x, Nx is the population over age x, and Dx+ is deaths over age x. The age group "birth rate," Nx/Nx+, is markedly lower for the age groups 70-74 and 75-79 for most republics, which may reflect the World War II experience of this cohort. Excluding these age groups and 80+ years, the estimated coverage for the former Soviet Union combined is 103 percent for females and 102 percent for males. Estimates of coverage using the growth balance method for each republic range from 65 to 120 percent, as shown in Table 6-2. These estimates follow no clear geographic pattern; Armenia, Georgia, Belarus, and Moldova have the lowest estimated coverage, rates below 70 percent. The assumptions underlying the growth balance method clearly do not hold at the republic level, making these estimates of coverage.⁴

TABLE 6-2

Estimated Coverage of Mortality Registration in the New Independent States, 1990.

The Bennett-Horiuchi technique for assessing vital registration completeness is a more powerful method that does not require assumption of a constant birth rate over the past 80 years, but does assume a closed population (Bennett and Horiuchi, 1981). As input, two censuses and all registered deaths by age and sex for the interval between the censuses are required. Censuses were conducted in each republic in 1979 and 1989; unfortunately, registered deaths by age and sex are available for the majority of years between 1979 and 1989, but not all. As a first approximation, we used the average number of registered deaths for each age group for all available years 1979-1989, multiplied by 10. For the former Soviet Union combined, the estimated completeness of registration for females is 99 percent and for males 102 percent. The estimated coverage may be somewhat exaggerated (over 100 percent, for example) because of overstatement of age at older ages (Bennett and Garson, 1983).

Table 6-2 provides the estimated coverage of death registration for each republic by sex. The median estimated completeness is severely affected by internal migration; those republics, such as Lithuania, which had substantial net immigration over the period 1979 to 1989 show overregistration of deaths, while those with net emigration show underregistration. The third column of Table 6-2 shows the estimated completeness of death registration for the population over age 50, which may be less affected by migration between republics. To the extent that the approximations used in the application of the Bennett-Horiuchi technique are plausible, registration is over 90 percent in all locations except for males in Uzbekistan, Kyrgyz, and Tajikistan and females in Azerbaijan, Kazakstan, Turkmenistan, Armenia, Uzbekistan, Kyrgyz, and Tajikistan.

The lower levels of vital registration coverage for many of the Central Asian republics and females in Azerbaijan are probably due to a combination of net emigration and lower completeness of vital registration. Given that vital registration for the Soviet Union as a whole is very close to complete, we suspect that internal migration in the former Soviet republics may play an important role in explaining the low coverage. Nevertheless, it is reasonable to suspect that vital registration coverage in Central Asia and Azerbaijan is lower than in other parts of the former Soviet Union. The estimates of registration coverage for all Central Asian republics, Georgia, and Azerbaijan are considerably lower for women than for men. This sex difference in vital registration coverage could be explained by more age overstatement by males than females or by sex bias in death registration. Further work using more detailed data on migration between republics by age and sex is needed to improve the estimates of sex-specific underregistration.

We conclude that for most republics, registration of adult deaths is 95 percent or more complete. Registration of adult deaths in the Central Asian republics is probably between 85 and 95 percent. Registration coverage of adult female deaths in Central Asia, Georgia, and Azerbaijan may be lower than that of adult male deaths. For our analysis, we have chosen not to adjust the reported levels of adult mortality based on the Bennett-Horiuchi technique. The 10 to 15 percent underestimation that may be present does not affect any of our major conclusions. Where appropriate, we draw attention to the effect corrections might have on the observed patterns of mortality and years of life lost.

Classifications of Causes of Death

There are two distinct sets of concerns with the attribution of causes of death in the republics of the former Soviet Union: the classification system and the quality of the coding of each individual death.

