Key findings

• Russia’s cause-of-death data is easily manipulated because of vague coding rules and differences across medical reporting systems, enabling officials to reassign causes in response to targets set by the 2012 ‘May decrees’¹.
• After 2012, reported cardiovascular disease (CVD) mortality fell, while deaths among those aged 60+ from nervous system diseases, endocrine disorders, mental illness, ill-defined causes and several other categories rose sharply – changes too abrupt to reflect ageing.
• Analysis of 2013–2019 data shows suspicious causes rising from about 8 per cent to 21 per cent of total mortality, while CVD fell from roughly 70 per cent to 55 per cent; several categories experienced exceptional annual spikes.
• Regional patterns vary markedly: some regions show rises exceeding 20 percentage points in suspicious causes, while others show almost none; these differences are unrelated to external-cause mortality or COVID-19 coding.
• The integrity of cause-specific mortality data has declined since 2012, undermining regional comparisons and complicating assessments of war-related deaths, making alternative big-data sources more reliable.

Introduction

Indicators of mortality by cause of death have been among the most susceptible to administrative control and distortion in Russia. Such manipulation is frequently political in nature, as demonstrated by the handling of COVID-19 mortality statistics. The principal mechanism of distortion lies in the physician’s ability to prioritise one cause of death over another when coding, in order to meet government-set performance targets.

The rules governing coding also vary across statistical systems. Hospital statistics record only the patient’s primary disease, whereas outpatient and polyclinic statistics include all diagnosed conditions except complications of the primary disease.

Within this framework, significant scope exists for politically motivated manipulation. Lower-level officials and politicians may seek to demonstrate improved performance and competitiveness to their superiors by altering or ‘doctoring’ statistics; because coding criteria are vague, such practices can be implemented at relatively low administrative cost.

A driver of such manipulation was the 2012 ‘May decrees’. On 7 May 2012, Vladimir Putin issued presidential order No. 598, ‘On improving state health policy’. It established several quantitative targets, including a reduction in mortality from cardiovascular diseases (CVD) to 649.4 per 100,000 by 2018. Mortality from neoplasms, tuberculosis and road accidents was also subject to explicit targets.

Additional key performance indicators included:

• A reduction in the death rate from malignant neoplasms in children to 192.8 per 100,000;
• A reduction in the tuberculosis death rate to 11.8 per 100,000;
• A reduction in deaths caused by traffic accidents to 10.6 per 100,000;
• A reduction in infant mortality to 7.5 per 1,000 newborns in regions with high baseline levels.

CVD mortality received particular emphasis. Regional governors were assessed in part according to progress towards these targets and therefore faced strong incentives to alter coding practices so as to report lower CVD mortality, or to exploit ambiguities in coding rules for this purpose.

The objective of this paper is therefore to examine the drivers of miscoding in greater detail and to test the hypothesis that presidential order No. 598 of 7 May 2012 prompted the systematic miscoding of deaths from circulatory system diseases as a result of opportunistic or fraudulent behaviour by political actors.

For details on the methodology, see the Appendix at the end of this article.

Evidence of misclassification in cause-of-death statistics

Since the decree came into force in 2012–2013, mortality in older age groups from several causes has begun to rise, while reported mortality from cardiovascular diseases (CVD) has declined sharply. To illustrate this pattern, age-specific death rates by cause of death for both sexes were drawn from the Russian Fertility and Mortality Database for the period 1999–2019. This data was used to calculate crude death rates (CDRs) by broad cause-of-death categories, largely based on International Classification of Diseases (ICD-10) classes. CDRs were employed to facilitate clearer interpretation. The analysis was restricted to individuals aged 60 years and older, as shifts in the structure of causes of death are more pronounced in older cohorts. Existing research also highlights that mortality statistics for these age groups exhibit greater inconsistency.

The target causes of death most likely to be subject to manipulation all fall within the broader cluster of cardiovascular diseases:

• Heart diseases (ICD-10: I00-I52);
• Cerebrovascular diseases (ICD-10: I60-I69);
• Other cardiovascular diseases (ICD-10: I70-I99).

