European Journal of Public Health, Vol., No., 64 68 ß The Author 1. Published by Oxford University Press on behalf of the European Public Health Association. All rights reserved. doi:1.193/eurpub/ckq11 Advance Access published on 24 February 1... Per capita alcohol consumption and alcohol-related harm in Belarus, 197 Thor Norström 1, Yury Razvodovsky 2 1 Swedish Institute for Social Research, Stockholm University, S-16 91 Stockholm, Sweden 2 Department of Psychiatry, Grodno State Medical University, Belarus Introduction Correspondence: Thor Norström, Swedish Institute for Social Research, Stockholm University, S-16 91 Stockholm, Sweden, tel: +46 8 16 23 14, fax: +46 8 46 7, e-mail: totto@sofi.su.se Received November 9, accepted January 1 Background: Although alcohol seems to be an important determinant of the mortality crisis in the former Soviet Republic of Belarus, little systematic research has been done on the relationship between alcohol consumption and harm at the aggregate level. The aims of the present study were to estimate the effect of per capita alcohol consumption on all-cause mortality, mortality from alcohol poisoning and hospital admissions for alcohol psychosis in Belarus. Methods: Annual data on the three outcomes and alcohol sale per capita for the period 197 were analysed using the Box Jenkins technique. Female mortality was included as a control variable and regarded as a proxy for other causal factors. To incorporate the lag structure, a weighted input was used in which a geometrical lag-scheme was applied. Results: The outcomes suggest that a 1 l increase in consumption was associated with an increase in male all-cause mortality of 2.3%. The corresponding figures for alcohol poisoning mortality and alcohol psychosis admissions are 12 and 2%. Conclusions: The present study strengthens the notion of alcohol consumption as an important determinant of population health in this part of the world, and thus the notion that alcohol control must be a key priority for Belorussian public health policy. Keywords: alcohol consumption, all-cause mortality, alcohol poisoning, alcohol psychosis, time-series analysis, Belarus... ince the 197s, there has been a steady increase in total Smortality in Belarus, interrupted only temporally in the mid-198s. There is a common belief that alcohol is a crucial factor in this mortality crisis in the former Soviet republics. 1 3 In Russia, for example, it has been estimated that alcohol may be responsible for >3% of all deaths, 2 while one recent study suggests that 43% of all deaths of male in the 2 4-year age range were attributed to hazardous drinking. 4 The high level of alcohol-related problems in the region is probably caused by a combination of the high overall level of alcohol consumption and harmful drinking patterns.,6 As for indications of harmful drinking patterns it may be mentioned that 6 8% of all alcohol in Belarus is consumed in the form of spirits 7 ; furthermore, findings from representative population surveys carried out in Grodno city suggest that 7% of men and 9% of women had a consumption pattern that was hazardous according to the AUDIT definition, while 28% of men and 2.8% of women were identified as being dependent. 8 These data are close to the results of the population study carried out in Arkhangelsk, Russia, which suggested that 61.9% of male and 2.7% of female industrial workers revealed scores on AUDIT corresponding to harmful drinking. 9 Unfortunately, there are no surveys for other points in time that would give a hint about the temporal pattern in such figures. In view of the high level of alcohol-related harm in Belarus, it is of interest to investigate to what degree harm rates are responsive to changes in overall consumption of alcohol. Since the latter is modifiable at least to some degree through alcohol political measures, such results would provide an assessment of to what degree these harms are preventable. In this article, we use Belarusian time-series data to quantify the impact of changes in per capita alcohol consumption on three different outcomes: all-cause mortality, mortality from alcohol poisoning and hospital admissions for alcohol psychosis. A second research question is to assess to what degree the increasing trend in harm rates can be accounted for by trends in per capita consumption of alcohol. These analyses extend previous aggregate-level studies based on Belorussian data. Razvodovsky 1 reported that changes in all-cause mortality were related changes in alcohol poisoning mortality as well as to changes in alcohol psychosis. 11 It seems plausible that both of these associations reflect influences from a common causal factor, i.e. per capita alcohol consumption. Methods In the analyses of all-cause mortality, we focused on male mortality as the outcome, while female mortality was included as a control variable. Female mortality is then regarded as an indirect measure of other causal factors that affect mortality in both sexes in a similar manner, such as advancements in medical technology and environmental factors in a broad sense. This is a more flexible version of the approach used in some studies in which male excess mortality is used as the outcome. 12 Cigarette sales were included as an additional control variable. Tobacco is a
