Dr. DeLuca's Addiction Website

The Role of Craving in Alcohol Use, Dependence, and Treatment

Barbara A. Flannery; Amanda J. Roberts; Ned Cooney; Robert M. Swift; Raymond F. Anton; Damaris J. Rohsenow


Craving is a poorly understood term used to describe a variety of phenomena related to subjective and objective states that may lead to alcohol consumption. Craving has been described and measured subjectively in terms of psychological phenomena such as urge, desire, and obsessive thoughts, and objectively in terms of physiological arousal, salivation in response to alcohol cues, decreased latency to drink, or increased consumption. The DSM IV (American Psychiatric Association, 1994) does not include craving as a criterion symptom of alcohol or other substance dependence but does state that craving (“a strong subjective drive to use the substance”) is probably a universal accompanying symptom. Craving as a conscious desire to use alcohol may be more readily perceived by the abusing as opposed to the dependent individual, because once a person moves from abuse to dependence, tolerance and/or withdrawal symptoms may be the factors that are viewed as precipitating more, continued, or relapse drinking.

Current theories of craving usually fall within one or more of the following categories: anticipation of euphoric effects (positive reinforcement, e.g., Rohsenow et al., 1990), removal of withdrawal symptoms (negative reinforcement, e.g., Glautier and Drummond, 1994), and as a function of conditioning to stimuli associated with alcohol or other abused substances (e.g., environment, people, affective state, see Childress et al., 1993 for review). The latter category, conditioned or associatively learned craving, may be experienced as the anticipation of euphoric effects or the elimination of withdrawal symptoms but not necessarily on a conscious level. Conditioned craving is thought to be a function of neural adaptations that result from chronic use of alcohol or other abused substances (e.g., Kalivas et al., 1998). These neurochemical changes are distinct from the acute and chronic neuropharmacological effects of a substance and may help explain why abstinent individuals relapse. To illustrate, if an abstinent, alcohol-dependent person goes to the bar where she or he used to drink regularly, the setting alone could trigger a neurochemical response to alcohol that, in turn, produces withdrawal symptoms such as anxiety, tremors, or sweating. The individual, knowing from experience that alcohol will relieve these symptoms, then may report a desire to drink. But, because conditioning occurs below the level of conscious awareness, the person may not ascribe the desire for alcohol or his or her relapse to craving.

Given the complex nature of craving, researchers have yet to reach a consensus definition of this phenomenon. However, a variety of approaches are being used to better understand the relationship of craving to drinking and relapse. The purpose of this symposium was to present the diverse techniques currently being used to explore craving. A variety of theoretical perspectives and empirical studies were described. Topics included preclinical models (Amanda Roberts), field assessment (Ned Cooney), effects of medications on craving (Robert Swift), clinical insights on craving (Raymond Anton), predictive utility of craving questionnaires (Barbara Flannery), and predicting posttreatment drinking by using in-treatment urge assessment (Damaris Rohsenow).


Amanda Roberts

Animal models are important tools for investigating neurobiological mechanisms that underlie craving and relapse and for testing putative anticraving and antirelapse medications. In laboratory animals, craving is defined as increased motivation for alcohol as reflected by excessive intake. Relapse, therefore, is excessive intake that follows an alcohol-free period. Four potential models of craving and relapse are currently under investigation at the Scripps Research Institute Alcohol Center: the alcohol deprivation effect, cue-precipitated reinstatement of alcohol self-administration, alcohol self-administration during withdrawal, and alcohol self-administration after periods of protracted abstinence.

The alcohol deprivation effect is characterized by a transient (1 or 2 day) increase in alcohol intake after a period of deprivation and is reliable only when animals have had extensive (at least 8 weeks) experience with alcohol. The model does not require that the animals are physically dependent on alcohol; therefore, this effect does not appear to be a result of physical withdrawal. It has been proposed that the alcohol deprivation effect may reflect changes in the reinforcing effects of alcohol (Heyser et al., 1997). In one study, chronic administration of the antirelapse compound, acamprosate, throughout the deprivation period completely eradicated the alcohol deprivation effect (Heyser et al., 1998). This paradigm is being investigated further in terms of neuropharmacological mechanisms that underlie craving.

In the cue-precipitated reinstatement of alcohol self-administration model (Katner and Weiss, 1999; Katner et al., 1999), rats are trained to lever press for alcohol in the presence of particular environmental stimuli. Lever-pressing behavior then is extinguished by removing both the alcohol and its associated stimuli. Subsequently, the rats are exposed to the alcohol-associated stimuli and their lever-pressing behavior is recorded. A significant increase in responding relative to extinction levels is reliably observed. This reinstatement occurs primarily in the presence of the smell of alcohol, which presumably triggers alcohol-seeking behavior. Naltrexone has been shown to block this effect. This model, like the alcohol deprivation effect model, is useful for exploring the neuropharmacological mechanisms involved in context-associated craving and relapse.

