The costs of tobacco, alcohol and illicit drug abuse to Australian Society in 2004/05
4. The epidemiological background to economic studies
Since a high proportion of the social costs of drug abuse results from drug-attributable death or sickness, the evidence quantifying the causal links between drug abuse and its health consequences represents data fundamental to social cost estimation. In recent years Australian epidemiological researchers have provided a series of comprehensive studies quantifying these causal links (see Holman et al., 1991; English et al., 1995 ; Ridolfo and Stevenson, 2001). The latest in the series is a research project on the Australian Burden of Disease (ABOD) undertaken by the School of Population Health at the University of Queensland (see Begg et al., 2007)
The strength of the causal link between abuse of a particular drug and its consequences for a particular health problem is represented by the aetiological fraction. "An aetiological fraction —also known as an attributable proportion or attributable risk—is a form of indirect quantification of morbidity and mortality due to a specified risk factor. In this case the risk factor is the consumption of tobacco, alcohol or an illicit drug" (Ridolfo and Stevenson, 2001, p. 2). Consider Table 7 below.
Table 7, Selected aetiological fractions for tobacco
Male Aged 35–39
Female Aged 35–39
|Ischaemic heart disease|
This table indicates that 40.8 per cent of all stroke deaths of Australian males in the age group 35 to 39 are estimated to be causally associated with tobacco. Among Australian females in the same age group, the percentage (33.3 per cent) is slightly lower. Slightly lower proportions of ischaemic heart disease deaths are attributable to smoking (40.1 per cent for males aged 35–39 and 32.6 per cent for females in the same age group). For an explanation of how aetiological fractions are calculated, see Ridolfo and Stevenson (2001, chapter 2).
An aetiological fraction which is positive but less than one indicates that the particular medical condition has more than one cause. The above table, for example, indicates that smoking is not the only cause of strokes or ischaemic heart disease. Occasionally these fractions can be negative, indicating that the drug in question has a protective effect against the medical condition under study.
Calculation of the aetiological fraction requires two fundamental pieces of information—the relative risk (measuring the causal relationship between exposure to the risky drug and the condition being studied) and prevalence (measuring the proportion of the relevant population engaging in the risky activity). Our current epidemiological work is fundamentally based upon the earlier work by Ridolfo and Stevenson (2001) and Holman et al. (1991) as well as on the recent University of Queensland ABOD study (Begg et al., 2007). Appendix C explains the derivation of the epidemiological estimates used in the current study.
Top of Page
Table 8 lists all the conditions which Appendix C concludes are causally linked to the abuse of alcohol, tobacco or illicit drugs.
Table 8, Conditions attributable to drug abuse, classified by drug
|Oropharyngeal cancer||Oropharyngeal cancer||Opiate dependence|
|Oesophageal cancer||Oesophageal cancer||Opiate abuse|
|Liver cancer||Stomach cancer||Opiate poisoning|
|Laryngeal cancer||Pancreatic cancer||Accidental opiate poisoning|
|Female breast cancer||Laryngeal cancer||Cannabis dependence|
|Alcoholic psychosis||Lung cancer||Cannabis abuse|
|Alcohol dependence/abuse||Cervical cancer||Amphetamine dependence|
|Alcoholic liver cirrhosis||Endometrial cancer||Amphetamine abuse|
|Road injuries||Bladder cancer||Cocaine dependence|
|Alcoholic poly-neuropathy||Kidney cancer||Cocaine abuse|
|Hypertension||Ischaemic heart disease||Psychostimulant poisoning|
|Ischaemic heart disease||Chronic obstructive pulmonary disease||Accidental poison by psychostimulants|
|Alcoholic cardiomyopathy||Tobacco abuse||Hallucinogen dependence|
|Supraventricular cardiac dysrhythmias||Parkinson's disease||Hallucinogen abuse|
|Heart failure||Pulmonary circulation disease||Hallucinogen poisoning|
|Stroke—haemorrhagic/ ischaemic||Cardiac dysrhythmias||Other psychotropic drug poisoning|
|Oesophageal varices||Heart failure||Accidental poisoning by hallucinogens|
|Gastro-oesophageal haemorrhage||Stroke||Anabolic steroid poisoning|
|Alcoholic gastritis||Peripheral vascular disease||Hepatitis B|
|Unspecified liver cirrhosis||Lower respiratory tract infection||Hepatitis C|
|Pancreatitis—acute/chronic||Ulcerative colitis||Infective endocarditis|
|Alcoholic beverage & other EtOH poisoning||Antepartum haemorrhage||Drug psychoses|
|Fall injuries||Low birthweight||Maternal drug dependence|
|Fire injuries||SIDS||Newborn drug toxicity|
|Drowning||Fire injuries||Antepartum haemorrhage|
|Aspiration||Asthma (under 15 years)||Low birthweight|
|Occupational and machine injuries||Macular degeneration||Road injuries|
|Suicide and self-inflicted injury||Otitis media||Suicide|
|Child abuse & assault||Lung cancer (passive)||Schizophrenia|
|Ischaemic heart disease (passive)||Licit/unspecified/combined|
Source: Appendix C.