Alcohol and alcohol-related harm in Belarus 6 well-established risk factor for numerous causes of death 13 ; furthermore, the use of alcohol and tobacco may well covary owing to their common economic determinants, such as income. A change in population drinking is expected to have an instantaneous impact on acute forms of harm (such as violent deaths), as well as a long-term effect on chronic harm, for instance liver cirrhosis. 14 We should thus expect that the impact of changes in drinking on all-cause mortality is distributed over a longer period of time. This phenomenon should be included in the modelling, or we are likely to underestimate the total effect. We will consider the time lag by using a weighted input series, where the lag weights are fixed a priori according to the following geometrical lag scheme: AW t ¼ A t þ A t 1 þ 2 A t 2 þ 3 A t 3 þþ n A t n We tried various values of the lag parameter () and found that a value equal to.3 was optimal, giving the best fit to the data. This indicates a faster mortality response to changes in population drinking than what has been found for other countries 17 where was equal to.7. A geometrical lag scheme was used for the cigarette indicator as well, but with equal to.8. 18 The lag scheme was truncated at lag, and the lag weights were rescaled to sum to unity. However, for the acute harm rates under study (mortality from alcohol poisoning and hospital admission for alcohol psychosis) no lag structure should be expected. All-cause mortality rate; Alcohol (litres) 1 In keeping with previous aggregate studies, we estimated semi-logarithmic models with logged output. For all-cause mortality the following model was estimated: rln MM t ¼ a þ 1 raw t þ 2 rcigw t þ 3 rln FM t þrn t where r means that the series is differenced, MM is male mortality, FM female mortality and a indicates the possible trend in mortality due to other factors than those included in the model. AW is the alcohol indicator and CIGW cigarette sales (both weighted as described earlier). The noise term, N, includes other etiological factors. The percentage increase in mortality associated with a 1 l increase in consumption is given by the expression: (exp ( 1 ) 1) 1. The noise term, N, includes other etiological factors. The models for alcohol poisoning mortality and incidence of alcohol psychosis include only the proxy for per capita alcohol consumption on the right-hand side. The observation period was 197. The data were analysed using the technique for time-series analysis that has been suggested by Box and Jenkins, 19 often referred to as ARIMA models. The method requires stationarity, while most of the series included exhibited strong time trends (figure 1). These were effectively removed by means of a simple differencing. That is, rather than analysing the relationship between trends in alcohol and harm rates, we analysed how annual changes in drinking impact on annual changes in harm. This procedure reduces greatly the risk of obtaining spurious correlations that are due to common trends. 4 3 3 2 1 Rates for alcohol poisoning and alcohol psychosis; Cigarettes 197 197 198 198 199 199 Male all-cause mortality Alcohol Alcohol poisoning Female all-cause mortality Cigarettes Alcohol psychosis Figure 1 Trends in male and female all-cause mortality per 1. Mortality per 1 from alcohol poisoning (males and females years), hospital admission per 1 for alcohol psychosis (males and females years), sales of cigarettes in hundreds per capita years and sales of alcohol in litres 1% alcohol per capita years
66 European Journal of Public Health After having estimated the effects of per capita alcohol consumption on total mortality, alcohol poisoning mortality and alcohol psychosis admissions, we proceeded to the second of our research question: to what degree can the alcohol factor account for the temporal variation in the three outcomes? To elucidate this, we compared the observed harm rates with the harm rates expected from the trajectory in alcohol consumption and the estimated alcohol effect according to: EH it ¼ c i expð i A t Þ where EH is the expected harm rate, the estimated alcohol effect parameter and A the alcohol indicator that was used in the estimation of the alcohol effect (in the calculation of the expected all-cause mortality the weighted construct, AW, was used). The constant (c i ) was chosen so that the observed and expected harm rates were equal for 197. Data on all-cause mortality, alcohol poisoning mortality (deaths per 1 population, years), hospital admissions for alcohol psychosis (per 1 population, years), alcohol