In the two self-administration models, rats are trained to lever press for alcohol and then are made physically dependent in alcohol vapor chambers. The rats are allowed to lever press for alcohol either during the early phase of withdrawal or after varying periods of protracted abstinence. It has been shown that rats will self-administer enough ethanol during early withdrawal to maintain significant blood alcohol levels and to ameliorate the physical signs of alcohol withdrawal (Roberts et al., 1996). This enhanced alcohol self-administration by dependent rats persists beyond the presence of overt withdrawal symptoms (Roberts et al., 2000). The enhanced self-administration observed both during withdrawal and after protracted abstinence can be blocked by a corticotropin-releasing factor receptor antagonist (Chan et al., 1999), which suggests that the stress system is critical in relapse.

The paradigms briefly discussed here model important aspects of alcoholism in humans that clinical researchers also are studying (i.e., anticipation, alcohol use, alcohol withdrawal, and protracted abstinence). These animal models also will be used to explore other factors shown to be important in craving and relapse in human alcoholics such as stress and affect (discussed by Dr. Ned Cooney) and obsessive thoughts, impulsive and compulsive behaviors, and changes in brain chemistry (discussed by Dr. Raymond Anton). In addition, these models show promise for providing insights into the development of additional pharmacotherapies for alcoholism.


Ned Cooney

Field assessment methodologies provide an alternative to the standard pencil-and-paper assessment technique. Field assessment encompasses a variety of different techniques, such as (1) maintenance of a written daily retrospective craving record, (2) reporting craving via an interactive voice response system, (3) experience sampling by using a pager or beeper watch to prompt written recordings, and (4) use of an electronic diary with real time recordings entered on a hand-held computer. Field assessment provides several advantages over retrospective or laboratory craving measurement. For example, this technique eliminates recall errors and demand characteristics that are inherent in the typical research methodology. Field assessment also offers benefits that overcome the artificiality of laboratory cue-induction studies where the perceived unavailability of alcohol may negatively impact a subject’s craving.

As with any type of craving measurement, field assessment has its problems. Study participants can fake compliance or find that reporting craving outside the research setting is a burden. Individuals also may be unwilling to respond after having relapsed or may fail to initiate craving reports when they actually occur (Litt et al., 1998). In an attempt to overcome some of these analytical dilemmas, (Stone and Shiffman 1994) initiated a variation of field assessment, termed ecological momentary assessment (EMA), that has been used in studies of smokers and drinkers. This technique involves the use of a hand-held computer that the study participants carry with them so that they can record craving-related behaviors, cognitions, and affects at the moment they occur and in the contexts within which they occur. Several sampling methods have been used to collect data through EMA such as interval, signal, and event contingent reporting.

In an 8 week trial in which 37 heavy drinkers were randomized to a program of moderation control or were put on a waiting list, (Collins et al. 1998) gave participants electronic diaries. They used EMA to track craving through random prompts (signal contingent recording), at the start of each day (interval contingent recording), and at the beginning and at the end of drinking episodes (event contingent recording). Compliance with the random prompts was excellent; subjects missed only 7.4% of the prompts and failed to complete 8.2% of the prompted recordings. Collins and colleagues found that most drinking episodes occurred during leisure or social activities and most frequently with a spouse, partner, or family member. Contrary to conventional thinking, heavy drinking (more than five drinks per occasion) often was preceded by a positive mood and, at least in this study, negative mood was not a significant predictor of excessive drinking.

An EMA study of 26 alcohol-dependent veterans (Litt et al., 2000) demonstrated that negative affect in conjunction with high arousal was the best predictor of urge to use alcohol (F = 33.78, df = 1, p > 0.0001) followed by a positive affect with high arousal (F = 9.96, df = 1, p > 0.001). There was a weak correlation between the subjects’ reported desire to drink during prior laboratory cue reactivity testing and the urge to drink and actual drinking during subsequent field assessment (r = 0.30, p < 0.10). A between-subjects analysis of field-assessed craving revealed that individuals who reported frequent urges to drink had greater alcohol dependence and higher trait anger and anxiety scores than did the nonreporters.

Only a small number of studies have used EMA methodology. Some of the limitations of EMA include the burden of data collection, interruption of recording after relapse, and uncertain compliance for event-contingent recording. On the other hand, there are several advantages of EMA methodology over traditional methods, which include reduced bias in participant recall and retrieval, inability to fake compliance, sensitivity to momentary changes (e.g., before and after drinking, before and after smoking), and sensitivity to momentary interactions among substances. This rich source of data has the potential to advance our understanding of craving and addiction.