In cases of miscoding, death certificates cannot simply be removed from the statistical system; Russian vital registration consistently records the total number of deaths. Manipulation therefore concerns the cause of death assigned. It follows that miscoding would manifest not only as a decline in recorded CVD mortality but also as a corresponding rise in other categories where misclassified cardiovascular deaths might be reassigned. The literature indicates that, at older ages, the greatest inconsistencies occur in mental and behavioural disorders, nervous system diseases, endocrine disorders, ill-defined conditions and certain cardiovascular subcategories (Danilova 2015, 2016).

Three major clusters appear particularly likely to absorb redistributed CVD mortality, a pattern visible in their marked absolute increases:

• Diseases of the nervous system (ICD-10: G00-G99);
• Endocrine, nutritional and metabolic diseases (ICD-10: E00-E90);
• Ill-defined conditions (ICD-10: R00-R99).

Four additional groups also appear plausible candidates for statistical reallocation:

• Mental and behavioural disorders (ICD-10: F00-F99);
• Diseases of the musculoskeletal system and connective tissue (ICD-10: M00-M99);
• Diseases of the genitourinary system (ICD-10: N00-N99);
• Diseases of the skin and subcutaneous tissue (ICD-10: L00-L99).

For example, mortality from ICD-10 class ‘Symptoms, signs and abnormal clinical and laboratory findings, not elsewhere classified’ (ill-defined conditions) rose noticeably. Prior to 2012, annual growth in the crude death rate (CDR) for this category did not exceed 10 per cent, with year-to-year fluctuations in both directions. In 2012, 2013 and 2014, however, the CDR increased by 12.4 per cent, 18.5 per cent and 22.8 per cent, respectively. This class includes ‘Senility’ (R54), a designation typically implying an unknown or unspecified underlying cause of death.

Similarly striking increases occurred across other categories. In 2014, the CDR for mental and behavioural disorders (including dementia, F00-F03) increased by 155.9 per cent, while mortality from nervous system diseases rose by 103.9 per cent. Other notable increases following the decree include:

• Endocrine, nutritional and metabolic diseases (+56.7 per cent in 2014);
• Musculoskeletal and connective tissue diseases (+73.4 per cent in 2015);
• Skin and subcutaneous tissue diseases (+39.6 per cent in 2016);
• Genitourinary diseases (+21.7 per cent in 2015).

Overall, the rise in mortality from these seven suspicious cause-of-death groups increased their combined share of the total CDR from 7–9 per cent in 2010–2012 to 21 per cent in 2017–2019. Over the same period, the share of deaths attributed to CVD declined from 68–71 per cent to 54–55 per cent. No comparable or rapid shift in the structure of old-age mortality by cause had been observed before 2012.

Regions with the highest levels of statistical manipulation

The results above were identified at the federal level. However, as noted earlier, the primary beneficiaries of such manipulation are regional politicians and bureaucrats; consequently, regional-level distortions in mortality statistics warrant particular attention.

The rise in mortality from suspicious causes after 2012 varied substantially across regions, consistent with expectations. In several regions, the difference between the maximum share of the CDR attributable to suspicious causes during 2013–2019 and the corresponding share in 2012 exceeded 20 percentage points. Differences of this magnitude may reflect the artificial reallocation of diagnoses.

Such increases could, of course, also be linked to changes in medically driven coding practices, such as improved diagnosis of dementia, diabetes and other conditions. If this were the sole explanation, however, more questions would arise:

• Why did these increases occur specifically after the 2012 ‘May decrees’?
• Why did certain regions fail to exhibit a comparable rise?

The regions with the highest likelihood of manipulation, as indicated by changes in the share of CDR from suspicious causes, were the Karachay-Cherkess Republic (a rise of 33.8 percentage points), Moscow Region (29.4), Nizhny Novgorod Region (29.4), Sakhalin Region (26.4), Amur Region (26.4), Yaroslavl Region (25.5), the Chechen Republic (24.6), Omsk Region (23.9), the Republic of Mordovia (23.2), Ryazan Region (22.8), the Republic of Mari El (22.5), the Kabardino-Balkar Republic (22.4), Tver Region (22.0), Rostov Region (21.6) and Tambov Region (20.8).