Top of PageIt is acknowledged that there are a number of conditions (such as depression and anxiety) for which aetiological fractions have not yet been developed but for which there is already evidence relating to drugs as a causal factor. However, without attributable fractions the associated costs cannot yet be quantified.
4.1 Causal interactions between drugsEnglish et al. (1995) acknowledged the possibility of double-counting of drug-caused deaths in a situation in which the estimated numbers of deaths attributable to each of the drugs (alcohol, tobacco and illicit drugs) were added together to give a total for all drug-caused deaths. After examining the epidemiological evidence they concluded that interaction occurred in relation to only three conditions—oropharyngeal cancer, laryngeal cancer and fire injuries. In all three cases the interaction was between alcohol and tobacco. On the assumption of double-counting of deaths from all three conditions, the number of deaths for 2004/05 would be overestimated by 2.18 per cent (366 cases) of total deaths from all drug-attributable conditions in that year.
The significance of this double-counting from the point of view of the present study is that, while estimates of the social costs of individual drugs are not affected by the problem, there is some slight overestimation of mortality-related and morbidity-related costs involved when individual drug costs are aggregated to yield total drug costs. Accordingly, to eliminate the possible effect of double counting we discount the aggregate estimates of these types of costs by 2.18 per cent, the extent of the estimated double counting.
4.2 Costs and benefits of alcohol consumptionInterpretation of estimates of the social costs of alcohol use and misuse is more complex than for tobacco or illicit drugs. For some medical conditions, alcohol consumption at appropriate levels can have a protective effect; that is, alcohol consumption can reduce the risk of illness or death. With minor exceptions in relation to tobacco, there is no evidence of any analogous health benefits from consumption of the other drugs.
Top of Page
Table 9 below presents a summary of the alcohol-attributable conditions for which the abstinence-based aetiological fractions are negative. Abstinence-based means that the aetiological fractions reflect the risks (and the benefits) of alcohol at all levels of consumption relative to a baseline of complete abstinence from alcohol. A negative aetiological fraction means that alcohol consumption has a protective effect against the relevant condition.
Table 9, Alcohol-attributable conditions for which the abstinence-based aetiological fractions are negative
Ischaemic heart disease
Ischaemic heart disease
Source: Appendix C.
Ridolfo and Stevenson (2001) also present aetiological fractions for hazardous and harmful alcohol consumption, as defined by the National Health and Medical Research Council (NHMRC), relative to low alcohol consumption. This reflects the approach adopted by English et al. (1995) in their earlier comprehensive study of drug-attributable mortality and morbidity in Australia. It represents the extra effect of alcohol consumption for the "unsafe" drinker compared with the responsible drinker. In all cases (except for minor protective effects against cholelithiasis) the aetiological fractions calculated on this basis are positive. That is, in all cases, they indicate the existence of harmful effects. Thus the English et al. approach eliminates the protective effects (the benefits) of alcohol consumption. English et al. justified their approach as follows:
- Given that the object of public health intervention in the alcohol education field is 'unsafe' drinking (i.e. hazardous and harmful drinking as defined by the NHMRC, 1992), it stands to reason that the exposure contrast of greatest interest in the underlying epidemiology should be that between the 'unsafe' drinker and the responsible drinker; not between the drinker and the non-drinker.
