The three year course of alcohol use disorders in the general population: DSM-IV, ICD-10 and the Craving Withdrawal Model

Aims  To determine the course of alcohol use disorders (AUD) in a prospective general population study using three different classification systems: Diagnostic and Statistical Manual version IV (DSM‐IV), International Classification of Diseases version 10 (ICD‐10) and the craving withdrawal model (CWM). The latter is an alternative classification, which requires craving and withdrawal for alcohol dependence and raises the alcohol abuse threshold to two criteria. Design, setting and participants  Data were derived from the Netherlands Mental Health Survey and Incidence Study (NEMESIS), a large (n = 7076) representative general population study with a baseline and 1‐ and 3‐year follow‐up assessments. Measurements  Diagnostic status according to DSM‐IV, ICD‐10 and CWM at baseline and at follow‐up was established using a structured interview (Composite International Diagnostic Interview: CIDI). Findings  DSM‐IV abuse, ICD‐10 harmful use and CWM abuse all showed a favourable course with remission rates of 81, 89 and 71%, respectively, at 1‐year follow‐up and 85, 92 and 79% at 3‐year follow‐up. Dependence according to DSM‐IV, ICD‐10 and CWM had a somewhat less favourable course, with remission rates (no dependence) of 67, 67 and 57% at 1‐year follow‐up and 74, 69 and 73% at 3‐year follow‐up, respectively. Subjects who were remitted at 1‐year follow‐up showed relapse‐rates of 0–14% for dependence and 4–12% for abuse at 3‐year follow‐up. Although CWM diagnoses tended towards greater diagnostic stability than DSM‐IV and ICD‐10, most differences were not significant. Conclusion  The conviction that addiction is a chronic relapsing disease may apply to treatment‐seeking alcoholics, but our data show a far more favourable course of alcohol use disorders in the general population.

Keywords: alcohol dependence; general population; recovery; relapse; Alcohol abuse

Addiction is regarded increasingly as a chronic relapsing disorder (e.g. [[1]]). This conviction is based on the long‐term course of treatment‐seeking alcoholics and drug addicts. However, clinicians tend to overestimate the chronicity of diseases, as clinical samples are biased toward severe cases of long duration [[2]].

Follow‐up studies of treated alcoholics do indeed show a serious course with a strongly elevated mortality risk. In an overview of 10 long‐term follow‐up studies of treated alcoholics, the annual remission rate for the survivors varied from 2.1% to 6.0%, while the annual mortality rate varied from 1.6% to 3.7% (1.6–4.7 times greater than expected) [[3]]. The majority of more recent clinical studies also show a chronic course and high mortality rates [[4]], although some long‐term studies show abstinence rates of up to 53% for surviving subjects [[8]]. However, Jin et al. studied former alcoholics who had been abstinent for a mean period of 4 years and found substantial relapse rates in this group (31%), indicating a chronic relapsing course [[10]].

Recently a large general population study on the course of alcohol dependence was published, showing that 25% of subjects with prior‐to‐past year dependence still met the full criteria of that diagnosis in the past year [[11]]. However, this was a retrospective study with the limitations of selective survival and recalling problems.

Also, some prospective general population studies of alcohol use disorders (AUD) according to Diagnostic and Statistical Manual version IV (DSM‐IV) have been performed. Booth et al. studied the 18‐month course of a community sample of 'at risk drinkers' (defined as having at least one life‐time AUD criterion and either displaying AUD criteria or drinking heavily during the past 12 months). Among subjects with a baseline AUD diagnosis they found stable remission rates of 38%, while 9% showed a stable chronic course. The remainder met criteria for AUD at some, but not at all follow‐up moments. Unfortunately, dependence and abuse were not considered separately [[12]]. Hasin et al. studied the 1‐year course of a community sample of heavy drinkers and found remission rates of 61% for subjects with alcohol abuse, but only 33% (29% no diagnosis, 4% abuse) for subjects with dependence [[13]]. This study among heavy drinkers has a bias towards more severe cases, because a substantial proportion of subjects with AUD, especially abuse, do not drink heavily [[14]]. Finally, Hasin et al. studied the 4‐year course of DSM‐IV AUD among all male current drinkers in a general population sample and found 46% remission (no AUD) for abuse and 54% (15% abuse and 39% no AUD) for dependence [[17]]. This study, however, has two important limitations; first, only male subjects were studied and secondly, no validated instrument was used to assess the diagnostic criteria and not all items of dependence were measured.