Robert Swift

Craving (usually defined as the conscious desire or urge to use alcohol) is an important heuristic construct with both scientific and treatment implications. Because high levels of craving have been associated with increased probability of relapse, especially in the early posttreatment period, interventions that reduce craving may have therapeutic benefits (Anton et al., 1996). Behavioral treatments that reduce craving, such as cue exposure combined with coping skills, have been shown to reduce subsequent alcohol use (Monti et al., 1993c).

In the past decade there has been increasing interest in the use of pharmacotherapy to improve the effectiveness of alcoholism treatment (Litten et al., 1996; Swift, 1999). The rationale for the use of medications is based on the premise that alcohol use is mediated through specific neurobiological and behavioral mechanisms that initiate and maintain alcohol drinking. Several medications have been demonstrated to reduce alcohol consumption, and some of these also have been shown to reduce alcohol craving.

The first medications used to treat alcohol dependence were aversive agents such as disulfiram and calcium carbimide. Because disulfiram inhibits dopamine-B-hydroxylase, a brain enzyme involved in dopamine and norepinephrine metabolism, it has been suggested that disulfiram could impact craving. However, neither of these drugs has been efficacious in treating alcohol dependence, nor have studies explicitly explored the effects of these medications on craving (Hughes and Cook, 1997).

Data from clinical trials support the efficacy of the opioid antagonists naltrexone and nalmefene for reducing spontaneous craving in recently abstinent alcohol-dependent individuals (Anton et al., 1999; Mason et al., 1994; O’Malley et al., 1992; Volpicelli et al., 1992). When latency to drink is used as an operational measure of craving, naltrexone also has been shown to reduce urge and/or motivation to drink (Davidson et al., 1996, 1999). On the other hand, laboratory cue reactivity studies show inconsistent effects of naltrexone on craving (Farren et al., 1999; Modesto-Lowe et al., 1997; Monti et al., 1999; Rohsenow, 1998).

A relatively new agent, acamprosate (calcium aceylhomotaurinate), has been posited to exert its actions at the polyamine regulatory site on the NMDA glutamatergic receptor complex and also to help restore balance between excitatory and inhibitory neurotransmitters within the nucleus accumbens (Litten et al., 1996). Although European trials have demonstrated that acamprosate increases the rate and duration of alcohol abstinence, its effects on craving have been equivocal. For example, in a French trial that used 1.2 and 2 g acamprosate doses, (Paille et al. 1995) found a decrease in craving among subjects who received the higher dose at 3 months but no effect at 6 or 12 months. Similarly, in a 48 week study, (Sass et al. 1996) found increased abstinence rates in the acamprosate-treated group compared with placebo. Self-reported craving as measured by visual analogue scale was variable over the course of the treatment period and was not correlated with drinking. In the U.S., acamprosate is still in the investigational stage of development.

The effects of various dopaminergic agents on alcohol craving have been assessed because it has been long known that dopamine plays a critical role in the reinforcing aspect of abused substances, which include alcohol (Koob and Weiss, 1992). For example, (Modell et al. 1993) conducted a placebo-controlled laboratory study by using the dopamine antagonist haloperidol. Although haloperidol was effective in reducing craving for and consumption of alcohol in a group of 16 alcohol-abusing or alcohol-dependent individuals, haloperidol’s general blunting of affect and behavior prohibits its use as a pharmacotherapy for alcohol dependence. Olanzepine, another dopamine antagonist, shows more promise. In a preliminary placebo-controlled, double-blind laboratory study, (Hutchison et al. 1998) reported that pretreatment with olanzepine reduced alcohol and cue-induced craving as measured by visual analogue scale. Tiapride is another dopaminergic antagonist that has been used in Europe to treat alcohol dependence. Although a 100 subject, placebo-controlled trial conducted in England showed tiapride to be more effective than placebo for relapse prevention, its impact on craving was not measured (Shaw et al., 1994).

Different serotonin (5-HT) receptor subtypes have been implicated in various aspects of alcohol use and dependence. Consumption, intoxication, and the development of tolerance may be mediated partially through action at 5-HT1 receptors. Activity at 5-HT2 receptors may contribute to withdrawal symptoms, and 5-HT2 and 5-HT3 (through modulation of dopamine release) receptor activity may play a part in the rewarding effects of alcohol. Although animal studies with serotonergic agents have demonstrated that these compounds are effective in reducing alcohol consumption, human clinical trials have had only modest success. For example, (Naranjo et al. 1994) reported that fluoxetine (a selective serotonin reuptake inhibitor) only slightly reduced the urge to drink. There have been equivocal findings for drinking and craving with the 5-HT1A agonist buspirone (Malec et al., 1996). The 5-HT3 antagonist ondansetron was shown to attenuate the desire to drink in a preliminary study of healthy adult males (Johnson et al., 1993). In more recent studies of ondansetron (Johnson et al., 2000b) and ondansetron and naltrexone (Johnson et al., 2000a), the effects of these medications on craving were not measured. The 5-HT2 antagonist ritanserin has thus far been disappointing in clinical trials. In a multicenter trial of 423 alcohol-dependent individuals who received either 5 mg/day of ritanserin or placebo, there was no different in craving or drinking between the active medication and the placebo groups (Johnson et al., 1996).