English et al. did, however, proceed to make the following point:
- Nevertheless, it is acknowledged that there are some specific policy initiatives for which an aetiological fraction based on zero consumption may still be necessary. For example, estimates of the overall social costs and benefits of alcohol consumption require that the baseline be no alcohol intake rather than a 'safe' level. (English et al., 1995, pp. 57–8.)
The present authors contend that economic estimation of the social costs of drug abuse should also take into account any social benefits—that is, negative social costs, of drug abuse. (As is indicated above, from a public drug policy perspective, private costs and benefits of drug consumption are almost always irrelevant.) Benefits resulting from protective effects of drug consumption arise to a significant extent only in the consumption of alcohol. It is for this reason that the present study, in common with our two previous studies of the social costs of drug abuse, calculates aetiological fractions on an abstinence basis.
Top of PageTo express this in another way, the assumed alcohol counterfactual situation of no previous alcohol abuse over an extended period implies both:
- no costs of alcohol, and
- no protective health benefits of alcohol consumption.
It does, incidentally, appear to be the case that even so-called 'responsible' levels of alcohol consumption can be dangerous in relation to certain medical conditions; for example, female breast cancer. In addition, even risky/high-risk drinking can prevent some deaths (while causing far more).
As a result of the existence of both harmful and protective effects of alcohol consumption, there exists the potential for substantial misinterpretation of alcohol cost data. This is particularly the case when considering policies designed to minimise alcohol-related harm.
Table 10 below analyses the impact of estimated alcohol-attributable deaths by "deaths caused" and by "deaths prevented". It also presents the same analysis for hospital bed days. This table illustrates the potential pitfalls of interpreting the aggregate death or bed day figures, without considering the components of those aggregate data.
Table 10, Alcohol-attributable deaths and hospital bed days, 2004/05
Hospital bed days
|Male and female|
|Total male and female|
Note: figures in brackets are negative
Top of Page
Consider the total male alcohol-attributable deaths of 1,206. This figure is derived by subtracting the number of deaths prevented, as a result of the protective effects of alcohol (1,376), from the number of deaths caused (2,582). It is estimated that in total in 2004/05 alcohol caused 3,494 deaths but prevented 2,437. When examining the potential benefits of policies designed to prevent alcohol misuse, the relevant number of deaths is 3,494, not the net figure of 1,057.
Similarly the number of potentially preventable hospital bed days is 1,031,660, not the net figure of 916,934. Of particular relevance to alcohol misuse policies is the fact that these deaths and hospital bed-days can be more readily linked to actual individuals, while the deaths and hospital bed-days prevented are theoretical. As always, interpretation of these data has to be undertaken with care.
A paper by Chikritzhs, Stockwell et al. (2002) makes the same point from a slightly different perspective. They examine the numbers of lives lost and lives saved in 1998 due to low risk and risky/high-risk drinking, compared with a baseline of complete abstinence. Their aggregate numbers of lives saved differ slightly from those presented in Collins and Lapsley (2002), largely as a result of the different time periods of the two analyses, but indicate a similar direction.
Table 11, Estimated numbers of lives lost and saved due to low risk and risky/highrisk drinking when compared to abstinence in Australia, 1998
Low risk drinking
Source: Chikritzhs, Stockwell et al. (2002), Table 1.
Note: figures in brackets represent numbers of lives saved.
They conclude that:
Top of Page
- It is recommended that, for future reports on alcohol-caused morbidity and mortality, there would be value in presenting a more detailed picture that identifies both the costs and the benefits of low risk drinking and risky/high risk drinking. In order to do this, an abstinence-based contrast must be adopted.
The present report, in its results presentation, provides a slightly different distinction— mortality caused and prevented, together with morbidity caused and prevented. However, the interpretational message remains the same.
Of the total lives saved as a result of alcohol consumption, 88 per cent occur at the age of 60 or above. 73 per cent of hospital bed days saved accrues to this age group.
4.3 Changes in alcohol-attributable fractionsTo a large extent the changes since the previous report in the attributable fractions for tobacco and illicit drugs result from changes in prevalence rates, rather than changes in relative risk. However, significant changes in the evidence of relative risk for alcohol have caused substantial changes in the number of deaths and hospital bed days, and therefore in hospital costs, estimated to be attributable to the consumption of alcohol. The changes in relative risk arise from several causes.
