Like the Soviet system on which it is based, the current Russian system for producing information on causes of death is decentralized. The extent of this decentralization has increased substantially since the country made the transition to a new system of cause-of-death coding in 1999. Before the transition, the network of regional Statistics Offices had been responsible for coding the underlying cause of death; but since 1999, this task has been delegated to the individual medical practitioners. This shift coincided with the transition to the ICD-10. The Statistics Service coders had to code the underlying cause of death in accordance with the Soviet Abridged Classification, which offered them only 184 diagnostic items to choose from. By contrast, medical practitioners now have to assign the cause using the complete ICD-10 classification, which contains over 10,000 nosological items.
According to some experts, this change led to a deterioration of the Russian system of coding and gathering information on the causes of death, in part because no unified training in cause-of-death coding for medical workers was provided [47]. Moreover, medical professionals were not even given any centralized instructions for filling in medical death certificates and coding the causes of death in accordance with ICD rules [48]. This lack of preparation has led to specific difficulties with and discrepancies in coding practices across subnational entities and over time.
The present study has identified several problems with the cause-specific mortality statistics across the Russian territories.
We have found that while certain causes of death (e.g., cancers, transport accidents) have roughly comparable cause-specific shares across the regional mortality structures, there is a much greater degree of inconsistency in the prevalence of other cause-specific shares across the regions. For some causes, the magnitude of these inconsistencies is too large, and is therefore more likely to be artificial than to be indicative of natural variation across regions. Thus, it is possible that the regional differences in mortality from these causes reflect variation in coding practices, rather than real differences in the prevalence of diseases.
The lowest levels of consistency among the causes of death we investigated were found for AIDS. However, the high degree of variability of AIDS diagnoses cannot be regarded as a problem of coding accuracy only. AIDS was a new cause of death at the start of our study period, and the number of people who were dying from this disease was clearly increasing as the period progressed. Over time, our understanding of and ability to detect the disease have improved, and coding practices have adapted accordingly. Mortality from AIDS has been rising rapidly in Russia over the last decade (Fig. 4). One piece of evidence that supports the claim that there is “natural” wide variation in AIDS mortality across different regions of Russia is the finding that there is a strong positive correlation between the registered prevalence of HIV in a given region [49] and the share of AIDS in the regional all-cause SDR (the correlation was 0.88 in 2012).
As is the case for other communicable diseases, AIDS has spread unevenly across the population. Some regions could be a nidus of infection, while others have had much lower incidence levels. Thus, the large degrees of spatial and temporal variation in the contributions of AIDS to overall mortality can be explained. However, some portion of the variation in the prevalence of AIDS mortality across Russian regions may have also been caused by discrepancies in the cause-of-death coding practices. In many countries, deaths from AIDS are systemically miscoded under tuberculosis, endocrine disorders, Kaposi’s sarcoma, meningitis, encephalitis, certain garbage codes, and other causes of death [50, 51]. C. Murray and co-authors estimated that the real number of deaths from AIDS in Russia in 2013 was 16,138 (95 % uncertainty interval 11,963 to 22,526) [51], or 52 % higher than the number that was officially reported by the Russian State Statistics Service.
In 2010, E. Tzybikova examined 6249 deaths that occurred among patients with newly diagnosed tuberculosis in 80 Russian regions. She found that in 61 regions some deaths from AIDS were mistakenly coded as tuberculosis. While tuberculosis was chosen as an underlying cause of death, AIDS was listed as an associated cause, which violates the ICD instructions for sequencing the causes of death [52]. The total number of such cases found by this study was 1,004. When we look at the 6784 deaths from AIDS recorded in the official statistics in 2010, it appears that a very significant fraction of the deaths were misclassified. Unfortunately, this study did not provide a detailed explanation of the study design, and did not investigate the regional peculiarities in the misclassification of AIDS and tuberculosis. However, the finding that there were incidents of misclassification in 61 of the 80 regions studied may indicate that the extent of the misclassification of deaths from AIDS also differs across regions.