The countries of Eastern Europe switched from the Soviet system of classifying causes of death after World War II; the NIS countries, however, have continued to use the Soviet system. Meslé et al. (1993) report that the Soviet system has undergone four major revisions since 1950; the last three revisions have been based on the International Classification of Diseases (ICD-7, ICD-8, and ICD-9), but contain many fewer causes (see also Kingkade and Arriaga, in this volume). The latest Soviet revision, in use since 1981, has also been slightly modified to include additional causes, such as AIDS (Goskomstat, 1987). Based on a translation of a bridge-coding manual prepared by the Soviet Central Statistical Administration (Goskomstat), we have mapped the Soviet codes to ICD-9. In turn, we have mapped the ICD-9 codes to the simplified list of diseases proposed by Murray and Lopez (1994). Without a formal bridge-coding exercise, whereby the same set of deaths is coded for both ICD-9 and the Soviet system, a potential error in interpretation is introduced. As discussed below, this is a significant problem only for complex groups of causes, such as cardiovascular diseases.

In addition, poor diagnostic skill in the NIS may introduce systematic error in the cause-of-death data. One of the only objective indicators of the quality of cause-of-death attribution is the proportion of deaths coded by physicians (Lopez, 1989). Even in Central Asia, more than 99 percent of deaths are coded by physicians (Goskomstat, 1987). Follow-up studies (where coding was reviewed by a panel of experts) from 1965 in central Russia, 1979 in Russia, and 1981-1982 in Belarus and Turkmenistan reveal that the percentage over- or underestimation for most large groups of causes, such as cardiovascular disease, is very small, e.g., 3.1 to 2.3 percent. The largest errors are in coding of respiratory disease, with errors of 1 1.3 to 17.2 percent (see Shkolnikov et al., in this volume).

Although nearly all registered deaths are coded by physicians, and the three follow-up studies demonstrate that the estimated population cause-specific mortality rates are reasonable, there may be substantial differences in diagnostic practice among countries. The results presented below, however, do not suggest that there is more diagnostic error in the data for the former Soviet republics than is observed for other developed countries.

Years of Life Lost Due to Premature Mortality and Excess Years of Life Lost

To capture the importance of death at different ages, we compute years of life lost due to premature mortality, using the methods outlined by Murray et al. (1994) and applied by Murray (1994).⁵ Estimation of years of life lost due to premature mortality provides a picture of the major causes of mortality, but not of avoidable premature death. To identify avoidable or excess years of life lost, we make comparisons with the rates of years of life lost observed for the Established Market Economies (Murray and Lopez, 1994). Excess years of life lost is then defined as the difference between observed years of life lost for each age and sex by cause and the number expected if the rates of the Established Market Economies are applied in a region. Excess years of life lost thus defined can be negative for a cause if the mortality rates by age and sex for a given disease are lower in a region than in the Established Market Economies.⁶

Geographic Patterns of Mortality, 1990

Summary results for each of the NIS and Eastern European countries comprising the Formerly Socialist Economies are provided in Table 6-3a for males and 6-3b for females. These tables provide ₅q₀ (the probability of death between birth and age 5); ₄₅q₁₅ (the probability of death between ages 15 and 60); ₁₀q₆₀ (the probability of death between ages 60 and 70); and e(0), or life expectancy at birth. Within the group of Formerly Socialist Economies, ₅q₀ ranges from 15 to 95 per 1,000 for boys and 11 to 78 for girls. Among adults, male ₄₅q₁₅ ranges from 194 to 305 per 1,000 and female ₄₅q₁₅ from 94 to 155. The high level and extensive range of adult male mortality is most remarkable. Adult male mortality in the Russian Federation, for example, is equal to that of India, whereas adult Russian women enjoy mortality that is 52 percent lower than in India.

TABLE 6-3a

Child and Adult Mortality in the Formerly Socialist Economies, Males, 1990.

TABLE 6-3b

Child and Adult Mortality in the Formerly Socialist Economies, Females, 1990.

