Reduction in Male Suicide Mortality Following the 2006 Russian Alcohol Policy: An Interrupted Time Series Analysis

Abstract

Objectives. We took advantage of a natural experiment to assess the impact on suicide mortality of a suite of Russian alcohol policies.

Methods. We obtained suicide counts from anonymous death records collected by the Russian Federal State Statistics Service. We used autoregressive integrated moving average (ARIMA) interrupted time series techniques to model the effect of the alcohol policy (implemented in January 2006) on monthly male and female suicide counts between January 2000 and December 2010.

Results. Monthly male and female suicide counts decreased during the period under study. Although the ARIMA analysis showed no impact of the policy on female suicide mortality, the results revealed an immediate and permanent reduction of about 9% in male suicides (Ln ω₀ = −0.096; P = .01).

Conclusions. Despite a recent decrease in mortality, rates of alcohol consumption and suicide in Russia remain among the highest in the world. Our analysis revealed that the 2006 alcohol policy in Russia led to a 9% reduction in male suicide mortality, meaning the policy was responsible for saving 4000 male lives annually that would otherwise have been lost to suicide. Together with recent similar findings elsewhere, our results suggest an important role for public health and other population level interventions, including alcohol policy, in reducing alcohol-related harm.

Alcohol-related harm represents a serious public health threat in many nations. It is responsible for 2.5 million deaths annually and is the third leading cause of premature mortality.¹ Because the programmatic efforts of funding agencies and the empirical literature on the alcohol–health relationship are increasingly focusing on individuals, it is important to determine if the population-level levers available to public health can be successful in reducing the burden posed by harmful alcohol consumption. Although population level associations provided by cross-sectional and especially time series designs are important in this respect, natural experiments provide an opportunity for an even more rigorous test of the efficacy of policy attempts to reduce alcohol-related harm. In this study, we took advantage of a natural experiment to determine if a new national alcohol policy in Russia was associated with a reduction in suicide mortality.

Russia provides an important setting for such a test. First, Russia possesses high levels of both alcohol consumption and suicide. Second, the nation has proven stubborn in its resistance to repeated alcohol policies over the centuries, with alcohol continuing to be responsible for about one fifth of all male deaths in Russia.¹ Third, although several population level studies reveal a strong association between alcohol and suicide in Russia, these results remain vulnerable to the charge of spuriousness—that high rates of population-level drinking and suicide are in fact both outcomes of other factors. Finally, and more generally, many consider suicide to be an extremely individual act and not amenable to public health efforts. Given this backdrop, if a rigorous natural experiment shows an alcohol policy to be associated with a reduction in suicide mortality, then we must take seriously the notion that public health efforts play an important role in the reduction of alcohol-related harm.

The level of alcohol consumption in Russia is among the highest in the world. Recent estimates put adult per capita consumption at about 16 liters of pure ethanol alcohol annually.¹ This high level of consumption in Russia is accompanied by a hazardous pattern of heavy episodic drinking, with alcohol researchers placing it in the highest risk category for harmful consumption.^2,3 Studies at the beginning of the 2000s estimated that the number of deaths attributable directly or indirectly to alcohol was 500 000 to 750 000 annually in the nation.^4,5 The level of consumption and the hazardous drinking pattern have been established as important contributors to the wide swings in mortality in Russia dating to the early 1990s, which were driven largely by alcohol-related mortality among working-age males.^6,7 Leon et al. estimated that more than 40% of all premature mortality among working-age Russian males was attributed to hazardous drinking.⁸

The Russian suicide rate is also among the highest in world. The 2010 suicide rate was about 50 per 100 000 for males aged 15 years or older and about 9 per 100 000 for females aged 15 years or older. The Russian male suicide rate ranks second only to Lithuania among nations recently reporting to the World Health Organization,⁹ and the Russian female suicide rate ranks in the top 10. The total suicide rate of about 28 per 100 000 Russians aged 15 years or older is about 2 to 3 times higher than in most European nations and the United States. Although the suicide rate spiked following the collapse of the Soviet Union and during the economic and political crises that followed, the nation also has exhibited high suicide mortality historically.

