Why We Should Do it Even Though We Shouldn’t Pay Too Much Attention to the Bottom-line Results
Paper presented at the annual meeting of the Kettil Bruun Society for Social and Epidemiological Research on Alcohol, Toronto, May 2001 by ERIC SINGLE and BRIAN EASTON.[1]
Keywords: Health Economics
Abstract
A coalition of provincial, national and international addictions agencies has sponsored a series of international symposia leading to the developing of international guidelines for estimating the costs of substance abuse. These guidelines have now been utilized in national studies in four continents, with more consistent and comparable results than in previous studies. Although the bottom-line results have been utilized to argue for alcohol issues having a higher place on the public policy agenda, the real value in such studies lies in the detailed results regarding mortality and morbidity attributable to alcohol, the relative contribution of acute vs. chronic conditions to overall problem levels, and the role of alcohol in adverse social consequences such as crime and economic productivity. Recent updated estimates are presented regarding the attributable proportion of various causes of disease and death due to alcohol misuse in Canada. There are a variety of factors which undermine the robustness of the findings, including lack of data, laying of assumptions and changes in the epidemiological knowledge base. It is argued that economic cost estimates should nonetheless be conducted and continually refined, as the detailed findings are of great utility to the design and targeting of prevention programming and policy.
Introduction
The costs of alcohol misuse represent an issue of key interest to stakeholders, policy makers and the media. Knowledge of the costs associated with the problems caused by alcohol informs decisions related to funding and to interventions to prevent or reduce adverse consequences. The purpose of this paper is to briefly describe current efforts to improve the methodology of cost estimation and discuss the many problems to be overcome to achieve more robust and comparable results.
Our major thesis is that while there are indeed many difficulties regarding the “bottom-line” estimates of total economic costs attributable to alcohol use, economic cost studies are nonetheless worthwhile research exercises that should be undertaken. As detailed below, there are indeed problems regarding cost estimates. But these are the problems of all most any aggregate statistical index including such widely used ones as a mortality rate or Gross Domestic Product. That we have called them an ‘index’ marks the issue. They combine literally millions of unique and idiosyncratic activities into a single measure, which has the property that in some sense the numbers rank the size of the phenomenon. Higher mortality rates imply relatively more deaths, higher GDP imply relatively more production. The implication is not simply because of the index construction, but because underlying the construction is a theory, which describes how the index can be constructed. The construction may not be perfect, and the theory itself often warns of the limits of interpretation. But if we understand these limitations, we not only understand the weaknesses of the indices. For all their failings, what serious demographer would abandon mortality rates, what serious economist would throw out GDP?
The same applies to the estimates of the social costs of alcohol and drug abuse. It is a single index that combines myriads of individual actions. Without it we are forced back to anecdotes, which may be misleading. As harsh as may be the experience of alcohol dependent persons and the communities in which they live, the implication of the measures of social costs of alcohol abuse is that, in the aggregate, alcohol abuse is an even more serious problem than suggested by these experiences. The cost estimates sum up all the individual anecdotes and provide a framework for comparing the aggregate societal impact of alcohol versus other psychoactive substances. While the problems experienced by narcotic users may be more severe in comparison to those experienced by the average person who experiences a problem due to his or her drinking, in total there are many more of the latter.
Alternatively we may use a less complex index, which again may be somewhat misleading. For example, examination of estimates of mortality attributable to substance abuse indicate that there are many more deaths from tobacco use than alcohol misuse, suggesting that tobacco is a far more serious problem than alcohol. But gross death rates ignore that alcohol-induced deaths typically occur at a much younger age than tobacco-induced deaths, and in terms of total years of life lost as a result of the abuse the two sources are much more similar. It also turns out the material losses from alcohol due to accidents and policy costs for the enforcement of alcohol laws are markedly higher than tobacco in most of the countries for which there have been analyses. Thus mortality rates are a useful indicator of the severity of particular forms of substance abuse, but economic cost estimates provide a more complete index.