In conclusion, the course of alcoholism has been studied extensively among clinical samples, but general population studies are few and have methodological shortcomings. In this study we examine the course (1 and 3 years) of abuse and dependence in a large representative general population sample (n = 7076), including all drinkers and using a validated diagnostic instrument.

In addition to DSM‐IV and International Classification of Diseases version 10 (ICD‐10), we use an alternative classification system for diagnosing AUD, the craving withdrawal model (CWM). CWM is an alternative classification for AUD that was designed in order to overcome two problems of the DSM‐IV AUD classification: absence of craving in the dependence criteria and low validity of the abuse category (see our earlier reports) [[14], [18]]. CWM has a more serious dependence category that requires both craving and withdrawal to be present. Furthermore, CWM increases the abuse threshold to two AUD items (see Table 1 for comparison with DSM‐IV and ICD‐10 diagnoses). So far, CWM has been validated by comparing it to DSM‐IV and ICD‐10 in two cross‐sectional studies on several items, such as drinking behaviour, social functioning and biochemical markers. These studies show that (a) DSM‐IV abuse subjects differed little from subjects with no AUD diagnosis, (b) ICD‐10 harmful use subjects were more severe, but they could hardly be distinguished from subjects with dependence and (c) CWM resulted in a clear distinction between no diagnosis and abuse and between abuse and dependence [[14], [18]]. Other aspects of validity of this model remain to be assessed. In this study we compare the course and diagnostic stability of CWM to DSM‐IV and ICD‐10.

1 Alcohol use disorders according to DSM‐IV, ICD‐10 and CWM.

Summarizing, the main question of the present study is: what is the course and diagnostic stability of alcohol abuse and dependence in the general population? A second question is whether the three classification systems, DSM‐IV, ICD‐10 and CWM, differ in the course of the diagnostic categories.

The data were derived from the Netherlands Mental Health Survey and Incidence Study (NEMESIS). NEMESIS is a prospective study collecting data in three waves (1996, 1997 and 1999) from a national, multi‐stage random sample (aged 18–64 years) in the Netherlands. At the first wave (T0) a total of 7076 people were interviewed (response rate 69.7%). The respondents reflected the Dutch population adequately. For more detailed information see an earlier report on NEMESIS [[19]]. Because the CIDI section on alcohol use disorders was only administered fully in subjects who drank at least 12 alcoholic units within 1 year at any time, the current study was restricted to the 6041 subjects (85.4%) who met this criterion. Of these subjects, 4853 (88.3% of T0) were re‐interviewed after 1 year (T1) and 4214 (69.8% of T0, 86.8% of T1) were re‐interviewed after 3 years (T2). After correction for gender, 12‐month alcohol use disorders at T0 did not increase the probability of loss to follow‐up at T1, but loss to follow‐up at T2 was increased for abuse at T0 according to DSM‐IV (OR = 1.35, 95% CI = 1.06–1.73) and CWM (OR = 1.44, 95% CI = 1.12–1.85) and for both harmful use and dependence at T0 according to ICD‐10 (OR = 1.51, 95% CI = 1.01–2.26 and OR = 1.63, 95% CI = 1.05–2.53, respectively).

The CIDI 1.1 was used to assess criteria of alcohol use disorders during the past 12 months at T0 and T1 and during the last 24 months at T2. The CIDI 1.1 is a reliable and validated, fully structured diagnostic interview, enabling to make diagnoses according to ICD‐10 and DSM‐III‐R criteria [[20]]. The interviewers had been given a 4‐day training course at the WHO–CIDI training and reference centre of the Academic Medical Centre in Amsterdam. Based on the CIDI 1.1, AUD diagnoses according to ICD‐10, DSM‐IV and CWM were made. As DSM‐IV uses the same AUD symptoms as DSM‐III‐R, we could make DSM‐IV diagnoses based on the CIDI answers. CWM dependence was diagnosed if craving and withdrawal were both present. The ICD‐10 criterion 'strong desire or compulsion to drink alcohol' was used for craving in the CWM diagnosis of dependence. For withdrawal, we required the presence of tremor. Tremor is the most robust clinical feature of withdrawal, giving withdrawal a stronger prognostic meaning [[21]]. CWM abuse was diagnosed if subjects did not meet CWM dependence criteria and met at least two of the other DSM‐IV (abuse and dependence) criteria (see Table 1).

