Alcohol control in its various guises potentially decreases the burden of disease substantially and thus is an important instrument for health-care policy. Alcohol policies include measures to reduce the availability of alcohol, alcohol taxation, restricting the sale of alcohol, regulating the drinking context, restrictions on alcohol marketing, drink-driving counter-measures, education and persuasion, and early intervention and treatment services [6]. Alcohol tax increase is a public policy tool that traditionally falls outside the scope of health policy. The reason for this might be that alcohol taxes usually are controlled by ministries of finance and that tax revenues are not part of the health-care budget. A notable exception in this respect is Thailand where excise taxes on alcohol and tobacco are used to fund major health-care programmes [7]. Still, from a public health perspective alcohol taxes are an important instrument, as they are known to be able to decrease alcohol consumption [6,8]. Tax increases have been shown to decrease, for instance, cirrhosis mortality and drink-driving deaths [2].

Since in most Western countries retail prices of alcohol are heavily influenced by alcohol excise taxes [3], an increase in alcohol taxes normally leads to an increase in prices of alcohol, which in turn leads to a decrease in the demand for alcohol. To measure the effect of price increases on alcohol consumption, economists estimate the price elasticity that indicates how much the consumption of a certain good changes if its price is changed. Clements et al [9] report price elasticity figures for seven countries (Finland, Sweden, Norway, UK, Canada, Australia and New Zealand), covering the period from the mid-1950s to the mid-1980s [9]. They find a price elasticity of -0.35 for beer, -0.68 for wine and -0.98 for spirits. These price elasticity figures imply that if, for instance, beer prices increase by 10%, beer consumption decreases by 3.5%.

So, increasing alcohol taxes may be an effective strategy to control alcohol consumption, but is it also cost-effective? The information needed to answer this question can be provided by an economic evaluation. An economic evaluation is a comparative analysis of the costs and effects of two or more interventions [10]. The outcome of an economic evaluation is expressed as a ratio of incremental costs to incremental effects: the incremental cost-effectiveness ratio (ICER). To our knowledge, the only previous study on the cost-effectiveness of alcohol taxation was conducted within the World Health Organization (WHO)-CHOICE project [11,12]. They found that alcohol control policies, particularly tax increases on alcohol, are cost-effective relative to other health interventions. However, in their estimates of the cost-effectiveness of alcohol taxes, they only took into account health effects derived from heavy drinkers and ignored the effects of taxation on the health of moderate drinkers.

To understand the effects of alcohol tax increases on health-care costs, Figure ​Figure33 displays differences in health-care costs between the Dutch and the current practice scenarios for mean values of the input parameters as displayed in Table ​Table2.2. The decrease in the incidence of diseases causally related to alcohol results in a decrease in health-care costs of those diseases. However, the gain in life-years causes an increase in the prevalence of all diseases unrelated to alcohol. From Figure ​Figure3,3, it can be seen that the savings in health-care costs of alcohol-related diseases are outweighed by increases in the health-care costs of diseases not related to alcohol in life-years gained.

Figures ​Figures44 and ​and55 display cumulative differences in costs and effects (both discounted) of the two alcohol tax scenarios compared with no tax increase over a period of 100 years for all values of the input parameters. This figure illustrates the strong correlation between health gains and health-care costs. This is due to the fact that the additional health-care costs are solely the result of increases in life expectancy from a reduction in alcohol consumption. Thus, the more QALYs gained the more additional health-care costs. The health gains in the Swedish scenario are much larger than in the Dutch scenario because the price increases are much higher in the Swedish scenario.

Table ​Table33 shows cumulative differences in health gains, cost differences and incremental cost-effectiveness ratios. Although the Swedish scenario results in many more costs and QALYs than the Dutch scenario, the costs per QALY are approximately equal. Costs per QALY are higher than costs per life-year gained in both scenarios.

Figure ​Figure66 displays the cost-effectiveness acceptability curves (CEAC) for the alcohol tax increases. A CEAC shows the probability that an intervention is cost-effective for different values of the threshold, that is, for different monetary values placed on a QALY. What can be derived from Figure ​Figure44 is that, if QALYs are, for instance, valued at €5000, a tax increase is cost-effective with a probability of 0.5. However, if society is willing to pay €10,000 per QALY (which is below the threshold value of €20,000 per QALY that is frequently cited in the Netherlands [23,24], and far below most other estimates of the value society is willing to pay for a QALY [25]), a tax increase is cost-effective with a probability of almost one.