Several nonserotonergic, anxiolytic agents have been reported to attenuate craving. For example, there have been reports that the -adrenergic blockers propranolol and atenolol reduce alcohol craving during withdrawal and early abstinence (Gottlieb et al., 1994). The alpha-2-antagonist clonidine has been reported to reduce alcohol as well as opioid and nicotine craving (Liskow and Goodwin, 1987). Although sedatives such as -hydroxybutyrate are reported to reduce alcohol consumption and to decrease craving when administered to abstinent alcoholic patients, such medications can cause sedative dependence and should be administered to alcoholics only with extreme caution and close monitoring (Addolorato et al., 1996).

In summary, several medications that reduce alcohol consumption in laboratory studies and clinical trials also reduce craving for alcohol. However, craving is not always associated with alcohol drinking, and there is not always a direct relationship between medication-induced reductions in craving and reductions in drinking. A better understanding of the concept of craving and its relationship to alcohol drinking may lead to a better understanding of the neurobiology of alcohol dependence and more effective clinical interventions.


Raymond Anton

At the Medical University of South Carolina (MUSC), Anton and colleagues have evaluated a neuroanatomical model and several methods for conceptualizing and quantifying alcohol’s rewarding effects and alcohol craving. Here, recent work is presented beginning with a neural model of craving and findings from a preliminary functional magnetic resonance imaging (fMRI) study of cue-induced craving. Some data from the Obsessive Compulsive Drinking Scale (OCDS), a self-report assessment developed at MUSC, also are presented. Finally, preliminary results from a laboratory cue induction and self-administration procedure for assessing medication effects are described.

Craving can be viewed as “a state of mind” that, in some instances, leads a person to ingest alcohol in excessive quantities, often with full knowledge of the adverse consequences of doing so. For many, craving is uncontrollable and likely triggered by internal and external stimuli. An alcoholic’s repetitive urge to drink and apparent resistance to this urge is quite similar to obsessive compulsive disorder (OCD). The similarities between alcohol dependence and OCD (i.e., obsessional thinking and compulsive uncontrollable behavior) suggest that they share a common or similar neural substrate. This complex neural system is hypothesized to increase the salience of alcohol and alcohol-related cues for alcohol-dependent individuals and may operate in the following manner. Alcohol and stimuli associated with alcohol elevate levels of dopamine within the nucleus accumbens (ventral striatum in humans). If craving is viewed as a multidimensional phenomenon that includes anticipation of reinforcement, has an emotional component, and involves activation of motivational systems that may lead to behavior, then it is possible that alcohol craving involves a neural network of various brain structures that project to and from the nucleus accumbens. Nucleus accumbal projections to the amygdala may provide a stress modulation and emotional tone to alcohol reward and craving. It is possible that the positive emotional impact of alcohol (or the removal of negative emotional states via alcohol effects, e.g., stress-relief drinking) and alcohol cues are “amplified” via amygdalar activation in alcohol-dependent individuals. Additional projections from the amygdala to the basal ganglia could activate stored repetitive action scenarios and generate stereotypical behaviors (e.g., alcohol-seeking behavior). Nucleus accumbal projections to prefrontal cortical areas may recruit functions such as decision-making, associative memory, and impulse control in the alcohol reward and craving. If frontal lobe dysfunction (such as attention-deficit disorder, depression, or toxic alcohol effects) exists, impulse control may be inhibited and lead to craving-initiated drinking. Alternatively, “highjacking” of the frontal-basal ganglia-thalamic neural circuits by alcohol could lead to obsessive thoughts of alcohol (craving) and compulsive (automatic) alcohol use.

At MUSC, an fMRI paradigm for assessing cue-induced craving is being developed. In the initial study, 10 non-treatment-seeking alcoholics and 10 age and sex-matched social drinking controls were presented with pictures of alcoholic or neutral beverages (with appropriate controls for visual effects). After the data were collapsed across groups, analysis revealed that alcoholics have significantly more left dorsolateral prefrontal cortical and thalamic neuronal activity than do control subjects in response to alcoholic beverage pictures. This initial work is consistent with findings from cocaine imaging studies and suggests that some of the brain structures of the model outlined previously are involved in cue-induced craving. Additional findings suggest that alcohol cues can activate specific structures implicated in craving without an individual consciously being aware of or labeling the activation as craving.

