Blackwell Science, LtdOxford, UKADDAddiction1360-0443© 2003 Society for the Study of Addiction to Alcohol and Other Drugs981217371746Original ArticleGallus

Bischof et al.Types of natural recovery from

alcohol dependence

RESEARCH REPORT

Types of natural recovery from alcohol dependence: a

cluster analytic approach
G. Bischof1, H. -J. Rumpf1, U. Hapke2, C. Meyer2 & U. John2
University of Lübeck, Department of Psychiatry and Psychotherapy, Germany1 and Ernst-Moritz-Arndt University Greifswald, Department of Epidemiology and
Social Medicine, Greifswald, Germany2

Correspondence to: ABSTRACT

Gallus Bischof
Department of Psychiatry and
Psychotherapy
Background Social capital and a low severity of alcohol-related problems have
Research Group STEP (Substance Abuse: been focused upon to explain the processes of natural recovery from alcohol
Treatment, Epidemiology and Prevention dependence. However, studies using control groups have not found significant
University of Lübeck
differences in these variables. Subtypes of natural remission which might
Ratzeburger Allee 160
23538 Lübeck
account for this inconsistency have only been described on grounds of qualita-
Tel. + 49 451 500 4860 tive data.
Fax: + 49 451 500 3480 Aims To identify subtypes of natural remitters using cluster analysis.
E-mail: gallus.bischof@ukl.mu-luebeck.de
Participants One hundred and seventy-eight media-recruited natural remitters
Submitted 22 January 2003;
were interviewed personally. Several triggering mechanisms and maintenance
initial review completed 4 April 2003; factors of remission were assessed using standardized questionnaires. Based on
final version accepted 20 August 2003 age of onset and severity of dependence, adverse consequences from drinking,
social pressure and social support, cluster analyses were performed.
Results Cluster analyses yielded three groups of natural remitters: one cluster
RESEARCH REPORT with a high severity of dependence, low alcohol-related problems and low social
support (‘low problems—low support’; n = 65), one group characterized by
high severity of dependence, high alcohol-related problems and medium social
support (‘high problems—medium support’; n = 37), and a third group which
consisted of subjects with high social support, late age of onset, low severity of
dependence, and low alcohol-related problems (‘low problems—high support’;
n = 76). Cluster solutions were confirmed using discriminant analyses. Analy-
ses of variance (ANOVAs) revealed further group differences on other triggering
and maintaining factors of remission.
Conclusions Failure to identify specific pointers to natural recovery in previous
research might be due to heterogeneous subgroups of natural remitters. In
order to build a conceptual framework for understanding the processes of
natural recovery, interactions of different independent variables should be
considered.

KEYWORDS Alcohol dependence, natural recovery, social capital, subtypes.

INTRODUCTION stance-related disorders without the utilization of profes-

sional help, give evidence that the majority of changes in
Since the mid-1970s, a growing number of studies on the addiction field taking place without professional assis-
recovery from substance use problems without profes- tance in various countries reveal rates between 66.7% in
sional help have been published (Vaillant 1982; Klinge- Germany (Rumpf et al. 2000) to 77% in Canada (Sobell,
mann 1991; Sobell, Sobell & Tonneatto 1992; Tucker & Cunningham & Sobell 1996).
Gladsjo 1993; Bischof et al. 2000; Bischof et al. 2001). Historically, research on natural recovery can be
Epidemiological studies, considering pathways out of sub- described as a twofold process; at the first stage, the main