For all diagnostic systems at baseline and at follow‐up, a subject was diagnosed as having either abuse or dependence only when the subject met full criteria. Subjects who were partially remitted did not obtain a diagnosis. Subjects with a past dependence diagnosis, who met present criteria for abuse only, were diagnosed as having abuse.

If statements are made about several of the diagnostic systems below, ICD‐10 harmful use is regarded as an abuse category and is not always mentioned separately.

Each respondent was diagnosed according to 12‐months DSM‐IV, ICD‐10 and CWM criteria at baseline. The diagnostic groups were compared regarding their diagnostic status at follow‐up. Mantel–Haenszel odds ratios (ORs) were assessed for abuse and dependence at baseline and at follow‐up, adjusting for the possible influence of gender.

The necessity of adjusting for sampling characteristics study design in analytical studies is the subject of ongoing debate. Most of the time overall conclusions do not change substantially when sample weights are used [[23]]. In addition, the weighted prevalence rates of the various DSM‐IV disorders measured in NEMESIS have already been described elsewhere [[26]]. Therefore, in the present analyses, we decided not to use sample weights.

All statistical analyses were performed with the Statistical Package for Social Sciences (SPSS for Windows, 12.0, 2003).

Table 2 displays the characteristics of the 6041 subjects at baseline. Furthermore, abuse and dependence rates are given for DSM‐IV, ICD‐10 and CWM. DSM‐IV dependence and ICD‐10 dependence showed good agreement (kappa 0.8). In contrast, 83% of the subjects with ICD‐10 harmful use, did not meet criteria for an AUD diagnosis according to either DSM‐IV or CWM. Of the subjects with DSM‐IV and ICD‐10 dependence, 22% and 24%, respectively, also met criteria for dependence according to CWM, the remainder was diagnosed with abuse according to CWM. Nearly half the subjects with DSM‐IV abuse did not obtain a CWM diagnosis.

2 Sample characteristics at baseline based on unweighted data.

Of the subjects with an AUD according to either diagnostic system, 2.4% (n = 12) had been seeking specialized treatment for alcohol use disorders in the last 12 months. The CWM dependence category contained the highest proportion of subjects seeking treatment: 25% (n = 5).

For more details on the baseline characteristics of the diagnostic groups according to CWM, DSM‐IV and ICD‐10 we refer to our earlier study comparing these classification systems [[14]].

Table 3 shows the course of baseline diagnostic categories according to DSM‐IV, ICD‐10 and CWM at 1 year and at 3‐year follow‐ups. Furthermore, ORs are displayed for abuse and dependence at baseline predicting diagnostic status at follow‐up.

3 Diagnostic categories at baseline (T 0) and at 1‐year (T 1) and 3‐year (T 2) follow‐ups.

1 Numbers displayed are percentages, numbers in between brackets are numbers of subjects. Odds ratios (ORs) are displayed for subjects with abuse opposed to subjects with no AUD diagnosis at baseline (T0) predicting abuse or dependence at follow‐up (T1 and T2). ORs are also displayed for subjects with dependence opposed to subjects with no dependence at baseline (T0) predicting dependence at follow‐up (T1 and T2). *Mantel–Haenszel (corrected for gender) significant at P < 0.05. †Odds ratio not calculated because no subject without a diagnosis developed dependence.

Overall, a large proportion of the subjects with abuse at baseline according to DSM‐IV, ICD‐10 or CWM remitted over a period of 1 year (81%, 89% and 71%, respectively) and 3 years (85%, 92% and 79%, respectively). Abuse significantly predicted having abuse or developing dependence at follow‐up for all diagnostic systems, except DSM‐IV abuse, which did not significantly predict dependence at 3‐year follow‐up (OR 2.6, 95% CI 0.9–7.9). Of the subjects with dependence at baseline according to DSM‐IV, ICD‐10 and CWM, 67%, 67% and 57%, respectively, did not meet diagnostic criteria for this diagnosis at 1‐year follow‐up and 74%, 69% and 73% no longer met dependence criteria at 3‐year follow‐up (either full or partial remission). Dependence significantly predicted having dependence at follow‐up for all diagnostic systems. Moreover, dependence had higher ORs than abuse for predicting dependence at follow‐up, although in some cases (ICD‐10 and CWM at 3‐year follow‐up) the confidence intervals overlapped.