During the last 3 centuries, Tsarist then Soviet then Russian governments used a series of alcohol policies in an attempt to limit alcohol-related harm. Gorbachev’s anti-alcohol campaign during the mid- to late 1980s is perhaps the best known. As part of that campaign, actions were taken to, among other things, restrict hours of alcohol sales, institute purchase quotas, shutter distilleries and breweries, and uproot vineyards.¹⁰ The Russian alcohol policy of interest in the present study occurred in the first decade of the 21st century. In the face of the Russian mortality crisis and alcohol’s central role in it, President Putin signed a law regulating production and sale of ethyl alcohol and alcohol-containing products in 2005, effective January 1, 2006.¹¹ Among other efforts, the law contained regulations (1) aimed at controlling the volume and quality of alcohol products and sales and (2) requiring the registration of alcohol production and distribution facilities. The new financial investments required by alcohol producers and sellers—registration fees, equipment costs, new excise stamp procedures, etc.—acted as a tax, resulting in fewer producers and distributors and increasing consumer prices.¹¹ Recent research reveals an increase in life expectancy in Russia during the last decade that is partially accounted for by a decrease in deaths attributable directly to alcohol¹²; the authors of that study hypothesized that the 2006 alcohol policy may be partially responsible for this decline. Our other studies show that the policy was, in fact, responsible for a reduction in male deaths attributed to transport accidents¹³ and in male deaths attributed to alcohol poisoning and in male and female deaths stemming from alcoholic cirrhosis of the liver.¹⁴

The goal of the present study was to determine if the 2006 Russian alcohol policy was associated with a decline in suicide mortality in the country. The answer to this question is important to public health efforts, and there is reason to believe there may be such an effect. First, cross-sectional studies show an association between population-level drinking and violence rates—suicide, homicide, or both—in Russia in recent years^15,16 and during the late Tsarist era a century ago.¹⁷ Second, more rigorous time series designs also reveal an association between population-level drinking and suicide rates in Russia. Again, evidence of this time series association dates back more than a century to the Tsarist era¹⁸ and continues through much of the Soviet era and into the contemporary period.^19,20 Third, a recent study using a national-level alcohol policy in another but much smaller East European nation, Slovenia, as a natural experiment revealed that it was responsible for an immediate and permanent 10% reduction in male suicide mortality.²¹ We took advantage of this rare natural experiment and employed interrupted time series techniques to determine if this Russian alcohol policy influenced suicide mortality in the country.

METHODS

The outcome variables in this study were the monthly number of male- and female-specific suicide deaths of those aged 15 years and older. We used counts because there was little change in Russia’s total population between 2000 (∼146 million) and 2010 (∼143 million). Russian cause-of-death data are based on medical death certificates, which are completed by a medical doctor (who either treated the deceased or established the cause of death), the coroner who completed the autopsy, or a forensic medical expert. During the 2000 to 2010 period of our study, more than 97% of suicide diagnoses resulted from a forensic autopsy. Deaths are classified according to the International Classification of Diseases, 10th Revision,²² with suicides coded X60-X84. Data for this study were obtained via a special tabulation of anonymous death records collected by the Russian Federal State Statistics Service. The 2 time series began January 2000 and ended December 2010. There were 132 monthly observations in each series, with 72 months of preintervention data and 60 months of postintervention data. The alcohol policy was implemented in January 2006.

We used autoregressive integrated moving average (ARIMA) interrupted time series techniques to model the intervention on the 2 outcome series. ARIMA procedures are well established in the literature on the impact of policy,^23,24 including alcohol policy,²¹ and so we provide only a brief description.

A fundamental concern associated with the evaluation of the efficacy of legislative and administrative initiatives is distinguishing their effects from other social processes that may be influencing an outcome series. ARIMA techniques, unlike simple pre- and postintervention means or percentage difference tests, explicitly take into account the potentially confounding effects of other causal mechanisms and, as a consequence, allow assessment of the change in the level of an outcome series independently of ongoing stochastic processes.²⁵

An ARIMA interrupted transfer function model consists of 2 parts. The first, the “noise” component, uses information from previous observations of an outcome series to model the systematic variation (i.e., autocorrelation) within the series. By applying the appropriate seasonal and nonseasonal differencing, along with estimating the appropriate seasonal and nonseasonal autoregressive and moving average parameters (i.e., prewhitening), one can separate the confounding influences of other causal processes from those associated with the intervention (in our case, the alcohol policy). Once a satisfactory noise component is identified and estimated, the intervention component is added to the transfer function equation. If the inclusion of a dummy series for the intervention (coded “0” for the period prior to the onset of the intervention and coded “1” beginning with the observation in which the intervention occurs and thereafter) increases the explanatory power of the model above and beyond that provided by the noise component (i.e., Granger causality), then we can conclude that the intervention significantly affects the outcome series.^25,26