sales (in litres 1% alcohol per capita years) and cigarette sales (per capita years) were obtained directly from the Ministry of Statistics of Belarus annual unpublished reports from 197 to. The National statistical agencies cause-of-death classification has been subjected to several changes over the last decades. Between 197 and 1988, Ministry of Statistics used the coding scheme based on ICD-8, and in 1989, ICD-9 was introduced. In 2, a new coding system, based on ICD-1, came into practice. There exists no assessment of the accuracy of the hospital admission data. However, its fairly close resemblance with the temporal pattern of alcohol poisoning mortality seems reassuring, as an earlier study confirmed the reliability of the mortality statistics for the Soviet and post-soviet periods. Inspection of the scattergram between the differenced alcohol and output series revealed a potential outlier (198), which might distort the outcome. We thus included a dummy variable that takes the value 1 in 198 and otherwise. Results As can be seen in figure 1, there was a steady increase in female as well as male mortality (1.7 and 2.1% per year, respectively). Alcohol poisoning mortality and incidence of alcohol psychosis were also increasing during the study period, although these indicators are much more volatile. Per capita alcohol consumption increased steadily during the first decade, then dropped markedly as a response to the anti-alcohol campaign (198 87), after which a new increasing trend was witnessed. Sales of cigarettes increased until the end of the 198s and then leveled off. Table 1 shows the outcome of the model estimations. Starting with the model for all-cause mortality, it is seen that the weighted alcohol indicator is clearly significant, but not the cigarette indicator. The estimated alcohol effect implies that a 1 l increase in per capita alcohol consumption is associated with an increase in male mortality of 2.3%. The corresponding figures for alcohol poisoning mortality and alcohol psychosis admissions are 12 and 2%; these estimates are also strongly significant. In actual numbers, these percentage increases would yield an additional 174 male deaths, 31 out of which would be due to alcohol poisoning, and 184 more alcohol psychosis admissions (these figures pertain to ). According to the diagnostic tests, the residuals from all three models were not different from white noise, and thus no noise parameters were needed. The results thus indicate that changes in per capita alcohol consumption affect changes in the three harm rates under study. But to what degree can the temporal pattern in consumption account for the major shifts and trends in harm rates? To elucidate this we now turn to the comparison of observed and predicted harm rates. The outcome (figure 2) shows that the marked decrease in allcause mortality in the mid-198s is mimicked reasonably well, while the following increase in mortality is poorly accounted for. From 1988, there was a 2.8% annual increase in total mortality, whereas the annual increase in predicted mortality was.7%, that is, one-fourth. The outcome is more or less the same with respect to the acute harm indicators; the trajectories before 199 are predicted reasonably well, whereas most of the sharp increase thereafter is poorly accounted for. Discussion We begin by addressing some potential limitations of the study that may have affected the outcome. First, it must be recognized that unrecorded alcohol comprises a considerable portion of overall alcohol consumption in the former Soviet republics, including Belarus. This is especially true for the transitional period after the collapse of the Soviet Union. Following the repeal of the state alcohol monopoly in 1992, the Belarusian alcohol market became highly fragmented, and the country was flooded by a wave of homemade, counterfeit and imported alcohol of low quality. 7 Using the indirect method suggested by Nemtsov, 21 the actual level of alcohol consumption in Belarus in 1999 has been estimated at 14 l of pure alcohol per capita. 7 However, there are no time-series data gauging unrecorded consumption that could be used to supplement the data on recorded consumption. The presence of unrecorded consumption represents a measurement error in Table 1 Estimated effects (ARIMA models) on all-cause mortality (males years), mortality from alcohol poisoning (males and females years) and hospital admission for alcohol psychosis (males and females years) All-cause mortality Alcohol poisoning Alcohol psychosis Effect SE Effect SE Effect SE Alcohol.23.6.111.22.221.3 Female mortality.71. 119 Cigarettes (weighted).162 ns.94 Dummy198.23 ns..86 ns.78.12 ns.16 Noise Constant.13..46.18.17.24 Diagnostics Q a () 2.74 (P >.74) 7.76 (P >.17) 8.88 (P >.11) Semi-logarithmic models estimated on differenced data for the period 197. a Box Ljung test for residual autocorrelation at lag. P <.1; P <.1; ns = not significant.