© 2003 Society for the Study of Addiction to Alcohol and Other Drugs Addiction, 98, 1737–1746
1738 Gallus Bischof et al.
interest of research lay in proving natural recovery to be a
real phenomenon. These studies usually argued from a
descriptive level, without taking control groups into con-
sideration (e.g. Tuchfeld 1981; Stall 1983). In the early
1990s, a second wave of research began to focus on dif-
ferences between treated and untreated recoverers
(Klingemann 1991; Sobell et al. 1992; Tucker & Gladsjo
1993; Blomquist 1999). However, most findings revealed
only small differences between treated and untreated
subjects. Although empirical evidence is still scarce, the
main topics in the literature on natural recovery are con-
cepts such as ‘social capital’ (Granfield & Cloud 1996) or
less severe alcohol-related problems (Cunningham 1999;
Klingemann et al. 2001) to explain resources viewed as
crucial for successful unassisted recoveries. Only one
study revealed a substantial impact of problem severity
on help-seeking and of social resources on natural recov-
ery (Blomquist 1999); however, these findings are
restricted due to small sample sizes.
Concepts such as problem severity and social
resources may interact reciprocally; e.g. when problems
are small, less social capital is necessary to overcome an
addictive disorder without treatment, while in the more
severe cases social capital might be a necessary prerequi-
site for recovery without specific treatment. Therefore,
lack of empirical evidence for concepts such as social cap-
ital in natural recovery might be due to the compensation
of different factors in heterogeneous populations.
Early studies on natural recovery based on qualitative
data emphasized the heterogeneous nature of unassisted
pathways out of addiction. For example, Klingemann’s
(1991) groundbreaking study on natural recovery
described a typology of motivation to stop on grounds of
qualitative data. At first, in concordance with the concept
of Alcoholics Anonymous (AA), one group of recoverers
can be described as ‘hitting bottom’, e.g. characterized by
physical, interactional and psychological collapse. How-
ever, hitting bottom may be put into perspective by two
other groups: first, one group was characterized by
Klingemann as ‘cross-road types’, who act on the basis of
a single crisis (such as health or psychological problems);
another group consists of ‘pressure-sensitive types’, who
react positively to social pressure which forces them to
choose between a life of conformity or an addiction
career. In contrast to this group, which can usually define
a specific turning point, another group was characterized
as slowly and harmoniously drifting out of addiction.
Their motivation to quit is based on positive changes in
their social environment. Another group was described
as having had esoteric or religious experiences as turning
points. However, Klingemann’s typology is based on qual-
itative data from interviews of the limited number of 60
former addicts, of whom 30 were formerly alcohol-
dependent and 30 were formerly heroin-dependent.
© 2003 Society for the Study of Addiction to Alcohol and Other Drugs
Broadly, the types generated by Klingemann differ on
at least two dimensions: problem severity (including
social pressure) and resources (social capital). In the
present analysis, we analysed data on remitters from
alcohol dependence without formal help, using cluster
analysis in order to find homogeneous subgroups. We
considered severity of dependence, adverse consequences
from drinking and social pressure to change drinking
behaviour as different aspects of problem severity. As
social resources, we took into account overall social sup-
port (see Methods section). Because general typologies of
alcohol-dependent subjects (e.g. Cloninger, Bohman &
Sigvardsson 1981) have emphasized further the role of
early development of dependence, we also considered age
at onset of dependence as a clustering variable. Cluster
solutions are cross-validated using analyses of variance
(ANOVAs) on other variables not used to form the
clusters.
METHOD
The sample described in this paper is part of the project
‘Transitions in Alcohol Consumption and Smoking’
(TACOS) (ANEPSA Research Group 1998). Recovery was
defined as meeting Diagnostic and Statistical Manual-4th
edition (DSM-IV) (American Psychiatric Association
1995) criteria of alcohol dependence life-time but not
within the last 12 months; participants fulfilled DSM-IV
remission specifiers of sustained full remission (meeting
none of the dependence criteria for at least 12 months).
Furthermore, subjects did not fulfill criteria for alcohol
abuse and did not exceed limits of risky alcohol consump-
tion according to the British Medical Association 1995),
defined as 30 g/alcohol daily for men and 20 g/alcohol
daily for women.
Individuals were recruited through media solicitation.
The media solicitation included 17 newspaper articles,
13 newspaper advertisements, two radio reports and one
television report. The recruitment strategy used is com-
parable to other studies (Sobell et al. 1992). The sampling
area covered the north of Germany, including the federal
states Schleswig-Holstein, Hamburg, and parts of Nieder-
sachsen and Mecklenburg-West Pomerania. Only three
individuals could be found through television and radio
reports. The majority responded to the newspaper articles
and advertisements. We included several different news-
papers covering a broad range of readers. In addition, we
included newspapers delivered to all households free of
charge. The heading of the newspaper advertisements
was as follows: ‘The Medical University of Lübeck seeks
individuals who have successfully overcome an alcohol
problem without treatment’. Four hundred and fifty-four
subjects responded to the media solicitation and were
Addiction, 98, 1737–1746