Generally, CWM diagnoses displayed a more serious course with lower remission rates and higher ORs than DSM‐IV and ICD‐10. At 1‐year follow‐up, the OR for CWM dependence was significantly higher than for ICD‐10 dependence; in all other cases the confidence intervals overlapped.

The small subgroup of subjects seeking specialized treatment (n = 12) displayed a more serious course. Of the subjects with dependence according to either diagnostic system, 67% still met the criteria for dependence at 1‐ and 3‐year follow‐ups.

We also examined the relapse rate between T1 and T2 of subjects with abuse or dependence at T0, who did not meet the criteria of an AUD diagnosis at T1.

Of the subjects with DSM‐IV dependence at T0 and no diagnosis at T1, only 13.6% (n = 3) relapsed into dependence and 4.5% (n = 1) met the criteria for abuse at T2. For ICD‐10, relapse rates were 11.5% (n = 3) into dependence and 3.8% (n = 1) had harmful use at T2. Of the subjects with CWM dependence at T0 and no diagnosis at T1, none relapsed into either abuse or dependence between T1 and T2.

Of the subjects with abuse at T0 and no diagnosis at T1, only one developed dependence according to either DSM‐IV, ICD‐10 or CWM at T2. Relapse rates into abuse were 10.0% (n = 15) for DSM‐IV, 3.8% (n = 2) for ICD‐10 and 12.0% (n = 15) for CWM.

This is the first prospective study in a representative general population sample, including both sexes, investigating the course of abuse and dependence with a standardized validated interview (CIDI). In this study, the course and diagnostic stability of three classification systems are compared. There are, however, some limitations to the study design that deserve comment.

First, it is necessary to be cautious about interpreting CIDI items as DSM and ICD‐10 criteria. However, CIDI has been well validated against other structured interviews. Although for other psychiatric disorders CIDI can be overly inclusive, this is not the case for alcohol use disorders [[27]].

Secondly, we did not weight our data in order to adjust for the special characteristics of the sampling procedure. In the NEMESIS study the 18–24 year age group, in which substance use disorders are common, was significantly under‐represented [[19]]. Therefore, our prevalence rates are an under‐estimation of true prevalence rates. The prevalence rates found by NEMESIS and the comparison to other epidemiological findings have been described extensively previously [[26]]. Furthermore, our prevalence rates are in general accordance with the 12‐month rates found in some other recent general population studies using the CIDI [[31]].

A third problem to consider is attrition. Having abuse according to any of the diagnostic systems and having ICD‐10 dependence at baseline predicted loss to follow‐up in the second wave. This might have led to a slight underestimation of the proportion that kept their diagnosis at 3‐year follow‐up. Another possible caveat resulting in a slight overestimation of the remission rates is that part of our findings might be explained by the statistical artefact of regression towards the mean.

Another limitation of the study design is the fact that the follow‐up was only 3 years. In order to evaluate the relapse rates of AUD in the general population, longer follow‐up studies are desirable.

CWM was designed as an alternative classification system, in order to overcome the main disadvantages of the DSM‐IV by focusing on craving and increasing the abuse threshold. In our earlier studies, comparing CWM to DSM‐IV and ICD‐10, we found that CWM increased the discriminant validity of AUD [[14], [18]]. One of our aims was to further validate this model. CWM dependence and abuse showed lower remission rates and higher ORs for predicting diagnosis at follow‐up than DSM‐IV and ICD‐10. However, due probably to the small number of cases with CWM dependence (n = 20), the confidence intervals overlapped in most cases. On the basis of our results, we cannot state that CWM has a better predictive validity than DSM‐IV or ICD‐10.

The prevalence and overlap data indicate that CWM dependence is a more serious subcategory of DSM‐IV and ICD‐10 dependence. In our earlier study on these subjects, we found that the group of subjects with CWM dependence had a mean drinking quantity of 93.7 alcoholic units per week at baseline [[14]]. Even of these serious drinkers only 43% and 27% still met the criteria for dependence at 1‐ and 3‐year follow‐ups.

Nearly half of the subjects with DSM‐IV abuse did not receive a CWM diagnosis. In our earlier study on these subjects, we found that over 90% of the subjects with DSM‐IV abuse obtained their diagnosis based on one symptom and 74% of them had their diagnosis based on the item 'drinking in situations in which it is hazardous'[[14]]. The present data indicate that over 80% remit in either 1 or 3 years. One could argue that these subjects did not have an alcoholic disease in the first place, but rather time‐limited irresponsible behaviour [[34]].

