In this issue of the Addiction, Meier, Purshouse & Brennan  (hereafter ‘the authors’) present modelling results of different alcohol pricing policies. Excessive alcohol consumption is related to many life-threatening diseases . Policies reducing alcohol abuse increase population health, with further beneficial effects to be expected outside the health care sector. The authors argue that heterogeneous modelling is needed to understand the differences in effects of policies affecting various types of alcohol consumption; for instance, consumption by youth in bars versus supermarket purchases by adult men. They present their outcomes as net present values, without paying explicit attention to the timing of effects. This commentary focuses upon the long-term effects of alcohol policies from an economic perspective. We argue that once different alcohol consumption subgroups are distinguished, the time dimension becomes crucial.
To illustrate the time dimension, let us consider the life of a hypothetical person called Joe. Joe works in a factory and each day after work drinks five beers. At weekends he goes to a bar with friends and has about 10 beers. At age 50 he suffers from a heart attack, which causes him to cut back on work for a few months. For the rest of his life Joe is on medication. He retires at age 65 and dies at age 70 from a fatal stroke. Joe's life could have taken a different course. Had he changed his alcohol consumption because of, for instance, a major tax increase, Joe would never have had a heart attack. In this parallel world of Joe, he would have gained quality of life as of age 50. Moreover, his boss would have been happy because Joe would not have needed to cut back on his work. He would not have had a stroke, but would have lived until 85, although suffering from dementia from age 75. That is, he would have lived longer and enjoyed more of his pension. However, in some of these so-called ‘added’ life years his quality of life would have been severely diminished. Of course, this is a stylized example, but it shows the important mechanism of postponement. By avoiding a heart attack, death is postponed. However, at later age substitute diseases cause losses in quality of life and add to health care utilization.
On a macro level a decrease in alcohol consumption leads to a decrease in the incidence of alcohol-related diseases, such as stroke and cardiovascular disease, but an increase in age-related diseases in a now older population. Similar mechanisms are at work in the case of tobacco control and overweight reduction [3,4]. With respect to health care costs, the increase in health care utilization might even outweigh the savings of avoiding alcohol-related diseases . If we broaden the scope and look at productivity, similar mechanisms are present. We saw in Joe's example that the avoidance of a heart attack resulted in net productivity gains, but also in additional years of retired life. In general, added years occur at older, usually less productive ages . While young people in general produce more than they consume, the reverse can be said of older people .
Turning to heterogeneity and combining it with the time dimension, all diseases related to alcohol occur more frequently at older ages, and therefore health effects are long-term effects. Different pricing policies affect different groups in society differently, as stressed by the authors. Policies that affect younger drinkers will have health effects further into the future than policies targeting older drinkers. Taking into account discounting of future effects, therefore, could mean that the same health effects would be valued less in a younger age group than those in an older age group . Ultimately, cost–benefit estimates could shed light on the value for money of the different policy options, accounting for all relevant health effects and related costs .
The main advantage of modelling is that by specifying mathematical relations between epidemiological quantities new insights and outcomes can be generated that cannot be estimated directly from data . Modelling can thus be seen as a form of evidence synthesis. The authors have presented an impressive modelling exercise in which they stress the importance of modelling heterogeneity. However, the modelling approach used by the authors does not allow us to quantify effects of policy over time, but rather attributes effects ex post. As we have argued above, the different policies probably differ in the timing of their effects. A next step therefore might be to quantify the dynamic effects of alcohol policies.