Another advantage of ARIMA modeling techniques over simple pre- and postintervention change scores is that they allow for examination of the functional form of the relationship between an intervention series and an outcome series. Crude mean and percentage difference tests assume that the effect of an intervention is well represented as an abrupt, permanent change in the level of the outcome series (at least for the remainder of the observations for a given time series). Although one can estimate this functional form as a zero-order transfer function using ARIMA modeling techniques, one can also examine the relative fit of competing adjustment models. It is possible that the effect of an intervention gradually reaches a new level (e.g., it takes some time for the intervention to reach its full effect) or that there is an instantaneous but short-lived effect (often reflecting a publicity effect). A first order transfer function can be estimated to model the former pattern of change in the level of a series, and a pulse function can be estimated to model the latter.²⁵

ARIMA model building is an iterative process. We derived the final intervention models by successively estimating the noise and intervention components and subjecting them to a number of diagnostic tests. For the statistical details involved in the identification and estimation of the noise and intervention components of ARIMA interrupted times series models, we refer the reader to popular and readily available published sources.^25,27

RESULTS

Monthly male and female suicide deaths are shown in Figure 1. Mean monthly suicide deaths for the entire period were 3830, 3193, and 637 for total population, males, and females, respectively. Preintervention means were 4348, 3646, and 702, postintervention means were 3205, 2649, and 556, respectively. These raw data reveal decreases in mean monthly suicide deaths of 27% for males and 21% for females following the implementation of the 2006 Russian alcohol policy. However, to determine if and what proportion of these changes were in fact attributable to the alcohol policy or simply to ongoing patterns in these time series data resulting from other causes, we must estimate the ARIMA models.

FIGURE 1—

Monthly male and female suicide deaths: Russia, 2000–2010.

Note. Alcohol policy was introduced in January 2006. Pre- and postintervention means for males are represented by solid horizontal lines. The postintervention decrease attributable specifically to the policy is represented by the dashed horizontal line.

The left column in Table 1 provides information about the form and statistical adequacy of the final univariate models for the male and female time series. This shows that both the male and female series required (1) log transformation to induce variance stationarity, (2) first-order differencing and first-order seasonal differencing to remove drift, and (3) first-order moving average and first-order seasonal moving average parameters to remove autocorrelation. The Q statistics for the final noise models for both males and females met the criterion that none of the autocorrelations was significant at the .05 level.

TABLE 1—

Final Noise and Intervention Models for the Effect of the 2006 Russian Alcohol Policy on the Number of Monthly Suicide Deaths

Noise model	Intervention model
Male suicide deaths
ARIMA Lg(0,1,1)(0,1,1)12	Yt* = at + ω0It
(1 – θ1B)(1 – θ12B12) at = (1 – B)(1 – B12)Yt	It = 0 for observations 1–72
θ1 = −0.589, P < .001	It = 1 for observations 73–132
θ12 = −0.692, P < .001	ω0 = −0.096; P = .01
	Q = 21.9; df = 18; P = .24
Female suicide deaths
ARIMA Lg(0,1,1)(0,1,1)12	Yt* = at + ω0It
(1 – θ1B)(1 – θ12B12) at = (1 – B)(1 – B12)Yt	It = 0 for observations 1–72
θ1 = −0.721; P < .001	It = 1 for observations 73–132
θ12 = −0.758; P < .001	ωo = −0.041; P = .34
	Q = 36.0; df = 18; P = .11

Note. ARIMA = autoregressive integrated moving average; Lg = natural logarithm transformation; θ = moving average parameter; B = backward shift operator; ω₀ = zero-order input parameter of a transfer function; I_t = intervention series; Q = Ljung-Box test statistic for the null hypothesis that the model residuals are distributed as white noise.

The right column of Table 1 shows the final transfer function models assessing the impact of the 2006 Russian alcohol policy on monthly male and female suicide deaths. The results suggested a zero-order response best fit the data for each series. The Q statistic is the Ljung-Box test statistic for the null hypothesis that the model residuals are distributed as white noise (i.e., they are uncorrelated). The models showed no impact of the alcohol policy on female suicide deaths (Ln ω₀ = −0.041, P = .34). However, the intervention model for males showed that the alcohol policy resulted in an immediate and permanent monthly decline in suicide deaths for males (Ln ω₀ = −0.096, P = .01). In this case, we found it necessary to transform the male time series by its natural logarithm to induce variance stationarity prior to the estimation of the transfer function equations, and thus the parameter ω₀ is in the log metric. This can be interpreted as the ratio of the postintervention series level to the preintervention series level,²⁷ meaning that the ratio is expressed as the percent change in the expected value of the process associated with the alcohol policy:

In our case, %Δ = (e ^–0.096 – 1)100 = −9.15%. Thus, the intervention was associated with a 9.15% reduction (or 334 in absolute terms) in the monthly number of male suicides. In Figure 1, the solid horizontal lines represent the pre- and postintervention monthly means for male suicide deaths. The dashed line represents that part of the overall decrease attributed to the alcohol policy.