The Formerly Socialist Economies are not a homogeneous group as most analyses tacitly assume. Figure 6-2 shows child mortality (₅q₀) on the x-axis and adult male mortality (₄₅q₁₅) on the y-axis. Three clusters of countries can be identified by simple inspection: a group with moderate child and moderate adult mortality, a group with low child and low adult mortality, and a group with low child and high adult mortality. Remarkably, each of these clusters contains a set of geographically contiguous countries. In fact, the countries are arrayed on the diagram in a manner that approximates a map of the Formerly Socialist Economies. Accordingly, we have divided the countries into three groups, which we term Central Asia, South FSE (for Formerly Socialist Economies), and North FSE. Notably, Kazakstan, which is sometimes included with the four Central Asian republics, is on the demographic boundary with North FSE in terms of the child-adult mortality map. We have included it with North FSE because of its high adult mortality. Summary measures for each of the three regions are provided in Table 6-4. As the table shows, even if adult mortality is adjusted for underregistration, the Central Asian republics remain a distinct cluster with high child and moderate adult mortality.

Figure 6-2

Adult male mortality vs. child mortality in the Formerly Socialist Economies of Europe and Central Asia.

TABLE 6-4

Basic Indicators and Mortality Figures for Three Regions of the Formerly Socialist Economies, Circa 1990.

Another way of putting in perspective the dissonance between child and adult mortality is to compare their current and expected levels. Expected levels can be determined using two different methods: the level of adult mortality expected from the level of child mortality based on a model life table (or vice versa), or the level of adult and child mortality expected on the basis of income per capita.

For the first method, the North model life table was chosen for comparison (Coale and Demeny, 1966). Table 6-5 shows the difference between observed adult mortality and adult mortality expected on the basis of observed child mortality from the model life table. The residuals confirm our clustering of countries into three groups. Central Asia has moderate child mortality (60 or more per 1,000) and expected or slightly higher levels of adult mortality. South FSE has low child mortality (less than 60 per 1,000) and moderately higher levels of adult mortality than expected (less than 10 percent excess adult mortality). And North FSE has low child mortality rates and very high levels of adult mortality (twice or more the model life table value). The deviation between female adult mortality and that expected based on model life table North is notably different than that for males. In Central Asia, adult female mortality is on average 42 per 1,000 lower than expected (although this may be explained in part by underregistration); in South FSE it is 5 per 1,000 lower than expected; and in North FSE it is 16 per 1,000 higher than expected. While women actually have better or close to expected mortality, the geographic pattern is symmetrical with that of males, confirming the significance of our three-part division of the Formerly Socialist Economies for both males and females.

TABLE 6-5

Deviations from Predicted Mortality, 1990.

Alternatively, we can compare observed levels of child and adult mortality with those expected on the basis of income per capita. Using data from the World Development Report 1993 (World Bank, 1993), we found that the relationship between the natural log of ₅q₀ and ₄₅q₁₅ and the natural log of income per capita in international dollars ⁷ is linear. The fitted regression equations can then be used to determine whether levels of adult and child mortality in each republic are above or below the levels expected on the basis of income per capita (Table 6-5). The average levels of adult and child mortality for the three regions are also shown. In Central Asia, child mortality is higher than expected, and adult mortality is slightly lower. South FSE has somewhat higher adult mortality than expected. But North FSE has markedly higher adult mortality than expected for its moderate income per capita. This method of defining excess mortality confirms that the three regions have distinct epidemiological profiles even when income per capita is considered.

Years of Life Lost Due to Premature Mortality by Cause, Age, and Sex

Combining the registered deaths for each of the countries in each region, we have created regional figures for North FSE, South FSE , and Central Asia. The structure of cause of death is different in each of the three regions, as presented in Table 6-6. North FSE is dominated by injuries and noncommunicable causes (Group II). The large share of years of life lost due to injuries in this region as compared with other regions in the world, such as the Established Market Economies, is notable. In South FSE, because of a younger population and slightly higher child mortality than in North FSE, Group I (communicable, maternal, and perinatal causes) is still an important cause of years of life lost. Injuries are much less important than in North FSE or Central Asia. In Central Asia, nearly half of all years of life lost is attributable to Group I. The division by large groups of causes divided into age groups reveals that in the population over age 5, injuries (Group III) claim a very large share of years of life lost.