screened by telephone. From the total group, 224 sub-
jects were excluded after the telephone screening because
they did not fulfill the criteria for natural recovery or were
living too far away. The remaining 230 eligible subjects
participated in a comprehensive interview (mean length:
142 min, s.d.: 34.5) conducted by psychologists and were
paid the equivalent of US$20. Payment was not
announced in advertisements and was mentioned for the
first time at the end of the interview. Twenty-six subjects
(11%) were excluded after interview for not fulfilling the
criteria. Subjects were asked to nominate one collateral
informant who could provide information corroborating
the subject’s drinking history and resolution. However, in
order to avoid biasing the sample in favour of subjects
with intact social networks, nomination of one collateral
informant was not a criterion for eligibility. In 89.4% of
all cases, data according to the alcohol dependence syn-
drome, utilization of help, date of remission and alcohol
consumption since remission were confirmed by collat-
eral interviews. If obvious inconsistencies between
respondent and collateral report occurred, respondents
were excluded from the study (2.2%; n = 5).
For the present analysis, 21 participants were
excluded because they did not fulfil DSM-IV specifications
for sustained full remission. The remaining 178 subjects
had never received any kind of alcohol treatment, defined
as counselling, in-patient or out-patient treatment for
alcohol dependence, self-help group participation and
who did not receive Antabuse (disulfiram) (n = 103) or
received minor formal help, defined as life-time contact
with alcohol treatment not exceeding nine self-help
group sessions, five counselling sessions by a physician
not specialized in addictive disorders or three counselling
sessions by a professional in the addiction treatment
(n = 75). A previous analysis revealed no severe differ-
ences between completely treatment-free subjects and
participants who received minor formal help (Bischof
et al. 2002). Any life-time in-patient or out-patient treat-
ment and any psychotherapy of comorbid psychiatric dis-
orders 2 years prior to and 1 year after remission was
defined as exclusion criteria.
Scales used to generate clusters
All scales used to form the clusters were assessed for the
time-frame 2 years prior to remission. Severity of alcohol
dependence was assessed using the SEverity Scale of Alco-
hol dependence (SESA; John & Hapke 2003). This instru-
ment was developed on the basis of the Short Alcohol
Dependence Data Questionnaire (SADD; Davidson &
Raistrick 1986), the Severity of Alcohol Dependence
Questionnaire (SADQ; Stockwell, Murphy & Hodgson
1983) and the Alcohol Dependence Scale (ADS; Skinner
& Allen 1982). SESA provides a measure of the severity of
© 2003 Society for the Study of Addiction to Alcohol and Other Drugs
Types of natural recovery from alcohol dependence 1739
dependence that is not confounded by social conse-
quences of drinking and differentiates between the fol-
lowing subscales based on factor analysis: withdrawal,
consumption to avoid withdrawal, craving, narrowing of
drinking repertoire, tolerance and reverse tolerance. The
standardized maximum score for subscales and overall
severity is 100. Adverse consequences from drinking
were assessed using a German translation of a nine-item
scale derived from the ‘Health and Daily Living Form’
(Moos et al. 1985). It focuses on adverse consequences
from drinking in nine different life domains: health, job,
money problems, family arguments, fights, trouble in the
neighbourhood, trouble with the police, trouble with
friends and driving while intoxicated. A five-point Likert-
type scale ranging from never to often was used. Social
pressure to quit drinking was assessed using a German
translation of a 15-item questionnaire (Hasin 1994; Bis-
chof et al. 2003a) covering social pressure by partner, rel-
atives, others, legal system, work and physician. Possible
scores range from 0 to 22.
Assessment of socio-demographic and
substance-related factors
Quantity and frequency of alcohol consumption was
measured within three time-frames (value of the highest
consumption period over the life-span, the year before
remission and the year prior to the interview). The aver-
age daily alcohol consumption was computed by multi-
plying the quantity and frequency of consumption for
each time-frame. Assessment of socio-demographic vari-
ables included years of schooling, gender, unemployment
prior to remission and marital status at time of interview.
Assessment of triggering mechanisms
We assessed variables that have been found to be related
to natural recovery or help-seeking behaviour (e.g. Sobell
et al. 1992; Tucker, Vuchinich & Gladsjo 1994; Finney &
Moos 1995; Tucker, Vuchinich & Pukish 1995). As the
time-frame for triggering mechanisms, we chose a period
of 2 years prior to the remission except for health-related
locus of control, which was assessed for the current status
at the time of the interview. Reasons for not seeking help
were assessed using a German translation of a 10-item
questionnaire (Michael et al. 2003) consisting of seven
items developed from a follow-up of a general population
study (Hingson et al. 1982) and an additional three items
derived from research on natural recovery (Sobell et al.
1992). The answer format is a five-point scale ranging
from ‘strongly disagree’ to ‘strongly agree’.
Life-events and the perceived relationship of problems
in several life-domains (work, partnership and health) on
the remission process were assessed using five-point
Addiction, 98, 1737–1746
1740 Gallus Bischof et al.
scales ranging from ‘not important at all’ to ‘very impor-