In short, the implementation of the 2006 Russian alcohol policy saved about 4000 male lives per year that otherwise would have been lost to suicide.

DISCUSSION

The suite of alcohol policies implemented in Russia in 2006 led to an immediate and permanent (i.e., for the length of our study) 9% reduction in male suicide mortality. There was no effect on female suicide mortality. We tested for alternative functional forms of the relationship between the alcohol policy and the response of suicide mortality, including (1) a more gradual impact over time relative to the immediate step change and (2) an immediate impact that dissipates within a few observations (i.e., a pulse function). Our results from these additional analyses revealed no evidence of either of these alternative functional forms and that the immediate step change model best fit the data. The impact on Russian suicide mortality of this set of alcohol policies is very similar in scope, size, and functional form as that found for a national alcohol policy in Slovenia by Pridemore and Snowden.²¹ They found the effect of the Slovenian policy to be limited to males, that the policy resulted in a 10% decrease in male suicide mortality, and that the effects were immediate and permanent.

The impact of this alcohol policy is not limited to suicide. In subsequent analyses not shown here, we estimated the impact of the policy on deaths from injuries of undetermined intent (Y10-Y34, Y87.2, Y89.9), the rate of which is abnormally high in Russia and which by definition will include undiagnosed suicide deaths. There was an immediate and permanent reduction in these deaths for both males (ω₀ = −304.69, P < .01) and females (ω₀ = −74.62, P < .01), in both cases representing a 10% reduction relative to preintervention levels. Our recent research also reveals that the 2006 alcohol policy led to immediate and permanent decreases in male deaths owing to transport accidents,¹³ and in male deaths attributed to alcohol poisoning and male and female deaths attributed to alcoholic cirrhosis of the liver.¹⁴ If there were effects on outcomes as serious as these various causes of mortality, we might also expect that the policy had beneficial effects for less serious but nevertheless important outcomes, including different types of morbidity, social behavior (including crime and violence), and economic indicators of health care costs, lost labor, etc.

Our finding that the 2006 Russian alcohol policy led to a reduction in male suicide mortality in the country is consistent with prior research. First, there is considerable evidence that alcohol policy can alter the level and pattern of drinking in populations.²⁸ Second, there is consistent cross-sectional and time series evidence of a population level association between alcohol and suicide in Russia that spans 3 centuries and successive Tsarist, Soviet, and Russian governments.^15,18–20 Third, the intense focus on harmful drinking and on the alcohol–suicide association in Russia led to the hypothesis that this association is especially strong in the country because of the intoxication-oriented consumption of distilled spirits, especially vodka,^15,16 which is the beverage of preference in Russia. In fact, time series research reveals Russian suicide rates to be associated with vodka sales but not with sales of beer and wine,^29,30 and that the population level alcohol--suicide association is stronger in nations, including Russia, that generally prefer distilled spirits.³¹ Similarly, prior research identified the extremely detrimental role of surrogate alcohols (e.g., colognes, medicinal products) in the Russian mortality crisis, especially among males,⁸ and this was noted even by President Putin in an April 25, 2005, address to the Federal Assembly. A goal of the anti-alcohol measures addressed in the present study was to reduce the availability of these alcohols. To the extent that they did so, this should also contribute to a reduction in alcohol-related harm, including suicide. Research has also shown that the population level alcohol–suicide association appears to be stronger in intoxication-oriented nations, including Russia.³² Finally, our findings are not only very similar to those showing an impact of a national alcohol policy on suicide mortality in Slovenia, but looking at Russia specifically Nemtsov³³ found that the prior, more encompassing anti-alcohol campaign of Gorbachev in the mid-1980s led to a reduction of blood alcohol-positive suicides in the country.

Our findings provide an important contribution to the public health and public policy literatures. First, Russia presents a rigorous test because of its high suicide rate and its high level and risky pattern of alcohol consumption, especially in the face of repeated attempts to curb harmful drinking in the country. If alcohol policy can be effective in this environment, then the prospects for success elsewhere are high. Second, both the programmatic efforts of federal agencies and the empirical literature on the association between alcohol and harm are becoming increasingly focused on individuals and on biology. Although such research is fundamental to our understanding alcohol’s effects on the body, our and related findings continue to show that public health, epidemiology, and public policy play a vital role in identifying specific risk and protective factors and in reducing alcohol-related harm. This is evidenced not only by the population-level association between alcohol and harm found in countless studies, but by the ability of population-level mechanisms like alcohol policy to reduce alcohol-related harm, including suicide. Simply put, policy can influence the amount and the pattern of alcohol consumption in populations,²⁸ which in turn is associated with alcohol-related harm.

