The trouble with alcohol abuse: what are we trying to measure, diagnose, count and prevent?


Alcohol abuse was once defined facetiously as mixing good liquor with bad soda, implying that the real meaning of the term was the abuse of expensive alcohols, not the drinker. Despite this logical inconsistency, the term alcohol abuse has grown in popularity during the last 30 years, in part because it seems to fill a terminological gap between alcohol dependence and ‘moderate’ alcohol use, and in part because it was adopted into the formal nomenclature of the American Psychiatric Association's (APA) Diagnostic and Statistical Manual in 1980 [1].

In the current version of the APA diagnostic manual [2], alcohol abuse is defined as a maladaptive pattern of alcohol use leading to clinically significant impairment or distress, as manifested by one or more of the following symptoms (i) failure to fulfill major role obligations at work, school or home; (ii) use in situations in which it is physically hazardous (e.g. driving an automobile); (iii) recurrent alcohol-related legal problems; (iv) continued substance use despite having recurrent social or interpersonal problems. The problem with this definition is suggested in an article by Keyes and Hasin published in this issue of Addiction[3], which reports that the rate of alcohol abuse increases with socio-economic status. The relationship between psychiatric problems and social class has been debated in the literature for quite some time [e.g. 4–6]. The problem in the particular case of abuse is that, as the article further suggests, this finding may be explained by the tendency for higher income drinkers to engage in drink-driving behavior, perhaps because they have greater access to automobiles. This is not the first time that drinking and driving, an indicator of hazardous alcohol use, has been linked with the likelihood of receiving a diagnosis of alcohol abuse [e.g. 7, 8]. Nor is it the first time that the abuse diagnosis has been questioned because of its susceptibility to the influence of social and cultural factors that may have little to do with the presence or absence of a psychiatric disorder [9]. Altogether, these findings raise important issues not only about the validity of the abuse diagnosis, but also about the way in which DSM-IV prevalence data derived from epidemiological surveys have been used in policy analysis, treatment need estimation, and in other applications.

The Keyes and Hasin paper [3] is based on an analysis of data in the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC), an extensive 2001–2002 national survey of the US adult population. Results show that 3.9% of the US adult population met diagnostic criteria for past year DSM-IV alcohol abuse. About 84% of the sample met criteria for abuse based on the hazardous use criterion alone, suggesting that at least in this US sample, abuse is mainly defined by hazardous use rather than legal problems, failed role obligations, or interpersonal problems linked to drinking. Further, the study showed that 69.3% of those diagnosed with abuse met criteria based solely on the drinking/driving items (either driving after having too much to drink or driving while drinking, as operationalized in the measuring instrument).

To examine the extent to which the findings reported by Keyes and Hasin [3] are present in other data sets, we repeated the analyses described above in three other general population surveys, all of which use multicluster sampling strategies to select respondents for interviewing: the 2006 US National Household Survey on Drug Use and Health (NSDUH), the 2007 Hispanic Americans Baseline Alcohol Survey (HABLAS) and the 2006 Brazilian National Alcohol Survey (BNAS). First, the identification of abuse in all of these surveys follows DSM-IV, but the operationalization of abuse criteria is slightly different across surveys. The NESARC uses the AUDADIS [10] to identify abuse and dependence. The NSDUH uses its own questions. The HABLAS and the BNAS use the Composite International Diagnostic Interview (CIDI) [11]. Differences in operationalization across these surveys are important because they occur in the hazardous drinking criterion, which seems to be responsible for the association between abuse and income in Keyes and Hasin's analysis. Basically, the AUDADIS hazardous use measure includes both driving after having too much to drink and drinking while driving. The CIDI asks whether the respondent is ‘often under the influence of alcohol in situations where you could get hurt, for example . . . when driving’. The NSDUH does not make reference to drinking and driving in its operationalization, but asks whether the respondent: ‘used alcohol regularly and then did something that might have put you in physical danger’. Because the NSDUH does not fully correspond to the DSM hazardous use criteria, we did not expect to find as strong a relationship between the abuse diagnosis, hazardous drinking and income. The HABLAS was selected because it allows for replication of the Keyes and Hasin analyses in a US ethnic minority group, whereas the Brazilian respondents in the BNAS were not expected to have the same level of access to cars found among those interviewed in the NESARC or the NSDUH. Therefore, alcohol abuse in the HABLAS and the BNAS should not be as closely related to drinking and driving and income as it is in the NESARC and NSDUH. Also, these surveys use slightly different skip patterns to decide who should be asked alcohol abuse questions. In the NSDUH only those who reported drinking on more than 5 days in the past year, or reported any drinking in the past year but did not report their frequency of past year use were asked the abuse questions. In the HABLAS and BNAS only past 12 month current drinkers were asked the abuse questions.

The analyses were all done with respondents of 18 years of age and older, using a ‘current’ (i.e. past 12 months) diagnostic time frame. Keyes and Hasin [3] excluded those with a lifetime diagnosis of alcohol dependence, which was not done with the HABLAS, the BNAS and the NSDUH because these surveys do not collect lifetime data on dependence. This should increase the denominator for the rates and thus decrease slightly the prevalence estimate for alcohol abuse in these three surveys. In the NESARC analysis reported by Keyes and Hasin, addition of respondents with lifetime diagnosis to the denominator lowers the alcohol abuse rate by about 0.5%, to 3.5% All analyses were conducted on weighted data. First, prevalence rates for alcohol abuse varied from 2.4% in the HABLAS, to 2.8% in Brazil and 4.4% in the NSDUH. These differences are in the expected direction, with lower rates of abuse in samples with less access to cars. Among those with abuse, proportions meeting the hazardous use criterion, which includes drinking and driving, were: HABLAS 42%; Brazil 38%; and NSDUH 78%. These rates are lower than the 90% found in NESARC data, but are in the expected direction, with lower rates in the HABLAS and Brazil. Examining the proportion of individuals who are identified as alcohol abusers based only on a positive response to the hazardous use criterion further confirms the expectations stated above: HABLAS 36%; Brazil 27%; NESARC 69%. Given that the NSDUH does not include driving under the influence in the hazardous use criterion, the proportion was not estimated. Finally, cross-tabulation of alcohol abuse and the hazardous use criterion with income data in the HABLAS, BNAS and NSDUH did not show a statistically significant association in any of the surveys.