Although in theory an alcohol tax increase can be implemented by legislation alone, some administrative costs and possible costs of law enforcement to keep smuggling to a minimum have to be made to successfully implement a tax increase. These costs, as well as additional tax revenues, are usually carried by sectors outside the health-care sector. Therefore, for an alcohol tax increase, taking the health-care perspective effectively illustrated the advantages compared with interventions whose costs traditionally fall inside the health sector, such as treatment of alcohol addiction or curative treatments in a hospital setting.

Like all economic evaluation modelling studies, the calculated costs, effects and cost-effectiveness ratios are conditional on various assumptions. Below we will discuss some of the assumptions we made to estimate the cost-effectiveness of increasing alcohol taxes. First, price elasticity is the same for moderate, excessive and dangerous drinkers. If, for instance, dangerous drinkers react less to price changes than moderate drinkers [3], health effects of tax increases may be smaller than we estimated. However, there are contradictory findings on the exact price elasticity of dangerous drinkers [26,27]. Second, all tax increases were translated into price increases. We did not take into account the possibility that producers do not pass on the tax increase to consumers, resulting in a decrease in their profit margin. However, if producers only partially pass on the tax increase to consumers, this will only decrease the amount of health gains but not influence the cost-effectiveness. Third, the effects of a tax increase on alcohol consumption will be sustained in the long run. This assumption is built on studies that argue that, since alcohol consumption is addictive, the long-run price elasticity is significantly higher than the short-run elasticity. Fourth, the price elasticity is the same for high as for small price increases. This is due to the fact that estimates of price elasticity are estimated on time series with mainly small price variations over time. Again, this assumption is probably more important for the estimation of the amount of health gains rather than the cost-effectiveness ratio. Fifth, we used price elasticities based on Clements et al [9]. We preferred to use specific Dutch data, but the available estimates [28] were not specific for the different alcohol products (beer, wine, spirits). In an analysis that did not include beverages sold in restaurants, bars and hotels, Leppänen et al [28] found a high value of total elasticity for alcohol demand in the Netherlands (-1.5). However, in another analysis, they estimated the price elasticity for alcohol in the Netherlands at -0.53, which is close to the weighted average in our analysis that was based on the Clements elasticities (-0.61). Nevertheless, assuming higher values of price elasticity would result in more QALYs gained and increased health-care costs due to increases in life expectancy. For instance, modelling an elasticity of -1.5 for all alcohol types, 13,000 QALYs will be gained at costs of €65,000,000. So, while both QALYs and costs will increase, the ICER would remain almost unchanged.

In the comparative quantification of health risks study, health effects of average alcohol consumption and patterns of drinking were estimated separately [29]. In this study we have limited ourselves to the effects of a tax increase on average alcohol consumption. This was done because the CDM models average drinking and the demand elasticity estimates for alcohol also refer to average alcohol drinking. Data on the influence of patterns of drinking are less available than data on overall consumption, but evidence is accumulating that patterns of drinking affect the link between alcohol and disease and mortality [29]. For example, the same overall average volume of alcohol can be consumed in small quantities regularly with meals (for example, two drinks a day with meals) or in large quantities on a few occasions (for example, two bottles of wine on a single occasion every Friday). This also implies that we did not model the effects of a tax increase on alcohol dependence, which is a disorder in itself. The simulation model we employed did not model all diseases considered to be related to be alcohol consumption separately. Some were only modelled indirectly through an elevated mortality risk. This means that we may have underestimated the impact of alcohol consumption on quality of life and health-care costs and have overestimated the cost-effectiveness ratio. Furthermore, the relative risks employed in the CDM are not based on the most recent meta-analyses. However, the study by Holman et al [18] was the only study that included relative risk estimates for the alcohol categories employed in the CDM for both diseases and mortality, and that provided estimates of all-cause mortality. Such a category of all-cause mortality was not used in other studies on the relative risks of alcohol consumption [23]. In the simulation model we used these estimates of all-cause mortality to estimate the effects for the causes of death that are not explicitly in our model. Moreover, a recent meta-analysis of relative risks on all-cause mortality yielded similar estimates [19], and a recent study from the comparative risk assessment collaboration group used comparable estimates of relative risks for specific alcohol-related diseases [30].