Modell and colleagues, noting the similarities between OCD and alcohol dependence, adapted the Yale Brown Obsessive Compulsive Interview Scale to quantify the obsessional thinking and compulsive behavioral aspects of alcohol dependence (Modell et al., 1992a,b). Anton and colleagues subsequently reformulated and validated the scale into an easily administered self-report measure known as the OCDS (Anton et al., 1995). The OCDS contains an obsessive subscale (items 1–6) that measures intensity, duration, and frequency of thoughts about drinking and the ability to resist these thought, and a compulsive subscale (items 7–14) that queries about drinking behaviors and drinking-related behaviors. Both subscales also address the interference with normal functioning that is caused by thoughts about drinking and drinking-related behaviors. Initial analyses revealed that the full-scale score varied with the severity of alcohol dependence and that the OCDS can discriminate social drinkers from individuals who are early stage alcoholics (Anton and Drobes, 1998; Moak et al., 1998). In a recent study, (Malcolm et al. 2000) found that individuals who had had several medical detoxifications had higher OCDS scores than those who underwent their first detoxification independent of recent drinking quantity and overall alcohol severity.

(Roberts et al. 2000) recently showed that the OCDS could be factor analyzed into a measure that contains three subscales. One of these subscales measures resistance/control impairment (items 5–8, 12, and 14, which measure ability to resist thoughts about drinking, and frequency and quantity of drinking in a specified period), and this subscale appeared to predict relapse drinking in the week after the assessment period. But note that two of the items ask directly about the frequency and quantity of drinking. Perhaps of equal or greater importance, this factor was sensitive to naltrexone treatment effects (at least when combined with cognitive behavioral therapy). Naltrexone-treated individuals had lower scores on this subscale than did placebo-treated subjects during a 12 week treatment trial (Anton et al., 1999; Roberts et al., 1999). Thus, naltrexone appeared to reduce urges and increase the ability to resist or control those urges.

One of the most recent investigations of craving at MUSC involves the development of a new clinical laboratory paradigm. This procedure ultimately will be used to evaluate the effects of medications on the subjective effects of alcohol (i.e., stimulation and sedation as measured by the Biphasic Alcohol Effects Scale) and alcohol craving. In the initial study that used this paradigm, responses of non-treatment-seeking alcoholics (considered early-stage alcohol dependent but generally younger and less severely alcohol dependent than treatment-seeking individuals) who drank 40 to 50 drinks per week were compared with the responses of age-, sex-, and ethic group-matched social drinkers (who drank less than 14 drinks per week). Participants were first exposed to alcohol cues (holding and sniffing an alcoholic beverage) and then consumed alcohol (achieving about 50–60 mg% peak blood alcohol concentration [BAC]) in a bar-like setting. After 1 hr, subjects were permitted up to four “mini-drinks” of alcohol (maximum BAC not to exceed 100 mg%). The non-treatment-seeking (early-stage alcohol-dependent) group experienced higher baseline (nonstimulated) craving as measured by a five-point rating scale (similar to the Alcohol Urge Questionnaire [AUQ], Bohn et al., 1995) than did the social drinking group. For the early-stage alcoholic compared with the social drinking group, craving, as assessed by the five-item rating scale, was greater after exposure to alcohol cues and after alcohol consumption. However, due to high baseline craving for the early-stage alcoholic group, this effect was not as robust as one would expect. Although the baseline and postdrinking craving scores were significantly higher for the early-stage alcoholics than for the social drinkers, the actual consumption of alcohol did not increase appreciably craving. It is interesting that, although the early-stage alcoholic group’s craving increased marginally after consumption, craving scores declined in the social drinking group. These results suggest that, at the BACs achieved, alcohol itself does not eliminate the high level of craving expressed by this group before alcohol consumption. Another finding of interest was that the early-stage alcoholics had much higher stimulation from alcohol than did social drinkers as measured by the Biphasic Alcohol Effects Scale. Sedation scores did not differ between the two groups. Although expectancy effects cannot be ruled out, there is some suggestion, contrary to belief, that alcoholics (early stage) are not tolerant to the sedating effects of alcohol (perhaps not measurable at the dose used and BAC achieved) but, rather, appear to be sensitized to the stimulating effects of alcohol. Preliminary analysis also suggests that the stimulating effects are independent of the experience of craving.


Barbara Flannery

Although craving assessment is a common feature of most clinical trials, often the data are not examined until the conclusion of the trial. It would be helpful for therapists to have access to clients’ craving profiles to determine whether craving is a significant component of their dependence and whether self-reported craving during the course of treatment indicates drinking during treatment. With such information, therapies could be better tailored to meet individuals’ needs. At present, a variety of different craving measurements are being used, but their utility in assessing risk for in-trial drinking has yet to be determined.