tant’. Health-related locus of control was assessed using a
German version of the Multidimensional Health Locus of
Control Scale (MHLC; Wallston & Wallston 1978; Lohaus
& Schmitt 1989). The questionnaire consists of the fol-
lowing subscales: internal health locus of control, power-
ful others health locus of control and chance health locus
of control. The standardized maximum score for sub-
scales is 100.
Assessment of maintenance factors
As studies on natural recovery focused mainly on trigger-
ing mechanisms, little is known about maintenance fac-
tors (Snow, Prochaska & Rossi 1994). We assessed factors
that have been shown to be important in studies on treat-
ment outcome and relapse (e.g. Booth et al. 1992; Miller
et al. 1996). These factors are assumed to be of similar
importance in natural recovery. One study by our
research group using the present sample revealed more
similarities than differences between successful remitters,
independently from help-seeking status (Bischof et al.
2000). Most of the maintenance factors were assessed for
the period of 1 year after remission. Temptation to drink,
self-effacy to abstain, mental health and satisfaction with
life were assessed for the current status at the time of the
interview.
Social support was assessed using the German trans-
lation of the Social Support Appraisal Scale (SS-A; Vaux
et al. 1986; Laireiter 1996), a 28-item questionnaire
which differentiates between the subscales support by
partner, family, friends and others. The standardized
maximum score for subscales is 100. Post-remission cop-
ing behaviour was assessed using a German translation
of the Coping Behaviour Inventory (CBI; Litman, Stapel-
ton et al. 1983; Bischof et al. 2003b), which differentiates
between four factors: positive thinking (e.g. ‘Pausing and
really thinking the whole alcoholic cycle through’), neg-
ative thinking (e.g. ‘Remembering how I have let my
friends and family down in the past’), avoidance/distrac-
tion (e.g. ‘Keeping away from people who drink’) and
seeking social support (e.g. ‘Telephoning a friend’) on a
four-point scale ranging from ‘never’ to ‘usual’. The stan-
dardized maximum score for subscales is 100. Tempta-
tion to drink and self-efficacy to remain abstinent in 20
different situations were assessed using a German trans-
lation of the Alcohol Abstinence Self-efficacy Scale
(AASE; DiClemente et al. 1994; Bott et al. 2003), which
differentiates between the subscales negative affect (e.g.
‘When I am very worried’), social/positive factors (e.g.
‘When I am excited and celebrating with others’), physi-
cal and other concerns (e.g. ‘When I have a headache’)
and withdrawal and urges (e.g. ‘When I am feeling a
physical need or craving for alcohol’). Temptations and
© 2003 Society for the Study of Addiction to Alcohol and Other Drugs
self-efficacy were assessed on a five-point Likert-type
scale. The standardized maximum score for both scales is
100. The number of people informed about previous
drinking problems was assessed as a maintenance factor
with scores ranging from 0 to 10. Perceived resolution
maintenance factors were rated for 12 different life-
domains on a five-point scale ranging from not important
at all to very important.
Mental health was tapped with the five-item Mental
Health Screening Test (MHI-5; Berwick et al. 1991),
ranging from 0 to 20, and a short questionnaire to assess
sense of coherence (SOC-3; Lundberg & Nyström Peck
1995), ranging from 0 to 6. Satisfaction with life was
tapped with a five-point rating scale differentiating eight
life domains (work, partnership, family, friends, financial
situation, living conditions, health) from the Question-
naire on Health Behaviour (Dlugosch & Krieger 1995)
and the Satisfaction With Life Scale consisting of five
items (Diener et al. 1985). In contrast to the original
instruments, the German translations of the MHI-5 and
the SWLS are used with five-point Likert-type scales. All
German translations of questionnaires were back-
translated by a native English speaker and compared with
the original versions. In case of inconsistencies, transla-
tions were modified.
Statistical analysis
Because clustering techniques tend to create clusters
arbitrarily, even when natural clusters or subtypes are
not inherent in the data structure (Skinner 1982),
whether or not the empirically identified subtypes are
arbitrary has to be evaluated. Critical issues include the
selection of variables and the selection of method. As to
the selection of variables, we used variables that were
shown to be theoretically relevant for understanding the
heterogeneity of natural remitters and that have differen-
tiated natural remitters in previous research. All analyses
were carried out using SPSS 10.0. To identify clusters,
quick cluster (K-means) analysis was performed (Norusis
1993). In contrast with the WARD algorithm, the K-
means algorithm is able to relocate subjects to clusters
during the analysis, in order to optimize homogeneity
within the clusters. As this analysis calls for the number
of clusters to be identified beforehand, a series of hierar-
chical cluster analyses was performed on two subsamples
as guidance for selecting the appropriate number of clus-
ters for the K-means analysis (Norusis 1993). All cluster
analyses were conducted using the variables age at onset
of dependence, adverse consequences from drinking,
severity of alcohol dependence, social support and social
pressure to change drinking behaviour. Clustering vari-
ables showed low (-0.03 for social support and social
pressure) to moderate (0.47 for social pressure and
Addiction, 98, 1737–1746
 Types of natural recovery from alcohol dependence 1741