The theory underpinning the social costs methodology is a well proven one, arising out of the neoclassical revolution of the end of the nineteenth century, and based on developments which began in the middle of the twentieth century in the measurement of GDP and the application of cost benefit analysis. Thus, the theory is the one that is probably used by the government economic advisers in most Western type democracies in virtually every practical public spending decision, as well as many other policy decisions. Cost estimates are in principle useful for public policy purposes, although there are practical limitations at its present state of development – limitations from data inadequacies and limitations because of theoretical problems arising from the applications of the theory. These limitations must be identified if we are to remedy them. It is these limitations that this paper addresses.
1. Estimating the costs of substance misuse: background
Studies that estimate the costs of substance abuse are still uncommon and relatively few countries have attempted to estimate these costs. In the past, estimates of the costs of substance abuse have often been fraught with methodological difficulties resulting in widely varying estimates.
In May 1994 the Canadian Centre on Substance Abuse (CCSA) organized an international symposium in Banff, Canada, to discuss the issues involved in estimating the social and economic costs of substance abuse, and to seek a consensus on the most appropriate model. The purpose of the meeting was to explore the feasibility of establishing an internationally acceptable common methodology for estimating the costs of drugs. The symposium in Banff brought together persons with experience and expertise in dealing with the issues of costs estimation. A working group was formed to explore the potential for developing guidelines on estimating the economic costs of substance abuse, consisting of myself, David Collins, Brian Easton, Rick Harwood, Helen Lapsley and Alan Maynard. These guidelines were drafted and finalized at a second International Symposium held in Montebello, Quebec, in 1996 (Single, Collins et al., 1996). In June 2000, the Third International Symposium on Estimating the Social and Economic Costs of Substance Abuse was held in Banff. This meeting focused on ongoing methodological issues and the expansion of cost estimation methods to developing economies. The guidelines are currently being revised and expanded to reflect the outcome of this meeting.
Studies have been carried out in Australia (Collins and Lapsley, 1995), Canada (Single et al., 1998) and the US (Harwood et al., 1999) utilizing these guidelines. Although the Canadian and American studies use a human capital approach to measure indirect productivity costs while the Australian studies uses a demographic approach, all of these studies use the Cost-of-Illness (COI) approach advocated in the guidelines.
Time does not permit a detailed comparison, but it is noteworthy that the results of all three studies show substantial material impacts resulting from alcohol misuse. In each case the largest cost is indirect productivity losses followed by health care costs and law enforcement costs. There appears to be some convergence in the results over time, with more comparable per capita costs between societies, possibly reflecting to some extent the agreement to use comparable methodology. The major divergency in results regarding alcohol costs is the higher per capita costs of alcohol in the US study compared with that of the Canadian study. However, it appears that most of the higher per capita estimates in the US study are not due to the costing methods but rather due to differences in the methods used to estimate alcohol-attributed morbidity and mortality
2. Sources of error in the bottom-line economic cost estimates
There are many reasons why one might choose not to give much attention to the bottom-line estimates of the overall economic costs attributable to alcohol in these studies. First, although the cost study symposia and the development of international guidelines have done much to increase comparability of results, there remains a lack of consensus regarding the appropriate methodology to employ in conducting cost estimation studies. The more commonly used Cost-of-Illness approach has been criticized for including indirect costs such as productivity costs, and some prefer the more conservative “externality” approach championed by US economist Willard Manning. Even among those who employ a COI approach, there are differences regarding the valuation of premature mortality caused by alcohol misuse. While most studies continue to use the human capital approach (which uses foregone income to estimate foregone productivity), Collins and Lapsely have developed an alternative “demographic” approach and several studies use new “willingness-to-pay” techniques (e.g., tobacco costs have been estimated in this fashion by Brian Easton in New Zealand and Claude Jeanrenaud in Switzerland). Exactly how to include the ‘human’ dimension is still a matter of discussion. Most studies recognize that there is a loss of actual or potential output as a result of morbidity and mortality. This can be measured by either the human capital approach (used in the US and Canadian cost studies) or the demographic approach (used in the Australian studies), which reflect different ways of dealing with the time element. However there is survey evidence from the willingness to pay studies, which supports the philosophical stance that we value life above that of the lost of production (so that the retired are considered of social value even though they are not producing). How to incorporate this into the social cost study is not fully agreed. However it is totally agreed that if there is an allowance for the value of life above that of the loss of output, then it is quite wrong to compare the full social cost estimate with that of GDP since the latter only applies output.