Other groups of infectious diseases that we studied, such as “tuberculosis” and “other infectious diseases,” had medium levels of inconsistency compared to other causes of death. As was shown above, regional variation in mortality from tuberculosis can be affected by the misclassification of AIDS. Difficulties in certifying deaths with AIDS/tuberculosis co-infection are common, especially in countries with a high burden of HIV [50, 53]. Nevertheless, it should be noted that during the period of observation the prevalence of tuberculosis in the mortality structure of Russian regions (\( \overline{{S}_{\bullet, c,t}} \) is equal to 1.3 %) was several times higher than mortality from AIDS (\( \overline{{S}_{\bullet, c,t}} \) is equal to 0.2 %). Hence, the miscoding of these two causes distorts the mortality statistics for AIDS much more significantly than for tuberculosis.
While high levels of spatial and temporal heterogeneity are normal in the transmission of infectious diseases, having to rely solely on the data reported by official statistics makes it difficult to determine whether high degrees of variation in mortality from infectious causes reflect real differences in the prevalence of disease, or are indicative of differences in coding practices as well. But for causes of death from non-communicable diseases, it seems rather unlikely that very high levels of within-country variation are natural. Our finding that some non-communicable diseases had much higher levels of spatial variation than some communicable diseases can serve as indirect proof that the level of variation we found for some non-communicable diseases is too high and cannot be accurate.
Very low levels of consistency were found for some groups of causes from the ICD chapter “diseases of the circulatory system.” For some of these causes, the level of consistency would have been greater if we had assigned them to broader groups of items. For instance, within the group of ischemic heart diseases the ratio between the inter-regional maximum and the inter-regional minimum values of V
r,c
calculated according to equation (2) amounts to 13.1 for “atherosclerotic heart disease,” 5.6 for “myocardial infarction,” and 4.3 for “other forms of ischemic heart diseases.” But if we combine all of these items into one group, “ischemic heart disease,” this ratio would be only 3.1. Similar results can also be obtained for the group of “cerebrovascular diseases.” These lower levels of inconsistency at higher levels of aggregation suggest that conflation often occurs when the possible causes of death are medically similar. Analyzing cause-specific mortality at higher levels of aggregation can reduce biases.
Coding discrepancies can undermine cause-specific analysis more significantly for causes that cannot be meaningfully grouped together with other items. Categories that represent complete ICD chapters, such as “diseases of the nervous system,” “endocrine, nutritional, and metabolic disorders,” and “mental and behavioral disorders” had very high levels of spatial and temporal inconsistencies in our analysis. Even more biases in the analysis of cause-specific mortality are caused by spatial and temporal differences in the use of garbage codes from the ICD-10 chapter XVIII, “symptoms, signs, and abnormal clinical and laboratory findings;” or groups of causes, such as “injuries of undetermined intent.” The propensity to assign garbage codes as underlying causes of death varied significantly across Russian regions. As garbage codes constitute a high share of the causes of death recorded in the Russian mortality structure, regional and period discordances can heavily affect the comparability of mortality indicators for other specific groups of causes of death that are misclassified with garbage codes.
In terms of spatial variations, a few regions can be pinpointed as having the cause-specific mortality structures that deviate the most starkly from the inter-regional average: the cities of Moscow and Saint Petersburg, which are constituent federal units; and the Republic of Dagestan, a Muslim region located in the North Caucasus. We offer several hypotheses for why these particular regions had the highest scores in our analysis.
First, these three regions have the lowest overall mortality levels of the 52 regions in our sample. Lower mortality levels are generally indicative of certain mortality structures. In particular, lower mortality levels usually correspond with a higher share of deaths from neoplasms in relation to other causes of death. Accordingly, it is quite apparent on the heatmap that there are substantial differences between Moscow and Saint Petersburg on the one hand and the other Russian regions on the other in terms of the share of deaths from neoplasms relative to overall mortality. But the deviating pattern for Dagestan is mainly attributable to the relative shares of other causes of death. It is important to note that the Republic of Dagestan differs considerably from the other regions in our sample, as it is the only national republic of North Caucasus selected for this analysis, and the Muslim regions of North Caucasus have much lower mortality levels from alcohol-related causes than the rest of Russia. In addition, in these regions there are long-term concerns about the understatement of mortality at infant and old ages due to the underreporting of deaths, and about the overstatement of age [54, 55].