adverse consequences from drinking) correlations. Table 1 Cross-tabulation of three cluster solutions using Ward’s
Adverse consequences from drinking showed most of the algorithm and a K-means algorithm.a
significant correlations with other variables, such as 3-Cluster solution K-means
r = -0.40 with social support, r = 0.37 with severity of
alcohol dependence and r = -0.26 with age at onset. 3-Cluster solution WARD Cluster 1 Cluster 2 Cluster 3

Cluster 1 2 31b 0 33
Replicability across cluster algorithms Cluster 2 3 5 73b 81
Cluster 3 60b 1 3 64
In a first step, we checked the number of clusters identi- 65 37 76 178
fied in the hierarchical cluster analysis for stability across
different methods and subsamples. Therefore, one ran- a 2
c = 272,75, 4 d.f, n = 178, P < 0.001. bCorresponding cluster: 50% of cases
dom sample representing 50% of all cases was selected from both row and column fall in cluster.

from the total sample of 178. The remaining sample was

used for replication. Two hierarchical cluster algorithms 0.8

(complete linkage, WARD) were applied to random sub-

samples of the primary sample after standardization of 0.7
variables. Because measurement scales, and therefore
means and standard deviations, differed substantially
0.6
between variables, standardization was necessary (Milli-
gan & Cooper 1987). Standardization was carried out
0.5
Standardized

using the range of the variable as divisor. This has been

found to give consistent superior recovery of the underly-
ing cluster structure, compared to the traditional Z- 0.4

standardization in a simulation study by Milligan &

Cooper (1987). 0.3

For each of the hierarchical analyses, the appropriate

LPLS
number of clusters was selected through examination of 0.2
the squared Euclidean distance coefficients (Norusis HPMS