The present analyses of three craving instruments was undertaken to determine and compare the utility of each in predicting drinking during the course of treatment. The three scales were part of the weekly assessment battery in a naltrexone trial recently completed at the Center for the Study of Addictions at the University of Pennsylvania. This trial compared the effect of a 9 month course of 50 mg/day of naltrexone with a 9 month course of placebo and with a 3 month course of 50 mg/day of naltrexone followed by 6 months of placebo. All three groups also received empathetic motivational therapy.

Subjects’ weekly scores on the Penn Alcohol Craving Scale (PACS, Flannery et al., 1999), the Obsessive subscale (items 1–6) of the OCDS (Anton et al., 1995), and the AUQ (Bohn et al., 1995) were analyzed via generalized estimating equations along with pretreatment drinking and weekly in-trial drinking. Of particular interest in these analyses were the relationships among weekly drinking, weekly craving, and drinking during the subsequent week of treatment.

The PACS (Flannery et al., 1999) is a five-item instrument that assesses intensity, frequency, and duration of craving along with ability to resist alcohol if it were available, and it asks responders to give an average craving rating for the past week. The response ranges from 0 to 6 for each item. The obsessive subscale (ODS) of the OCDS (Anton et al., 1995) was chosen rather than the full-scale OCDS because the six items contained in this subscale ask responders to rate their thoughts (craving) about drinking. The compulsive subscale was excluded from analyses because the questions deal with drinking and drinking-related behaviors. These items were thought to measure the potential consequences of craving by asking subjects to report their craving over the last week or 2 week period. The AUQ (Bohn et al., 1995) is an eight-item “here and now” account with a seven-point Likert scale scoring system. Individuals are asked about their desire for alcohol through such statements as, “I want a drink so bad I can almost taste it,” and “If I had the chance to have a drink, I don’t think that I would drink it.”

Surprisingly, pretreatment number of drinks per day in the 90 days that preceded study entry was not a significant predictor of drinking during the trial (2 = 2.75, NS). However, each of the baseline craving measures when analyzed in conjunction with pretreatment drinking proved to be a significant predictor of number of drinks per week during treatment. The OCD was the strongest predictor (2 = 14.02, p = 0.0002) followed closely by the PACS (2 = 11.99, p = 0.0005). The baseline AUQ was not as strong a predictor as the other two scales (2 = 9.31, p = 0.002).

Because the relationship between drinking and craving is unclear (i.e., does craving precede drinking, does it occur in conjunction with drinking, or is it a consequence of drinking?), weekly number of drinks was entered into the generalized estimating equations along with the scores from each craving assessment. In each analysis, the prior week’s craving proved to be a stronger predictor of drinking during the following week than was previous week’s drinking. In this analysis, the PACS was the best predictor (2 = 28.36, p < 0.0001) followed by the ODS (2 = 14.00, p < 0.0002), with the AUQ being the weakest predictor (2 = 6.32, df = 1, p = 0.01). Another important finding was that the weekly in-trial PACS scores clearly differentiated relapsers from those who remained abstinent or drank less than five drinks per week. The ODS also differentiated these two groups but much less clearly than did the PACS. Relapsers and nonrelapsers did not show a difference in their AUQ scores during treatment.

These analyses demonstrate that craving assessments provide useful and predictive information concerning subjects’ drinking during treatment but differ in their ability to do so. Most importantly, weekly PACS and ODS scores were better predictors of subjects’ drinking during the subsequent week of treatment than was subjects’ drinking during the prior week. And, although AUQ scores were also predictive of subsequent drinking, this relationship was not strong compared with that of the PACS and the ODS. These finding suggest that retrospective craving measurement during treatment may be a more useful tool for assessment and treatment of craving than current or “here and now” craving self-reports. Instruments such as the AUQ appear to be better suited to laboratory self-administration and cue-reactivity studies.


Damaris Rohsenow

In keeping with (Kozlowski et al. 1989), Dr. Rohsenow of the Center for Alcohol and Addiction Studies, Brown University, advocated the use of the word urge instead of craving to describe desire, want, wish, thirst, or craving for alcohol. The term urge is more inclusive because urge is used to refer to a full range of desire, whereas craving frequently is used to describe intense or overpowering desires to drink. Furthermore, the term urge separates this component of alcohol dependence from other states and behaviors that may or may not be associated with alcohol urges (e.g., expectancies, intentions, or actual use; Monti et al., 2000a).

Of particular interest with alcoholic patients is their ability to control urges (i.e., when intending not to drink). In assessing whether urge to drink during treatment can be used to determine whether individuals will drink once treatment has ceased, a variety of studies show that the context within which urge is measured is critical. For example, data from several studies show that during inpatient treatment, urge is typically low (e.g., a mean of 1.8 on a 0–10 scale, Monti et al., 1993c) and thus appears to be a poor predictor of future drinking. Weekly urge assessment during outpatient treatment has it own drawbacks. Some of the methodological factors that can pose a problem include the reduced validity of retrospective self-reports and floor effects that may occur as a result of treatment or from avoiding any drinking-related stimuli early in treatment. Additionally, because of confounding that may occur when a person resumes drinking, it is difficult to disentangle urges that precede drinking from urges precipitated by drinking. One can overcome some of these difficulties by examining urges elicited by simulated high-risk situations. Cue-reactivity and role-play methodologies are two contextual methods that have been used to determine whether an individual’s urge predicts subsequent drinking.