So far, CWM has several advantages over DSM‐IV. Several aspects of validity still require further research. In order to provide more conclusive data on the predictive validity, clinical studies examining the course of CWM must be conducted, resulting in larger numbers for the dependence group. Given the emphasis of CWM dependence on craving and withdrawal, it would be interesting to find out whether CWM abuse and dependence result in different findings in neurobiological and treatment studies. The usefulness and validity of CWM should be expanded to other addictive substances as well, considering both physiological and psychological aspects of withdrawal.

Our data show that, in the general population, a large percentage of the subjects with abuse and even dependence remit and relapse rates are low. Our prevalence rates at T2 might even be a slight overestimation (and the remission rates a slight underestimation), as the second follow‐up concerned 24‐month prevalence instead of 12‐month prevalence. The favourable course over 4 years is an important finding for disputing unnecessary pessimism and a lack of treatment motivation by primary care physicians [[36]].

Our remission rates differ from the rates found by Hasin et al. [[13], [17]]. As already stated in the Introduction, these studies differ from ours in some aspects. One study was restricted to heavy drinkers [[13]], the other was restricted to men and did not use a validated diagnostic instrument [[17]]. Furthermore, in both studies partially remitted subjects were regarded as still having dependence at follow‐up, explaining part of their lower remission rates. In our study, we diagnosed subjects at baseline and at follow‐up if they met full criteria. A potential risk of this approach is that a waxing and waning course might be regarded as a remission. However, our low relapse rates were reassuring in this aspect. Our high remission rates converge with Booth et al. [[12]], who found that only 9% of their general population sample with a baseline AUD diagnosis showed a stable chronic course over 18 months. At 18‐month follow‐up, 77% were in remission. A disadvantage of this study is that abuse and dependence were not considered separately, while in our study (and other studies) those diagnoses differ substantially both in severity and course. Finally, our results converge with the results of the retrospective study on alcohol dependence by Dawson et al. who found in their general population sample that only 25% of subjects with a past dependence diagnosis still met the criteria for this diagnosis in the last year [[11]]. However, the remission rates reported in the Dawson et al. study probably reflect remission over a longer period of time.

Only a small minority of our sample had sought specialized treatment for their alcohol problems. It usually takes more than 10 years before patients with alcohol‐related problems enter addiction treatment [[37]]. The finding that a substantial part of alcohol‐dependent individuals recover without formal treatment has been described frequently since Vaillant published his follow‐up studies [[38]]. The Vaillant data concern two samples, former Harvard undergraduates and non‐delinquent socially disadvantaged subjects. He followed them every 2 years for a period of 60 years, thereby collecting valuable information on the course of alcoholism in the community. Several other studies have described aspects of 'natural versus treated remissions', such as reasons for seeking treatment [[39]] and the course and characteristics of the subjects who recover with and without treatment [[41]]. Caetano suggested that some cases of alcohol dependence in the general population should be viewed as less severe forms of the phenomena seen in clinical samples. He expected this 'less severe syndrome' to be less stable in time [[15]]. Our findings confirm that the 3‐year course of alcohol dependence in the general population is far more favourable than findings from treatment‐seeking populations. Also, our subgroup of subjects seeking treatment displayed a more serious course, but this finding has to be interpreted with caution because of the small numbers.

This is the third study indicating that the CWM classification has several advantages over the DSM‐IV AUD classification. However, other aspects of the validity of CWM remain to be assessed.

Furthermore, the notion that alcoholism is a chronic relapsing disease applies to populations seeking treatment in specialized treatment centres. However, the 3‐year course of AUD in our general population sample shows high remission rates (either full or partial) and low relapse rates.

McLellan suggested that recognizing the chronic nature of addiction implicates that primary care physicians should be better educated in screening and diagnosis of addiction, brief interventions, medication management and referral criteria [[44]]. Our findings suggest that it is also important to realize the difference between the chronic, severely ill alcoholics who need long‐term or even life‐long specialized treatment, and the patients who are much more common in the general population and who can apparently overcome this disease with a brief intervention or even without formal treatment. The statement that addiction is a chronic relapsing disorder seems to represent a classic case of what has been called the 'clinicians' illusion'[[2]].