1993), and for each either a three- or four-cluster solu- LPHS

0.1
tion was indicated. In a second step, WARD and K-means Age onset Social pressure Social suport
Adverse consequences Severity dependence
analysis were conducted for the total sample of 178 sub-
jects in order to control for the influence of the chosen Figure 1 Cluster solution K-means; mean values on range standardized
method on the overall results. Both methods revealed a variables for LPLS (‘low problems—low support’), HPMS (‘high
very high correspondence, as displayed in Table 1. All problems—medium support’) and LPHS (‘low problems—high support’)
three clusters were identified independently from the cho-
sen method. The overall correspondence rate between
(mean 8.6; SD 1.9), and medium social support (mean
both methods was 92.1%.
46.2; SD 15.0). The third cluster (n = 76) is characterized
by late onset (mean 37.0; SD 10.8), low adverse conse-
quences from drinking (mean 5.7; SD 3.1), low severity of
RESULTS
dependence (mean 48.8; SD 17.5), low social pressure
(mean 3.8; SD 2.3) and high social support (mean 68.7;
A K-means cluster analysis was used for further analysis.
SD 15.9). In order to rename the groups using meaningful
The following analyses compare solutions for the whole
names, we used social pressure and adverse consequences
sample (n = 178). Mean values and standard deviations
from drinking as indicators of alcohol-associated prob-
are displayed in parenthesis. As displayed in Fig. 1, the first
lems, especially as both variables revealed the highest cor-
cluster (n = 65) is characterized by early onset (nean
relation. Cluster 1 was labelled ‘low problems—low
28.8 years, SD 8,9) with medium adverse consequences
support’ (LPLS), cluster 2 was renamed ‘high problems—
from drinking (mean 9.3; SD 3.9), a high severity of
medium support’ (HPMS) and cluster 3 was labelled ‘low
dependence (mean 64.0; SD 15.4), low social pressure
problems—high support’ (LPHS).
(mean 3.4; SD 1.8) and low social support (mean 34.3; SD
14.3). The second cluster (n = 37) consists of subjects with
Confirmation of results using discriminant analyses
early age onset (mean 26.8; SD 7.0), high adverse conse-
quences from drinking (mean 14.3; SD 5.2), high severity Cluster solutions were confirmed using discriminant
of dependence (mean 66.9; SD 17.6), high social pressure analyses. As shown in Table 2, 99.4% of all cases in the K-

© 2003 Society for the Study of Addiction to Alcohol and Other Drugs Addiction, 98, 1737–1746
1742 Gallus Bischof et al.

means cluster groups were identified correctly. The rate of
correct identification ranged between 98.5% in the first
cluster and 100% in the second and the third clusters.
Differences between groups in variables not associated
directly with cluster variables
To analyse differences between the groups identified in
the cluster analysis, other factors such as influence of life
events, satisfaction with life and impact of partnership,
work and illness were investigated using univariate anal-
ysis of variance (ANOVA). Table 3 displays differences on
socio-demographic and substance-related variables. The
first cluster revealed a higher number of women, were
less often married and revealed a high rate of unemploy-
ment. Subjects in the second cluster were less likely to be
female and were younger. Subjects in the third group had
a higher school education level, and revealed a lower
average daily alcohol intake prior to remission.
Significant differences between the clusters regarding
variables triggering and maintaining the remission pro-
cess are displayed in Table 4. No difference concerning
reasons for not seeking help were identified between
groups. With regard to triggering factors of the remission
process, considerable differences could be identified for
variables associated with satisfaction with life and vari-
Table 2 Discriminant analysis: cross-tabulation of predicted and
actual group membership according to 3-cluster solution K-means.
Predicted group membership
3-Cluster solution
K-means Cluster 1 Cluster 2 Cluster 3
Cluster 1 64 1 0
Cluster 2 0 37 0 37
Cluster 3 0 0 76
64 38 76
ables assessing direct subjective impact of problems or life
events on the remission process. LPHS subjects revealed
higher satisfaction with life domains prior to remission
compared to both the other groups, except for satisfaction
with living conditions, where no differences were found
between LPHS and HPMS. Concerning the impact of life
domains on the remission process, HPMS subjects gave
more emphasis to the impact of partnership on remission
than LPLS subjects. Furthermore, HPMS subjects gave
higher emphasis to the impact of financial and legal
events than did both of the other groups.
Concerning maintenance factors, no differences
between temptation to drink and self-efficacy in remain-
ing abstinent in various high-risk situations, mental
health and sense of coherence were identified. The most
specific pattern of maintenance factors was identified in
HPMS. These subjects informed more people about their
former drinking problems compared to both the other
groups. HPMS revealed more support in remaining absti-
nent by partners and friends compared to LPLS, more
support by their financial situation compared to LPHS
and more support by living conditions compared to both
the other groups. Concerning coping behaviour, HPMS
subjects sought social support more often and used
avoidant coping strategies compared to LPHS and used
negative thinking more often compared to LPLS.
HPMS and LPHS revealed more social support by part-
ners and family even after resolution compared to LPLS,
while LPHS reported more social support by others com-
pared to LPLS. Furthermore, LPHS compared to LPLS
revealed more satisfaction with life post-remission.
DISCUSSION
61
76 This is the first study to analyse subgroups of natural
178 remitters from alcohol dependence using cluster analysis.
Although results from cluster analysis are highly depen-