In the cue-reactivity procedure (e.g., Monti et al., 1993b), patients are exposed to their alcoholic beverage of choice and asked to hold and sniff the glass whenever signaled during 3 min periods. Subjective reactions are assessed by using 11-point Likert rating scales of urge and attention (to beverage and to one’s internal reactions; Rohsenow et al., 1994). Objective reactions include cardiovascular responses and measuring the degree of salivation in response to the alcohol by using cotton dental rolls. Subjects also are exposed to water or juice as a control condition. In the role-play procedure, individuals are asked to respond to a variety of situations that pose a high risk for drinking (Monti et al., 1993a). The role-plays are videotaped and rated by trained staff for resistance skill and anxiety. After each situation, patients complete assessments of urge, anxiety, and difficulty in resisting urges. Both methods have excellent reliability (Monti et al., 2000b).

Urge in response to laboratory cues has been an inconsistent predictor of drinking outcome. Contrary to expectations, higher self-reported urge in response to alcohol cues early in treatment predicted less frequent posttreatment drinking (Monti et al. 1993c) and a longer latency to relapse (Drummond and Glautier, 1994). On the other hand, (Cooney et al. 1997) found that subjects who reported a higher urge (when alcohol cues were coupled with negative affect induction) had a shorter latency to relapse. When (Rohsenow et al. 1994) examined salivation and urge together, they found that salivation, but not urge, was predictive of posttreatment drinking frequency. In a second analysis, higher salivary responses and lower self-reported attention to the alcohol cues were both predictive of greater drinking frequency. In a recently completed study, Rohsenow and colleagues (D. Rohsenow et al., unpublished data, 1994) found that urge and attention to alcohol stimuli were not significant predictors of relapse after treatment but that salivary response did predict posttreatment relapse. These findings suggest that salivation may be a marker of automatic drug-seeking processes, whereas attention is indicative of nonautomatic processes that may inhibit drinking.

The predictive power of urges assessed by using the role-play procedure has been investigated in two studies. (Monti et al. 1990) found that higher urge (averaged across 10 simulated drinking situations) predicted greater drinking during a 6 month posttreatment follow-up period. Urge proved to be an even better predictor of subsequent drinking when assessed immediately after each 3 min relaxation period of individual role-play scenarios. In a recently completed study, P. M. Monti and colleagues (unpublished data, 2000) found that higher urge during posttreatment role-play scenarios predicted a significantly higher quantity and greater frequency of drinking during a 3 month posttreatment follow-up period.

These studies have several important implications for research and treatment. First, urge in response to simple situations such as the presence of an alcoholic beverage may be easier for individuals to handle because theses situation require an obvious and simple coping response: drink refusal and escape. Urges in these simple situations may provide an obvious danger signal and thereby may allow individuals to immediately mobilize coping strategies to prevent drinking. Therefore, urges in response to cue exposure may not be a reliable predictor of subsequent drinking. Alternatively, high-risk situations, such as those represented in the alcohol-specific role-play task, may represent a context for urge assessment that is a particularly good predictor of risk for poor outcomes. Urges for alcohol in the presence of complex social and emotional cues (as in the situations presented in the role-play task) may be more difficult to recognize and handle. First, individuals may not recognize their urges in response to these complex situations due to the presence of other strong emotions and a high cognitive load and, thus, may be less aware of the need to cope with these urges. Second, these complex emotional and interpersonal situations appear to be more difficult to cope with effectively and thus require more complex sets of cognitive and behavioral coping skills (Monti et al., 1989). Therefore, even when aware of the need to cope, many alcoholics may not be able to provide an effective set of coping responses, which increases their risk of drinking. Using the role-play methodology to identify individuals with a greater need for additional treatment could be a useful adjunct to clinical practice.


The importance of craving lies in its relationship to drinking and relapse. Each of the speakers presented information about craving (or urge to drink) that detailed how its occurrence impacts subsequent drinking. Through her discussion of various paradigms, Dr. Roberts showed that preclinical research is being used successfully as a tool to assess how stimuli associated with alcohol elicit alcohol-seeking behaviors. These alcohol-seeking or anticipatory behaviors are used to model human craving. Factors such as alcohol withdrawal and protracted abstinence were shown to increase anticipatory behaviors, and pretreatment with pharmacological agents (e.g., naltrexone and acamprosate) decreased these behaviors. A recent finding which showed that D-Phe-CRF, a corticotropin-releasing factor antagonist, blocks self-administration provides evidence that the stress response system is involved in relapse and most likely plays a role in craving.