Second, data are invariably lacking on cost items in almost all cost studies. This is particularly true for developing economies which often lack reliable reporting systems. Even in developed economies, there is sparse information on many cost elements. The proportion of crime attributable to alcohol is highly contentious. There is a lack of data on the costs of specific alcohol-related productivity problems such as absenteeism, job turnover, lower on-the-job productivity due to alcohol use, alcohol-attributable disability and so forth. In some countries, estimates of alcohol consumption do not exist for the year under investigation, and must be estimated by interpolating the prevalence from other years. It is frequently difficult to determine from government budgets what proportion of policy costs (prevention, research and law enforcement costs) that can be attributed to alcohol. In the absence of complete data, decisions have to be made because to ignore a cost is in effect to count the cost as zero, which is generally more erroneous than making a decision on the basis of incomplete information.
Third, even when relatively complete data are available, the prevailing methods for estimating economic costs attributable to alcohol use involve a layering of multiple assumptions. For example, estimates of alcohol-attributable morbidity and mortality are required to underpin estimates of productivity costs and costs to the health care system. The morbidity and mortality estimates are made by combining information on (a) the relative risk of consuming alcohol at different levels to various causes of disease and death from meta-analyses of the epidemiological literature with (b) prevalence data on the number of persons consuming alcohol at levels associated with a higher relative risk in order to generate (c) etiologic fractions of the proportion of all such causes of disease and death that can be causally ascribed to alcohol use. These etiologic fractions are then applied to the reported number of hospitalizations and deaths by cause to estimate morbidity and mortality attributable to alcohol use. This procedure must necessarily make the following assumptions:
It is assumed that all alcohol-related causes are included. This may not always be the case, depending on many factors. Even the best meta-analysis of alcohol-related health consequences (English et al., 1995) failed to consider excessive cold, probably because it is virtually unknown in Australia. In the Canadian territories, however, freezing to death accounts for 20% of overall morality and it is frequently attributable to alcohol misuse.
Some causes involve both alcohol and other causes (e.g., drug interactions, injuries stemming from fires involving both smoking and alcohol intoxication) and arbitrary decisions must be made concerning the division of attribution.
It may be assumed that relative risk estimates from studies in one country can be used to estimate relative risk in another, when there are no local data for that country.
It is assumed that confounders are adequately controlled for in the studies used to estimate relative risk.
It is assumed that age and gender are adequately controlled for in estimating relative risk.
In most cases (but not in the Canadian study), the estimates of relative risk are derived from both morbidity and mortality studies, thus assuming that the risk of morbidity is equivalent to the risk of mortality.
It is assumed that the reported number of hospitalizations and deaths is accurately counted and complete, and that the cause is accurately recorded. We know, for example, that some disorders only recently described in the medical literature (such as fetal alcohol syndrome) are not being reliably recorded yet.
Even when one has reasonably accurate estimates of alcohol-attributable deaths and hospitalizations, there are yet other assumptions involved in arriving at estimates of health care costs. In most studies, a per diem cost is used to estimate hospitalization costs on the assumption that costs for alcohol-related conditions are similar to other conditions. Depending on the cost study, there is also often a need to use cost estimates from a particular region to make national estimates. Thus, for example, in the Canadian study we had very good estimates for some health care costs such as prescription drugs for some provinces that were then applied on a per case basis to other provinces.