Alcohol and alcohol-related harm in Belarus 67 4 Rate for all-cause mortality 1 197 the explanatory variable (X) that may induce various forms of bias. Generally, a random measurement error yields a downward bias of the estimated effect of X. However, unrecorded consumption may also be systematically related to recorded consumption. This may bias the estimated alcohol effect upwards (if there is a positive correlation between recorded and unrecorded consumption) as well as downwards (negative correlation between recorded and unrecorded consumption). There is hardly any information we can use to assess which, if any, of these forms of bias is most likely. At any rate, it is clear that if there actually was an increasing trend in unrecorded consumption after the end of the 198s, as suggested above, there would be a corresponding underestimation of the projected alcohol-related mortality during that period. As indicated in figure 2, our findings suggest that this is indeed the case. Another potential source of bias is the influence of some omitted variable that is correlated with alcohol as well as mortality. As already noted, a possible candidate in this context is smoking. In his study based on Canadian data, Norström 16 reports that the estimated alcohol effect decreased from 2.9 to 1.7% per litre when tobacco was included in the model. The higher prevalence of smoking among Belarusian men probably explains part of their excess mortality. 7 However, in the present analysis, the inclusion of cigarette sales did not affect the estimate of the alcohol effect. This is in all probability explained by the fact that the changes in the alcohol and cigarette indicators are not correlated (correlation between differenced weighted series, r =.24, SE =.). It should be recognized that including female mortality as control variable may imply that some of the alcohol effect is controlled away, leading to a downward bias. The estimated alcohol effect should thus be regarded as conservative. 197 198 Male all-cause mortality Alcohol poisoning Alcohol psychosis 198 199 199 It is of interest to compare our findings with results for other countries, although such a comparison must be confined to all-cause mortality in the lack of estimates for the other harm rates. We may first contrast the annual 2% increase in male mortality with the average decrease in male mortality of 1% per year that was found for the 14 Western European countries included in the ECAS study. The difference is a reminder of the widening mortality gap between East and West. As to the relationship between population drinking and harm rates, previous findings exhibit a conspicuous pattern in the way that this relationship tends to be stronger in countries where the drinking culture is characterized by heavy drinking episodes, and thus weaker in countries where drinking is typically associated with meals and is more evenly spread over the days of the week. 22,23 Thus, Norström reported a stronger alcohol effect on all-cause mortality in northern Europe (3% per litre) than in mid- Europe and southern Europe (1% per litre). Subsequent findings confirm this pattern. 16,17 The estimate for Belarus (2.3% per litre) fits reasonably well into this pattern: it is markedly larger than the 1% estimate observed for countries in regions with a less hazardous drinking pattern (southern and mid-europe), and similar to those observed 17 in countries with more detrimental drinking patterns, 24 but it is still a bit lower than what was obtained for northern Europe. However, to fully appreciate the impact of the alcohol factor, the alcohol effect should be weighed together with exposure, i.e. the level of alcohol consumption. A feasible way of doing that is to compute the alcohol attributable fraction (AAF) on the basis of our aggregate findings. 2 Such a calculation (based on average consumption for the whole study period) yields an AAF equal to 16.9%; the corresponding figure for the period is 19.2%. That is, our estimates suggest that close to one-fifth of male mortality in Belarus is attributable to alcohol. 3 3 2 1 Male all-cause mortality predicted Alcohol poisoning predicted Alcohol psychosis predicted Figure 2 Trends in observed harm rates and harm rates predicted from trends in alcohol consumption Rates for alcohol poisoning and alcohol psychosis
