When it was first proposed in 1976, the Alcohol Dependence Syndrome (ADS) concept was promoted as a provisional construct that would require significant work to confirm its assumptions. As described in the paper based on the David Archibald Lecture [1], there has been a considerable amount of clinical, neurobiological and epidemiological research conducted in the past 30 years that is relevant to a scientific evaluation of the ADS concept, with much of it supporting its validity. Nevertheless, some research findings do not agree with the concept, particularly in relation to studies using methods of modern psychiatric epidemiology. In this commentary I would like to suggest that the problem may be less in the ADS construct than in the limitations of our measurement procedures.

One issue has been the general failure of psychiatric epidemiologists to recognize the underlying theory behind the ADS concept, as described in a later paper by Edwards et al. [2]. That paper clearly established ADS within the broad domain of psychological learning theory, especially the operant, classical, cognitive–behavioral and social learning traditions. Instead of characterizing the ADS elements in terms of pharmacological symptoms, ‘maladaptive responses to drinking’ and the severity of consequences, as do Li et al. [1], learning theory would suggest that the symptoms are better organized into physiological, behavioral and cognitive elements that often interact with one another to maintain drinking once dependence has become established. This explains the emphasis on the persistence of drinking despite consequences (increased salience), rather than on the consequences themselves. The salience of alcohol reinforcement means that dependent people will drink to excess despite the ‘punishment’ that often results from their drinking. Although alcohol-related problems are often a key indicator of alcohol dependence, they should not be part of the core syndrome because problem indicators distract attention from the underlying dynamics of addiction: i.e. neuroadaptation (tolerance and withdrawal), maladaptive cognitions (not only illogical thinking but also subjective craving for alcohol) and impaired control over drinking.

Common sense suggests that neuroadaptation, maladaptive thought processes and inability to control one's drinking are conceptually and functionally different from having an alcohol-related injury and other problems. Of course, problems are likely to be correlated with dependence symptoms, but that does not mean that alcohol-related injuries or liver cirrhosis tell us much about why some people drink themselves to death. The application of high-powered statistical procedures to symptom measures may reveal that dependence symptoms cluster or problem indicators correlate, but they will never be able to ‘prove’ that problems are part of a dependence syndrome in a functional or etiological sense.

The role of consequences in ADS theory is thus tangential, which explains why they were relegated to the residual category of harmful drinking in ICD-10. The definition of harmful drinking was limited further to harms that could be attributed clearly to alcohol's toxicity or impairment in an implied dose–response manner, rather than allowing social interpretations to enter into the definition. Thus, alcohol-related pancreatitis and injuries qualify as harmful drinking in ICD, whereas marital problems do not, because they are likely to vary from one culture to another, or even from one marriage to another. The lack of a clear rationale for the alcohol abuse criteria in DSM-IV resulted in the inclusion of a mix of dependence and harm, dictated apparently by blind empiricism (looking at correlations among items rated by heavy drinkers) rather than from any conceptual or theoretical rationale. The National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) data suggest that measures of three of the four ‘abuse’ criteria correlate more with the dependence syndrome criteria, whereas the fourth is somewhat independent. This is not surprising, because the abuse criteria are theoretically closer to dependence than to harmful consequences (e.g. drinking despite problems is a measure of salience, not consequences). Although abuse and dependence criteria can be scaled along a single continuum of severity, that does not mean that we should be mixing apples and oranges in our enumeration of diagnostic criteria. Ideally, the dependence criteria should focus upon the essential features of the disorder, not the epiphenomena, especially those (such as social problems) that are likely to vary with social conditions.

Another measurement issue with conceptual implications can be stated as follows: to what extent can we make inferences about a complex set of behavioral, cognitive and physiological symptoms when they are all measured by small numbers of interview questions that rely not only upon respondents' honesty, but also upon their ability to retrieve very complex types of information with little or no guidance as to what is being measured? Moreover, the items included in most of the structured interviews were designed and validated to measure the presence or absence of the diagnostic entity (i.e. dependence, abuse, harmful use), not necessarily the individual symptoms. To now test a theory on the basis of these questions creates the risk of far exceeding our measurement capacity.

One of the most puzzling findings emerging from recent psychiatric epidemiology research related to the ADS concept is the extraordinarily high prevalence rate of alcohol dependence among young adults. Harford et al. ([3], Table 1), for example, report that 4.6% of US male adolescents (aged 12–17) meet criteria for past-year dependence, and the rate increases to 8.5% in the 18–23-year age group. Thereafter, prevalence declines for each succeeding age cohort. Taken at face value, the population prevalence data are at variance with the general assumption that alcohol dependence is cumulative and progressive, and that early onset of alcohol dependence (as noted in the clinical typology literature) is associated with a more serious form of alcohol dependence characterized by a severe course. Why would alcohol dependence be so prevalent in the younger age cohorts, and does this not represent an epidemic of alcoholism that will continue to grow as the younger cohorts become older? A recent paper [4] suggests that, far from an epidemic of youthful alcoholism, the answer may be measurement error, with the most likely culprits being our self-report measures of tolerance, withdrawal and salience.

If psychiatric epidemiology has failed to provide an adequate test of the ADS theory, where do we go from here? As noted by Caetano & Babor [4], there is a need to explore symptom level data more thoroughly to determine whether young adults are confusing the sequelae of acute intoxication with alcohol withdrawal, and rapid initial tolerance with the alcoholic's ability to consume significant amounts of alcohol without apparent behavioral impairment. Secondly, epidemiologists and clinicians need to be more skeptical of the tendency of structured psychiatric interviews to classify young adults as being alcohol-dependent, especially in the absence of chronic heavy drinking. Thirdly, it may be useful to revisit Jellinek's [5] (1960) notion of alpha alcoholism, which represented a binge drinking pattern that may be harmful in the absence of pronounced tolerance, withdrawal and loss of control, and which shows no signs of progression. This would depend upon our ability to develop better measures of those dependence elements that seem to be endorsed with high frequency by younger binge drinkers. Marshall MacLuhan was fond of saying that ‘we shape our tools, and thereafter our tools shape us [6]’. It would be unfortunate if our concept of dependence was shaped more by the limitations of our measurement tools than by the clinical and epidemiological realities we are trying to understand.