TABLE 6-6

Years of Life Lost Disaggregated by Large Groups of Causes for Each Subregion, 1990.

More detailed information on years of life lost by cause is provided in Table 6-7. In North FSE, the results are notable in Group I for a considerable burden of tuberculosis and respiratory infections. Lung cancer causes over 4 percent of all years of life lost, reflecting the prominent role of smoking in defining the health problems of adults. Other cancers causing more than 1 percent of the total burden include lymphoma/leukemia and cancers of the stomach, colon/rectum, and breast. Cardiovascular diseases cause 35 percent of the total years of life lost—ischemic heart disease representing nearly half of this total, followed by cerebrovascular disease. The residual category, ''other cardiovascular," still causes 7.6 percent of the total; this category needs to be further defined to determine the contributing components. Unexpectedly, alcoholic cirrhosis is not a large cause of death in this region. Alcohol-associated deaths fall to a large extent under the categories of neuropsychiatric (alcohol dependence) and poisoning (58 percent of years of life lost due to adult poisoning in North FSE is attributable to alcohol). The consumption of hard liquor in preference to other forms of alcohol may explain the distinctive manifestation of alcohol in the mortality data for this region. Injuries cause an extraordinary 22 percent of total years of life lost in North FSE. Motor vehicle accidents, suicides, poisonings, and homicides, in descending order, are the largest contributors to this total. Treml (1982 and in this volume) has pointed out that a considerable portion of the poisonings is probably due to alcohol intoxication.

TABLE 6-7

Percent Distribution of Years of Life Lost by Major Causes of Death, by Region, 1990.

The results for South FSE are notable for the much larger share of years of life lost (7.1 percent) attributable to respiratory infections, mostly among children (68 percent). Reflecting a slightly less advanced smoking epidemic, lung cancer causes 3.4 percent of years of life lost, followed by cancers of the stomach and breast, lymphoma/leukemia, and cancer of the colon/rectum. As in North FSE, cardiovascular diseases are the most important cause of death, accounting for 38 percent of years of life lost. The pattern within this category, however, is distinctly different. Ischemic heart disease and cerebrovascular disease have nearly equal shares, 12.6 and 11.0 percent, respectively. Other cardiovascular diseases represent the largest component, 14.0 percent. This residual category is worrisome. Local coding practices may be responsible for assigning to this category some ischemic heart disease deaths, or possibly deaths from some other major cardiovascular causes, such as cardiomyopathy or arrhythmia. Further work on defining the specific cardiovascular causes coded in this group is urgently required. Digestive diseases cause over 5.3 percent of years of life lost in South FSE, with cirrhosis being responsible for more than half of this amount. As a group, injuries are much less important in South than in North FSE. Motor vehicle accidents cause 2.3 percent of deaths, followed by suicides (1.3 percent).

In Central Asia, years of life lost is dominated by child deaths; thus the associated causes are more in Group I. Respiratory infections (29.7 percent), diarrheal diseases (8.6 percent), and perinatal causes (9.5 percent) are the most important. Among cancers, lymphoma/leukemia and cancers of the lung and stomach account for more than 1 percent of years of life lost each. Cardiovascular diseases cause 15.6 percent of deaths; of these, over one-half are attributable to ischemic heart disease and about one-fourth to cerebrovascular disease. The injury pattern is similar to that of South FSE, except for the prominent role of drownings (1.7 percent), suicides (1.1 percent), and homicides (0.8 percent).