68 European Journal of Public Health This is very close to the AAF (based on individual-level data) that Rehm et al. 26 report for Russia as of 2 (17.9%), a country that is on a par with Belarus with regard to consumption level and drinking pattern. In this context it may be mentioned that an even larger AAF (41%) was reported by Leon et al. 4 on the basis of a case control study undertaken in the industrial city Izhevsk. In conclusion, this study strengthens the notion that alcohol both in terms of consumption level as well as drinking patterns is an important determinant of population health in this part of the world and thus alcohol control should be a key priority for Belorussian public health policy. Acknowledgements Thor Norström gratefully acknowledges project grant (4-183) from the Swedish Council for Working Life and Social Research. Conflicts of interest: None declared. Key points Several studies indicate a relationship between per capita alcohol consumption and various alcoholrelated outcomes. However, there is a scarcity of such studies pertaining to the Eastern Europe. Based on analyses of time-series data for Belarus, our results suggest that a 1 l increase in consumption yields an increase in male all-cause mortality of 2.3%. The corresponding figures for alcohol poisoning mortality and alcohol psychosis admissions are 12 and 2%. The results strengthen the notion that alcohol consumption is an important determinant of population health in this part of the world, and thus that alcohol control should be a key priority for Belorussian public health policy. References 1 Leon DA, Chenet L, Shkolnikov VM, et al. Huge variation in Russian mortality rates 1984-94: artefact, alcohol, or what? Lancet 1997;3:383 8. 2 Nemtsov AV. Alcohol-related human losses in Russia in the 198s and 199s. Addiction 2;97:1413 2. 3 Razvodovsky YE. Alcohol-related problems in Belarus. Alcologia ;12:1 14. 4 Leon DA, Saburova L, Tomkins S, et al. Hazardous alcohol drinking and premature mortality in Russia: a population based case-control study. Lancet 7;369:1 9. Stickley A, Leinsalu M, Andreev E, et al. Alcohol poisoning in Russia and the countries in the European part of the former Soviet Union, 197 2. Eur J Public Health 7;17:444 9. 6 Pridemore WA. Heavy drinking and suicide in Russia. Soc Forces 6;8:413 3. 7 Razvodovsky YE. Alcohol and mortality crisis in Belarus. Grodno: Medical University Press, 3. 8 Razvodovsky YE. A psychometric analysis of the Russian version of the AUDIT. Alcohol Alcohol ;38(Suppl l):31. 9 Nilssen O, Averina M, Brenn T, et al. Alcohol consumption and its relation to risk factors for cardiovascular disease in the north-west of Russia: the Arkhangelsk study. Int J Epidemiol ;34:781 8. 1 Razvodovsky YE. All-cause mortality and fatal alcohol poisoning in Belarus, 197. Drug Alcohol Rev 8;27:62. 11 Razvodovsky YE. Alcohol psychoses and all-cause mortality in Belarus. Adicciones 8;:39 46. 12 Bandel R. Die spezifische Mannersterblichkeit als Masstab der Alkoholsterblichkeit. Ergebnisse der Socialen Hygiene und Gesundheitsfursorge, Bd II [Specific male mortality as measure of alcoholrelated mortality]. In: Results of Social Hygiene and Health Care. Leipzig: Georg Thieme Verlag, 193, 424 92. 13 Ezzati M, Lopez AD, Rodgers A, et al. Selected major risk factors and global and regional burden of disease. Lancet 2;36:1347 6. 14 Skog O-J. The risk function for liver cirrhosis from lifetime alcohol consumption. J Studies Alcohol 1984;4:199 8. Norström T. Per capita alcohol consumption and all-cause mortality in 14 European countries. Addiction 1;96:113 28. 16 Norström T. Per capita alcohol consumption and all-cause mortality in Canada, 19 98. Addiction 4;99:1274 8. 17 Norström T. Alcohol consumption and all-cause mortality in the United States, 19-2. Contemp Drug Prob 7;34:13 2. 18 Hemström O.Male susceptibility and female emancipation: studies of the gender difference in mortality. Stockholm, Sweden: Almqvist & Wiksell, 1999. 19 Box GEP, Jenkins GM. Time series analysis: forecasting and control. London: Holden-Day Inc., 1976. Wasserman D, Varnik A. Reliability of statistics on violent deaths and suicide in the former USSR, 197 199. Acta Psychiat Scand 1998;98(Suppl 394): 34 41. 21 Nemtsov A. Estimates of total alcohol consumption in Russia, 198-1994. Drug Alcohol Depend ;8:133 42. 22 Norström T, HemströmÖ, Ramstedt M, et al. Mortality and population drinking. In: Norström T, editor. Alcohol in postwar Europe: consumption, drinking patterns, consequences and policy responses in European countries. Stockholm: Almqvist & Wiksell, 2, 7 76. 23 Norström T, Ramstedt M. Mortality and population drinking: a review of the literature. Drug Alcohol Rev ;24:37 47. 24 Rehm J, Taylor B, Patra J. Volume of alcohol consumption, pattern of drinking and burden of disease in the European region. Addiction 6;11:186 9. 2 Norström T. The use of aggregate data in alcohol epidemiology. Br J Addict 1989;84:969 77. 26 Rehm J, Sulkowska U, Manczuk M, et al. Alcohol accounts for a high proportion of premature mortality in central and eastern Europe. Int J Epidemiol 7;36:48 67.