It does seem therefore that using drinking and driving as an indicator of hazardous drinking in surveys with populations with a high rate of access to cars makes the diagnosis of alcohol abuse sensitive to the surrounding environment. In this case not only does the diagnosis become sensitive to socioeconomic status, it may also be influenced by factors such as the extent to which drinking and driving is seen as a dangerous, deviant behavior, or the extent to which drinking and driving laws are enforced by local police. This may violate the assumptions and intent of the original diagnostic construct. For instance, when discussing shortcomings in DSM-III criteria, Rounsaville et al.[9] underlined the limitations arising from relying on the presence of ‘social and occupational consequences’ for a positive diagnosis of abuse or dependence. They correctly argued then that the ‘wide variability in acceptable uses of psychoactive substances’ would make the diagnosis particularly sensitive to social forces in the environment. They go on to state that ‘the current system (DSM-III) is vulnerable to powerful, swiftly changing social forces such as the tightening of laws restricting alcohol use while driving’. This statement, written over 20 years ago, represents well a limitation in the abuse diagnosis if it is largely based on issues related to drinking and driving. The fact that problem rates for the other three abuse criteria are very low indicates that for a significant proportion of the population who meet the diagnostic threshold for abuse (one of four symptoms), the defining diagnostic indicator is drinking in situations that may increase risk of harm. This does not seem to fit the original intent of the abuse criteria.

Indeed, the World Health Organization considered and then rejected the incorporation of a hazardous alcohol use category in its diagnostic system because it did not fit the definition of a psychiatric disorder. For the same reason, one may question the appropriateness of the diagnosis of abuse in US general population surveys if it is driven mainly by the hazardous use criterion. Is it correct to render a psychiatric diagnosis solely for drinking in a way that may increase the risk of an accident or injury? Would it not be better to understand this as an indicator of a personality disorder? More importantly, is it appropriate to report the abuse diagnosis prevalence rates as part of treatment need estimates where abuse is often combined with dependence as an indicator of the need for resources to treat ‘alcohol use disorders’. No wonder so many people who are in this category report that they are not in need of treatment. Rather than being in a ‘state of denial’, they may not really need any more treatment than a brief intervention that points out the potential consequences of drinking and driving and the other risks of heavy drinking. Finally, when the abuse diagnosis is combined with dependence in providing national estimates of ‘alcohol use disorders’, such a practice would tend to dramatically inflate the dimensions of the alcohol problem. This may lead to faulty decision making and misinformed resource allocation.

What should be done about this situation? Here are some suggestions for epidemiologists, nosologists and alcohol researchers:

  • • Conduct further research on the abuse diagnosis as measured in psychiatric interviews, especially in cases of epidemiological studies in the general population. Interviews with general population samples are relatively brief, which does not allow for the development of extensive probing of symptoms and may thus lead to false positives.
  • • Do not report rates of DSM-IV alcohol abuse in prevalence estimates without also reporting item-level rates so that the data will not be misinterpreted. Readers will then be able to estimate the extent to which the identification of alcohol abuse is based on hazardous drinking alone, and especially on drinking and driving, or the occurrence of other types of problems.
  • • Avoid using the abuse diagnosis as an independent or dependent variable in research unless preliminary analyses are conducted at the item level.
  • • Do not combine abuse and dependence rates into the more general category of alcohol use disorders without explaining what proportion of the total prevalence is attributable to hazardous use, which technically may not fit the definition of a psychiatric disorder.
  • • Change the threshold used for positive identification of alcohol abuse from one criterion to at least two. This will ensure that abusers report at least one more problem besides drinking and driving. Another option is to eliminate alcohol abuse entirely from the classification system, focusing instead on repeated acts of acute intoxication in the absence of major dependence symptoms. Alcohol-related problems now included as indicators of abuse can be reported in epidemiological surveys as such, together with other problems such as those that are job related (e.g. absenteeism), family related (e.g. divorce, separation) and others.

As implied in the title of this editorial, the trouble with the term ‘alcohol abuse’ is that the attempt to define it in terms of a disparate set of behavioral symptoms has perhaps resulted in what philosophers call the fallacy of misplaced concreteness. This fallacy occurs when one mistakes an abstract concept about the way things are for a physical or ‘concrete’ reality. Using a different type of terminology, the term and the concept appear to be in need of ‘deconstruction’. When deconstructed as an operational definition, the hazardous use symptom seems to account for an inordinate amount of the prevalence of alcohol abuse. When deconstructed as a diagnostic category, it does not seem appropriate to label individuals with an alcohol abuse disorder when their only symptom is drinking in a way that may increase the potential for harm.


Raul Caetano's work on this paper was supported by a grant (RO1-AA013642) from the National Institute on Alcohol Abuse and Alcoholism (NIAAA) to the University of Texas School of Public Health. Thomas Babor was supported by NIAAA grant # 2P60AA03510 to the University of Connecticut School of Medicine.