Dr. Swift presented clinical data about the effectiveness of various pharmacological agents in the treatment of alcohol dependence and craving. Although numerous medications have been tested, their effectiveness for reducing consumption and craving generally has been modest. According to Dr. Swift, a better understanding of the concept of craving and its relationship to drinking will lead to a better understanding of the neurobiology that underlies craving and drinking and to the development of more effective pharmacological treatments. Continued exploration of the neuropharmacology of alcohol dependence through preclinical research appears to be critical for attaining this goal.

Drs. Cooney, Anton, Flannery, and Rohsenow detailed different techniques for assessing craving and for using craving as a predictor of subsequent drinking. Dr. Cooney presented information about a relatively new and sophisticated technique for measuring craving—EMA. Participants in clinical trials are given hand-held computers on which they can record not only craving but also thoughts and behaviors related to craving. This technique provides more information than does the standard weekly pencil-and-paper measurement of craving, because states and behaviors can be reported at the time they occur and within the context within which they occur. As with any assessment technique, EMA also has disadvantages but appears to be a promising method for increasing our understanding of craving and alcohol dependence.

Dr. Anton’s group is using several different techniques to explore craving, which include within-treatment self-report by using the OCDS, fMRI of cue-elicited craving, and a laboratory cue exposure and self-administration procedure. The OCDS has been shown to be useful in differentiating social drinkers from alcohol-dependent individuals and for assessing severity of alcohol dependence. Recent analyses have shown that treatment with naltrexone reduces scores on the resistance/control impairment OCDS subscale, which suggests that this medication may be useful for controlling urge (craving) to drink. Preliminary imaging studies of cue-induced craving suggest that craving is mediated by several brain substrates, which include the left dorsolateral prefrontal cortex and thalamus. The recently developed laboratory paradigm described by Dr. Anton appears to be useful for the study of alcohol’s stimulating and sedating effects as well as cue-elicited and alcohol-elicited craving. The procedure has yielded several interesting preliminary findings: early-stage alcohol-dependent individuals appear to be sensitized to alcohol’s stimulating effects, and alcohol consumption has differential effects on craving for social drinkers and early-stage alcoholics. Although social drinkers report decreased craving after consumption, early-stage alcoholics maintain a level of craving that is comparable to the level reported before consumption.

Dr. Flannery provided important data which showed that weekly in-treatment assessment of craving can be a useful tool to predict subsequent in-treatment drinking. Analyses of three craving assessments (the PACS, the ODS, and the AUQ) used in a 9-month naltrexone trial revealed that craving scores were actually stronger predictors of drinking during the week after assessment than was the number of drinks in the prior treatment week. If such information were made available during treatment, it could help clinicians tailor therapy to meet individual clients’ needs. For individuals with a strong correlation between craving and drinking, therapists could offer strategies for coping with and resisting urges to drink.

Dr. Rohsenow presented data on the relative usefulness of two types of craving assessment procedures for predicting drinking after treatment. Although the cue reactivity paradigm has been used widely, subjects’ levels of craving and attention and emotional state in response to alcohol have been inconsistent predictors of subsequent drinking. The alcohol-specific role-play task, on the other hand, shows more promise. Urge to drink assessed after exposure to individualized high-risk drinking situations was a strong predictor of drinking during 3 and 6 month posttreatment follow-up periods. Dr. Rohsenow suggested that the differential predictive utility of these procedures lies in the complexity of the situation. Cue exposure represents a relatively simple stimulus-response sequence: Alcohol is presented in isolation, and the response is obvious—resistance through removing oneself from the danger signal. In the role-play task, the individual is presented with a complicated real-life situation that includes multiple cognitive and emotional components. In the presence of complex social and emotional cues, urge may not be recognized and/or may be more difficult to overcome.

There is considerable debate about the precise meaning of the term craving. Indeed, some of the speakers at this roundtable symposium presented different interpretations of craving (e.g., obsessional thoughts about alcohol versus any desire to consume alcohol). There was no disagreement, however, about the pivotal role that this phenomenon plays in alcohol dependence as evidenced by the quantity of empirical studies that focus on craving. Craving represents a unique research challenge because we continue to develop techniques to measure it yet are unable to precisely define what we are measuring. Although the techniques and data presented in this symposium may not bring us closer to a consensus definition of craving, they do show that researchers willing to take on the challenge of craving are increasing our understanding of this complex phenomenon.

Many thanks to Helen Pettinati for co-chairing this roundtable symposium and to Melanie Butler for help with manuscript preparation.


Alexander DeLuca, M.D., FASAM.
Copyright © 1999. All rights reserved.                            [Top of Page]
Revised: June 16, 2001.
Dr. DeLuca's Addiction Website