The current method sees the exclusion (compared with the previous work) of epilepsy and psoriasis. A number of other conditions (oropharyngeal cancer, falls, fire, occupational) are defined by a wider range of codes, resulting in increased numbers of attributable cases. Hypertension has a narrower range of defining codes due to the exclusion of renal-related hypertension.
The net number of alcohol-caused deaths/hospitalisations was greater in the current study than would have been previously estimated due to a combination of the above factors but, more importantly, due to revisions of the attributable fractions (AFs) applied to a number of key diseases. These revisions have arisen from work recently undertaken by the University of Queensland Australian Burden of Disease (ABOD) research project. The ABOD report (Begg et al., 2007, p. 84) makes the following comment:
- This study reports a substantially lower health benefit due to alcohol compared to the previous Australian burden study (AIHW: Mathers et al 1999, AIHW: Ridolfo and Stevenson 2001) with only an stimated 2,346 deaths being prevented in 2003 compared to 7,157 deaths in 1996. This is due to the previous study underestimating the number of people who abstain from alcohol or drink less than 0.25 drinks per day.
For some diseases, the ABOD report utilised a lower AF for alcohol-caused cancers, hypertension, IHD, stroke, aspiration, drowning and suicide. While this reduced the numbers of eaths/hospitalisation due to conditions such as aspiration, drowning and suicide, the impact of smaller AFs was much greater in reducing the number of deaths/hospitalisations for IHD and stroke prevented by the consumption of alcohol. In general, the lower AFs are due to the lower prevalence of hazardous and harmful drinking reported in the ABOD report than that used by Ridolfo and Stevenson. While this contributes to the AFs for stroke and 'fire, scolds and burns', other analysis by the ABOD group using different information also contributes to the reduction in the attribution of these conditions to alcohol.
Top of Page
The problems with the earlier alcohol attributable fractions identified by the ABOD report have implications for the demographic estimates used in the present study. These implications arise for two reasons:
- It has not proved possible to recalculate the earlier alcohol-attributable fractions.
- As explained in Appendix A, in the demographic calculations the English et al. (1995) attributable fractions are assumed to apply prior to 1988 (but without illicit drug deaths prior to 1962) and those of Ridolfo and Stevenson in 1998, with intermediate fractions for the period 1989–1997. Those of Codde (see Appendix C) are assumed to apply in 2006, with intermediate fractions from 1999 and 2005.
Because the Ricardo and Stevenson attributable fractions imply an overestimate of the number of alcohol-attributable deaths prevented, and thus an underestimate of net deaths, the demographic estimates used here will tend to understate the alcohol-attributable reduction in the 2004/05 population. It follows that the alcohol-attributable social costs presented below will also be underestimated.
It is unfortunate, from the perspective of this economic study, that the change in the epidemiological method of estimating the protective effects of alcohol has eliminated any comparability between the alcohol cost estimates presented in this study and those for 1998/99 presented in Collins and Lapsley (2002). The epidemiological data affect not only the health cost estimates but also, through their demographic implications, the estimated labour costs in both the workforce and the household. Thus, in order to make inter-temporal comparisons of alcohol costs, the 1998/99 estimates would have needed to be recalculated using the revised epidemiological data. For further discussion of this issue see section "9.2 Recalculation of the 1998/99 alcohol cost estimates" on page 80.
4.4 Costs and benefits of tobacco consumptionIn general, the costs of tobacco consumption, in terms of deaths and hospital bed days caused, are much higher than for alcohol. Tobacco consumption yields some benefits, although, unlike alcohol, these benefits are proportionately extremely small. Nevertheless, for consistency it is appropriate to provide details of these benefits.
Table 12 gives details of the tobacco-attributable conditions with negative aetiological fractions; that is, against which tobacco provides some protective effect.
Table 12, Tobacco-attributable conditions for which the aetiological fractions are negative
Source: Appendix C.
Top of Page
The breakdown of tobacco-attributable deaths and hospital bed days into the caused and saved categories is presented in Table 13.
Table 13, Tobacco-attributable deaths and hospital bed days, 2004/05
Hospital bed days
|Male and female|
|Total male and female|
The savings in lives and bed days are trivial compared with the costs imposed. Of the total lives saved as a result of tobacco consumption, 97 per cent occur at the age of 60 or above. 73 per cent of hospital bed days saved accrues to this age group.
Top of Page