Annex Table 6-1 provides estimates of years of life lost by cause for each of the NIS countries, to facilitate more detailed comparisons among states within each of the epidemiological regions.

Excess Years of Life Lost

Table 6-8 shows the distribution by age, sex, and region of excess years of life lost, by disease group and all causes of death. Figures 6-3a, b, and c present the excess years of life lost by age in South FSE, North FSE, and Central Asia, respectively. Total years of life lost in Central Asia is 80 percent higher than expected based on rates of the Established Market Economies, and in North FSE and South FSE is 67 and 50 percent higher, respectively. The excess can be apportioned among different age and sex groups. In Central Asia, 80 percent of the excess is due to child mortality (under age 5). The excess is concentrated among children because of high fertility and a young age structure, combined with moderately high levels of child mortality even by developing world standards. Adult men and women account for 9.9 percent of the excess years of life lost in the region. In South FSE, excess years of life lost is distributed across nearly all age groups, with 28.9 percent being among children under age 5 and 25.3 percent being among adult males. In North FSE, almost half of the excess (44.8 percent) is among adult men aged 15-59, confirming the unique mortality pattern of this region. Just over 15 percent is among children under age 5, and a further 10.6 percent is among women age 15-59. Mortality among the population over age 60 contributes 26 percent to the total excess.

TABLE 6-8

Excess Years of Life Lost, by Region, Sex, Age Group, and Cause, 1990 Region.

Figure 6-3a

Excess years of life lost by age, South FSE region.

Figure 6-3b

Excess years of life lost by age, North FSE region.

Figure 6-3c

Excess years of life lost by age, Central Asia region.

In Central Asia, the major problem is excess mortality in the age group 0-4 years. Figure 6-4 provides the distribution of excess years of life lost among this age group by major causes. Nearly 90 percent is from communicable and perinatal causes. More specifically, 56 percent is due to respiratory infections, a pattern characteristic of a developing country. The second-largest share (17 percent) is attributable to diarrheal diseases, followed by perinatal causes (11 percent), hepatitis (4 percent), and drowning (2 percent). The prominent role of hepatitis in this age group is highly unusual; further efforts are needed to confirm the coding and validity of this burden. Measles, diphtheria, pertussis, and tuberculosis are not large contributors to the excess mortality among children in this region, indicating the effectiveness of immunization programs in the region, at least up until 1990.

Figure 6-4

Excess years of life lost by cause for ages 0-4, Central Asia region.

North FSE males account for nearly one-third of all excess years of life lost in the entire region of the Formerly Socialist Economies. Approximately half of this excess is due to noncommunicable diseases and half to injuries. The literature on the rising mortality among adult males, particularly in North FSE, has stressed that most of the increase is due to cardiovascular disease. Here, we are examining not only the cause of the increase, but also the difference in the level of mortality. Some of this excess already existed before the increases in adult male mortality began in 1965. Differences in level or trend viewpoints can lead to different health priorities (see also Anderson and Silver, in this volume). Clearly, for the population affected, differences in current mortality levels or years of life lost by cause are the more important.

Figure 6-5 allocates the excess years of life lost in these age groups by more detailed causes. The figure shows that 21 percent is due to ischemic heart disease. In descending order of magnitude, road traffic accidents, suicides, poisoning (which includes acute alcohol ingestion), cerebrovascular disease, lung cancer, drowning, and homicide each contribute more than 5 percent to the total excess years of life lost. Over 16 percent is distributed across a large number of more specific causes, each of which contributes less than 1 percent to the total. Given the real concern about heavy intake of hard alcohol in the region, it is surprising that there is no excess death due to cirrhosis. This may reflect a coding practice, however. In some countries, alcohol-related deaths are coded to various other causes, such as acute alcohol poisoning; alcohol dependence; and drug-use, subcode alcohol.

Figure 6-5

Excess years of life lost by cause for males, ages 15-59, North FSE region.

We turn now to a more detailed examination of the mortality patterns in North FSE.