Although the results presented in this study are encouraging, there are reasons to remain pessimistic about Russia. First, the high level and risky pattern of alcohol consumption in Russia remain. Second, Russian suicide mortality is still unacceptably high, especially for males, and there is little doubt that alcohol remains associated with it. Third, our analysis of the impact of this alcohol policy on several other types of mortality that might be expected to respond to this type of policy—including alcoholic cardiomyopathy, exposure to natural cold, homicide, and certain types of cancers—revealed no association, although we did find immediate and permanent decreases in male deaths attributed to transport accidents,¹³ and male deaths attributed to alcohol poisonings and male and female deaths associated with alcoholic cirrhosis of the liver.¹⁴ Fourth, even when summed, the annual number of deaths caused by suicide, injuries of undetermined intent, transport accidents, alcohol poisonings, and alcoholic cirrhosis of the liver prevented by the alcohol policy are a very small proportion (well below 10%) of the estimated 500 000 to 750 000 annual deaths in Russia attributable directly or indirectly to alcohol.^4,5 Fifth, research shows that some of the suite of policies contained in the overarching law were slow, ineffective, or otherwise problematic in their implementation.^11,34 Sixth, although meant to reduce alcohol-related harm indirectly, the implementation of the policy was largely devoid of public health elements. The focus of this set of policies was regulatory in nature—for example controlling the volume and quality of alcohol products and requiring registration of alcohol production and distribution facilities—with little attention given to attempting to change public perceptions of harmful drinking and its impact on individuals, families, neighborhoods, and society as a whole. Nevertheless, our findings do provide some hope; if the simple regulatory features of the 2006 law were able to have a welcome impact, then we might expect even better outcomes from a more comprehensive approach using the levers available to public health.

Limitations

A key threat to the validity of interrupted time series designs is history. In our case, this would be other policies or social forces that could be expected to result in a decrease in suicide mortality and that occurred at approximately the same time as the alcohol policy. We found no other population-level candidates for this possibility. A second alternative explanation might be that the decrease in Russian male suicide mortality is simply following a region-wide pattern that happened to coincide with the policy. This seems unlikely. Russia is an enormous region unto itself that is influenced by and shares borders with European, Central Asian, and Asian nations. Furthermore, as described above, other analyses revealed that the 2006 Russian alcohol policy was responsible for decreases in some other types of mortality (e.g., deaths from transport accidents, and from alcohol poisonings and alcoholic cirrhosis of the liver) that we would be expect to be affected by the policy but not otherwise expect to be associated with population-level suicide rates. Nevertheless, it would be worthwhile to undertake subnational analyses to determine where suicide rates are declining the most and how this might covary with the administrative effectiveness of the implementation of the policy. Finally, our aggregate data do not provide information on the blood alcohol content (BAC) of individual decedents. We might expect a decline in BAC-positive suicide deaths and no effect on suicide deaths where alcohol was not present, although such an approach has limitations as there are pathways for alcohol to affect suicide (e.g., family, employment, health) even if the person was not under the influence at the time of the suicide itself. Still, Nemtsov³³ has shown just such an effect, finding a strong correlation between changes in alcohol consumption and changes in BAC-positive deaths in Russia during the period 1965 to 1999.

Conclusions

We took advantage of a natural experiment to determine if an alcohol policy led to declines in suicide mortality. Although their use is still relatively rare, this and other types of natural experiments increasingly are being used in the social sciences and social epidemiology.³⁵ These designs allow for greater assurance when drawing causal inferences, and thus provide rigorous evaluations of the efficacy of policy and other public health interventions. We believe also that Russia, because of its history of a risky level and pattern of alcohol consumption and its high suicide rate, provides an important contextual setting for this type of analysis. Our results revealed an immediate and permanent 9% decline in suicide mortality among Russian males, and the scope, size, and functional form of this impact on suicide mortality was remarkably similar to that found for the effects of another recent national alcohol policy in Slovenia.²¹ The absolute size of this decrease means the policy saved more than 4000 male lives annually that otherwise would have been lost to suicide. Nevertheless, premature mortality and the rate of other health and social problems attributable directly or indirectly to alcohol in Russia remain extremely high, and officials must do more to reduce these burdens. Still, these and similar results provide an important contribution to the literature because they reveal the ability of public policy, public health interventions, and various other population-level mechanisms to reduce harm, even for seemingly individual-level behaviors like hazardous drinking and suicide.