Data source: Russian Birth and Mortality Database (RosBRiS)However, there are also regions in which the structure of mortality by cause exhibited minimal change. For example, Tomsk Region recorded an increase of just 0.5 percentage points, the Jewish Autonomous Region likewise 0.5, Arkhangelsk Region only 0.1, while Chukotka Autonomous District and Krasnoyarsk Krai showed slight declines of 0.2 percentage points.

How systematic is the pattern?

A central question is the extent to which this effect is systematic. To explore this, the results were compared with patterns in other causes of death that might also be susceptible to manipulation. A plausible candidate is mortality from external causes, particularly the category of ‘events of undetermined intent’, which forms part of the broader set of external causes lacking clear classification. However, no statistical evidence was found to suggest any meaningful association between changes in this category and the rise in suspicious causes, implying that miscoding does not follow a uniform national pattern. Instead, it appears that regions pursue their performance targets using distinct, locally specific strategies.

To investigate this further, the ratio of deaths from events of undetermined intent to other external causes of death was calculated, and the relationship between this ratio and the share of suspicious cases was examined for 2012 (‘pre-May decrees’), 2019 (pre-COVID-19 and after adjustments to coding practices prompted by the ‘May decrees’) and 2022 (the period of COVID-19 and the war).

No significant association was observed between coding practices during the COVID-19 pandemic (that is, causes of death among those aged 60+ coded as caused by COVID-19) and the earlier manipulations following the ‘May decrees’.

Potential applications

Since 2012, the quality of cause-of-death statistics has deteriorated markedly. A sharp decline in reported mortality from circulatory system diseases (CSD) has been accompanied by increases in several other cause-of-death categories. In principle, shifts in the causes of mortality are to be expected: an ageing population and revisions to diagnostic guidelines naturally produce gradual changes. In this instance, however, the speed and magnitude of the increases observed in certain categories are difficult to explain without reference to political incentives.

Consequently, in some regions the distribution of mortality by cause has diverged so substantially from underlying health conditions that it no longer provides a reliable reflection of reality. In such contexts, robust analysis of cause-specific mortality becomes effectively impossible. It is also plausible that statistical manipulation in these regions extends beyond cause-of-death classification.

These observations have important implications for the analysis of Russian-Ukrainian war casualties. Because post-2022 mortality statistics are no longer publicly accessible, they cannot be assessed with any degree of accuracy. Even the data available for 2022 is problematic: deaths recorded under external causes – the category most relevant to conflict-related fatalities – may have been doctored, miscoded or otherwise manipulated. Under these conditions, alternative sources, including large-scale big-data approaches such as those employed by Mediazona, provide a more reliable basis for estimating wartime losses.

Appendix: Methodology

One of the goals of this research was to identify the regions exhibiting the highest levels of statistical manipulation in the coding of causes of death. The extent of such manipulation was measured in percentage points, defined as the difference in the share of miscoded cases among those aged 60 and older before and after the ‘May decrees’.

For this purpose, the seven cause-of-death groups (see Evidence of misclassification in cause-of-death statistics) were combined into a single broad category labelled ‘suspicious causes’. The share of deaths attributed to this category was calculated as a proportion of all deaths. For each region, the maximum share of suspicious-cause mortality observed during 2013–2019 was then identified and compared with the corresponding share in 2012, the year preceding the strong incentives to manipulate statistics introduced by the ‘May decrees’.

The resulting indicator is the percentage-point difference between the maximum share of CDR from suspicious causes within total CDR observed between 2013 and 2019 and the ‘initial’ share of CDR from suspicious causes recorded in 2012.

This metric was calculated for each region as follows:

Δ = (ωCDR₂₀₁₃, ωCDR₂₀₁₄, …, ωCDR₂₀₁₉) – ωCDR₂₀₁₂

ωCDR_y = CDR_y^suspicious / CDR_y^total × 100%

CDR_y^suspicious = CDR from suspicious causes in year y
CDR_y^total = all-cause CDR in year y

A simple moving average was applied to smooth the series. This was calculated using three adjacent data points, except in the case of 2019, for which the average of 2018 and 2019 was used.

1. ‘May decrees’ refers to the sets of presidential decrees that Vladimir Putin traditionally issued after his May inaugurations in 2012, 2018 and 2024. They usually outline ambitious targets for economic and social policy, intended as a development roadmap but chronically underfunded and often met only on paper ↩︎