The estimation of productivity costs also entails considerable assumptions. There is remarkably little information regarding specific negative consequences of alcohol use in the workplace such as the extent of tardiness, absenteeism, employee turnover and lower on-site productivity due to impairment. Sometimes the cost estimation studies must assume that these productivity costs can be estimated by considering the lower wages of alcohol dependent persons. This is at best a crude approximation of productivity costs and rests on the assumption that wages are a true reflection of productivity. Furthermore, the choice of discount rate–the rate at which future earnings are “discounted” or converted into current dollars–can have a very large impact on the total cost estimates (Single et al., 1998).[2]
Even law enforcement and criminal justice costs, which derive mainly from government budgetary data, are subject to certain assumptions. For example, police costs due to alcohol misuse are typically estimated by examining the proportion of offences (violations that are officially drawn to the attention of the police) that are due to alcohol and multiplying this percentage by total police costs. Court and corrections are similarly estimated in terms of percentage of total charges or jail sentences. This assumes that the amount of police, court and corrections time and resources spent on alcohol offences is the same as for other offences.[3] Thus, as with other types of costs, the estimation of criminal justice costs involves considerable assumptions.[4]
The fourth major reason for caution in interpreting the bottom-line cost estimates concerns changes in the epidemiological database and what we know about the effects of alcohol use. There are constant improvements in diagnostic practices. As noted above, the best method currently available for estimating alcohol-attributable morbidity and mortality relies on reliable diagnoses of alcohol-related causes of death and hospitalization. Conditions such as fetal alcohol syndrome have only recently been described and accepted in the medical literature, and such conditions will likely be underreported for some time. More importantly, new research is continually emerging concerning the link between alcohol use and various causes of disease and death. Ten years ago, there was insufficient data to conclude that there is a causal connection between alcohol use and breast cancer. Now the evidence is compelling. Although alcohol accounts for less than 3% of breast cancer fatalities in Canada, it is such a big killer that it represents the third leading cause of alcohol-attributable death among women (Single et al., 2000). A study conducted just one decade earlier would likely have not even included breast cancer in the cost calculations.
3. Why do cost studies if the results are so uncertain?
Thus, even with improvements in methods for estimating alcohol-related mortality, morbidity and economic costs, there are still significant sources of error in the bottom-line estimates of total economic costs caused by alcohol misuse–errors arising from incomplete data, layering of assumptions and changes in the epidemiological data base. Given these uncertainties, why do cost estimation studies at all?
There are several responses to this question. The first answer is that doing the exercises exposes data deficiencies, and forces us to improve the statistical base and our understanding of the processes involved. Economic cost studies help to identify information gaps, research needs and desirable refinements to national statistical reporting systems. There is no better way to lay out a national research agenda that to conduct a cost estimation study. Cost studies are an excellent device for the identification of data development and research needs. For example, the Canadian cost study identified a strong need for improved estimation of the proportion of crime that can be attributed to alcohol and drug misuse and spawned a study on this topic currently being undertaken by Kai Pernanen and Serge Brochu.
The second concerns quality control. Policy makers need and use cost estimates, explicitly or implicitly, in set priorities among competing concerns. Despite all of the uncertainties involved, economic cost estimates are frequently used to argue that policies on alcohol and other psychoactive substances should be given a high priority on the public policy agenda. The public is entitled to a quality standard against which individual cost estimation studies can be assessed. Without such a standard there will be a tendency by the advocates for each social problem to overbid, adding in additional items to make their concern a suitably high (even exaggerated) number.
Third, cost estimates help to appropriately target specific problems and policies. It is important to know which aspects of alcohol misuse involve the greatest economic costs, what specific problems are most likely to occur and in what demographic or geographic groups. The nature and magnitude of costs draw our attention to specific areas which need public attention, or where specific measures may be effective. For example, Table 1 presents the relative risks and etiological fractions for alcohol-attributable causes of disease and death in Canada in 1992, while Table 2 presents updated estimates of alcohol-attributed hospitalizations and deaths for 1995 in Canada. The information on relative risk and the proportion of cases for each cause that is attributable to alcohol is of considerable interest in and of itself. It is important for clinicians and for prevention programming to know the proportion different types of cancer or the proportion of accidents that are caused by alcohol use. It is noteworthy that acute causes account for nearly one half of the total estimated alcohol-attributed mortality and approximately two third of years of life lost. The relative importance of acute consequences has significant implications to alcohol policy and prevention programming.
Table 1: Relative Risks and Etiologic Fractions For Conditions Partially Attributable to Alcohol, Tobacco and Illicit Drug Use, by Cause & Gender, Canada, 1992 Table Available from either author
Table 2: Deaths, potential years of life lost and hospitalizations due to alcohol, tobacco and illicit drugs by cause and gender, Canada, 1995 Table Available from either author
Last but not least, the development of improved estimates of the costs of substance abuse offers the potential, although generally not yet realised, to provide baselines measures for more sophisticated economic analyses to determine which policies and programmes are the most effective in reducing the harm associated with alcohol and other drug use. For instance, the costs studies in some countries are raising the significance of the interaction between alcohol and crime and leading to greater policy attention to this nexus. Ultimately, cost estimates could be used to construct social cost functions for optimal tax policy and national target setting.