Second, the populations of the cities of Moscow and Saint Petersburg are entirely urban. Dagestan, by contrast, is the only region in the sample in which the urban population is still smaller than the rural population. Therefore, it is possible that the significant differences in the mortality structures between these three regions and the other regions of Russia are at least partly attributable to the differences between urban and rural populations.
The other possible explanation is a registration effect. A death in Russia can be registered either at the location of the deceased’s permanent residence or at the location of death. This may result in certain biases in mortality statistics at the regional level, which can be especially large for Moscow and Saint Petersburg. First, there are a number of large federal medical centers in these two cities that specialize in the treatment of specific diseases and especially of cancers. In addition to residents of Moscow and Saint Petersburg, residents of other regions may be treated in these centers. Among all deaths from cancers in Moscow in 1990–1994, 4.8 % of the men and 5.6 % of the women who died were non-residents [56]. Additionally, Moscow and Saint Petersburg have huge migration inflows. The cause-specific mortality structures of these cities may therefore be affected by the selectivity in the health status of arriving migrants. Arkhangelsky and co-authors found that the cause-specific mortality structures of residents and non-residents in Moscow are very different [57]. Non-residents are, for example, more likely than residents to die from external causes, infectious diseases, and ill-defined conditions.
The results obtained in our study suggest that a complex series of actions will be needed to standardize regional approaches to cause-of-death coding and to improve the comparability of cause-specific mortality data within Russia. These actions should focus on strengthening the legal and regulatory framework for mortality statistics, improving the quality of human resources, and ensuring the full implementation of ICD standards. A national “gold standard” of training on death certification should be developed for medical practitioners. To increase the likelihood that medical workers will adhere to a uniform set of coding principles, the training procedures should be standardized to the greatest possible extent. Ideally, an automated, centralized coding and/or training software application would be designed and implemented across the country. The regular monitoring of the comparability of cause-specific mortality data reported by regions is also essential. In our study we took the average region/cause deviations for an 11-year period; thus, only the long-term deviations from the inter-regional average level were highlighted. Surprisingly, the number of such long-term deviations was found to be quite large. This finding suggests that regions can follow different coding practices for a long period without these discrepancies being discovered by the responsible federal authorities. Additional checks must be carried out in cases in which mortality from a certain cause in a certain region obviously deviates disproportionately from the average level, and the origins of these kinds of deviations should be thoroughly investigated.
We also suggest producing an aggregated list of causes of death that can be used in analyses of regional mortality patterns with a minimal risk of inter-regional incomparability and biases. Such a list should be regarded exclusively as a stopgap measure. Developing and implementing a national plan for strengthening the quality of cause-of-death statistics is essential, and should still be seen as the highest priority. But as making substantial improvements takes time, in the interim the aggregated list can be useful for analyzing cause-specific mortality in Russia at the regional level.
Limitations
Our study has several limitations. The first arises from the indirect character of the method proposed. As we analyzed the official cause-specific mortality data as they are, we can make only indirect assessments of the quality and validity of these data. Although we can observe spatial and temporal variations in cause-specific shares, we cannot be certain whether they are caused by real mortality differences or by discrepancies in coding practices. While we can be reasonably sure that such discrepancies are present when the regional deviations in the causes of death are especially large, we cannot judge the less obvious cases. Further research is needed to determine why there are problems in the data.
The second limitation follows from the grouping of causes of death in the Russian Abridged Classification. It is impossible to extract from these data certain groups of ill-defined cancers and ill-defined cardiovascular diseases, which are also regarded as so-called garbage codes [58–60]. In most cases, such garbage codes are combined in the RC-1999 with some well-defined codes under the heading “other and unspecified.” For instance, the group “cancers of other and independent (primary) multiple sites” in the RC-1999 includes, in addition to the codes for ill-defined cancers (C76, C80, C97), codes that correspond to neoplasms with specific localization, and that cannot be referred to as garbage codes, such as “cancer of eye and adnexa (C69)” and “cancers of thyroid and other endocrine glands (C73-C75).” Because these codes are ill-defined, we could not compare their prevalence across the regions, yet this is an important criterion for evaluation the quality of cause-of-death coding [2].
Third, our study was based on death counts obtained at the regional level. However, while the coding procedure in Russia is performed in a completely decentralized manner at the level of medical practitioners, there may also be some important discrepancies within the regions themselves.