The concept of Gross Domestic Product (GDP) was subject to much the same sources of error and criticisms that economic cost estimates currently face. Despite similar issues of lack of complete data and layering of assumptions, the estimation of GDP has been continually refined and improved, and it has become one of the most useful tools for economic analysis and policy development. Through international cooperation and the development of an on-going process to continually update and refine the methodology, economic cost studies will similarly become more reliable over time, and perhaps even become an important cornerstone for comparative analyses of alcohol policy and interventions much as GDP is today.
Conclusion
In conclusion, bottom-line economic cost estimates are not the sole consideration in the determination of political priorities, nor should they be. There are a host of other considerations that policy makers and others involved in policy development must necessarily consider. But economic cost studies provide an extremely useful framework for identifying key leverage points in policy development and for the development of priorities in research as well as for treatment, prevention and other interventions. The cost estimation symposia and international guidelines have done much to reduce differences in economic modeling and enhance the comparability of results of cost estimation studies in different countries. Most of the remaining sources of error in cost studies reflect data deficiencies that research must necessarily address if we are to achieve more effective interventions and outcomes. Despite the many sources of error in current cost estimates, we believe that it is indeed a valuable research exercise to estimate the costs of alcohol misuse and continually improve upon the methodology for conducting such studies. And ultimately, if we do not do these cost studies well, others will do them badly, compounding the sorts of problems we have described here, while failing to provide the sort of benefits that good studies can do, and so simply adding to the confusion in a field which is already very difficult.
References:
Collins D. & H. Lapsley (1995) Estimating the Economic Costs of Drug Abuse in Australia. Canberra: Australian Publishing Services.
Harwood, H et al (1999) The Economic Costs of Alcohol and Drug Abuse in the United States-1992, Washington: National Institute on Drug Abuse (available on their website: www.nida.nih.gov/EconomicCosts)
English D, D. Holman, E. Milne E, M. Winter, G. Hulse, G. Codde, G. Bower, B. Corti, C. De Klerk, G. Lewin, M. Knuiman, J. Kurinczuk & G. Ryan (1995) The Quantification of Drug Caused Morbidity and Mortality in Australia, 1992. Canberra: Commonwealth Department of Human Services and Health.
Single E, D. Collins, B. Easton, H. Harwood, H. Lapsley & A. Maynard (1996) International Guidelines on Estimating the Costs of Substance Abuse Ottawa: Canadian Centre on Substance Abuse. (Second Edition 2001)
E. Single, L. Robson, X. Xie, J. Rehm (1998) “The economic costs of alcohol, tobacco and illicit drugs in Canada, 1992”, Addiction 93:7 (1998): 983-998.
Endnotes
[1] Eric Single is Professor of Public Health Sciences, University of Toronto; Senior Associate, Canadian Centre on Substance Abuse; Honorary Professor, Curtin University, Perth, Australia; Associate, Centre for Addiction and Mental Health; President, Single and Associates, Research Consulting Ltd. (6 Mervyn Avenue, Etobicoke, Ontario M9B 1M6, Canada; email: e.single@utoronto.ca).
The contributions of our colleagues David Collins, Helen Lapsley and Rick Harwood to the ideas expressed in this paper are gratefully acknowledged.
[2] Under some counterfactual scenarios, a discount rate is unnecessary, but usually the interpretation of the results are more limited.
[3] A paradoxical implication of this procedure is that if more effort were put into the successful prevention of alcohol abuse, with a resulting reduction in the charges as a consequence, police costs allocated to the social costs of alcohol misuse would fall.
[4] Arguably the social costs arising from the justice system are the least developed in almost all studies, reflecting a dearth of data. Given the proportion of those, say, in prisons that are involved with some from of drug abuse, this may be the single greatest lacuna in the studies.