Table 3 Demographic characteristics and substance-related variables between LPLS (‘low problems—low support’), HPMS (‘high prob-
lems—medium support’) and LPHS (‘low problems—high support’).

LPLS (n = 65) HPMS (n = 37) LPHS (n = 76) Sign. P

Female (%) 46.2 5.4 18.4 <0.001

Married at interview (%) 44.6 62.2 67.1 0.022
Mean age at interview (SD) 46.7 (9.3) 43.0 (8.8) 53.6 (9.8) 0.001a
<10 years of schooling (%) 57.8 62.2 33.3 0.003
Unemployed prior to remission (%) 41.5 18.9 13.2 0.015
Duration of remission (months) 84.3 (67.3) 91.6 (72.6) 72.6 (59.1) NS
Average daily alcohol consumption (gr.) 316.0 (206.4) 368.1 (219.1) 212.5 (128.1) 0.001
Abstinent recovery (%) 92.3 97.3 92.19 NS
Unsuccessful quit attempts (%) 44.6 51.4 46.1 NS

a
LPHS differed (P < 0.05) from LPLS and HPMS (Scheffé-test).

© 2003 Society for the Study of Addiction to Alcohol and Other Drugs Addiction, 98, 1737–1746
 Types of natural recovery from alcohol dependence 1743

Table 4 Significant differences in triggering mechanisms and maintenance factors between LPLS (‘low problems—low support’), HPMS
(‘high problems—medium support’) and LPHS (‘low problems—high support’).

Variable LPLS (n = 65) HPMS (n = 37) LPHS (n = 76) Sign. P (ANOVA)

Triggering mechanisms
Satisfaction with life domains prior to remission 2.8 (1.4) 2.4 (1.3) 3.2 (1.6) <0.001a
Partnership 2.8 (1.4) 2.4 (1.3) 3.2 (1.6) <0.001a
Family 2.9 (1.5) 2.3 (1.6) 3.8 (1.2) <0.001a
Friends 3.0 (1.4) 3.1 (1.1) 3.9 (1.0) <0.001a
Financial situation 2.8 (1.5) 2.0 (1.2) 3.6 (1.4) <0.001b
Living conditions 3.4 (1.5) 3.7 (1.5) 4.1 (1.3) 0.012c
Relationship of life-events on remission 1.9 (1.5) 2.8 (1.8) 2.5 (1.6) 0.014d
Partnership 1.9 (1.5) 2.8 (1.8) 2.5 (1.6) 0.014d
Financial events 1.5 (1.2) 2.0 (1.3) 1.1 (0.7) 0.001a
Legal events 1.1 (0.6) 1.9 (1.5) 1.3 (0.9) <0.000a
Maintenance factors
Number of individuals informed about previous drinking problems 4.8 (2.0) 6.3 (1.6) 5.2 (2.2) <0.002f
Coping behaviour
Seeking social support 11.6 (15.3) 16.0 (16.3) 8.1 (14.0) 0.030e
Avoidance/distraction 14.4 (16.1) 21.1 (15.5) 12.0 (12.3) 0.028e
Negative thinking 32.9 (24.2) 45.8 (20.1) 53.0 (26.9) 0.055d
Supportive life-domains
Partnership 2.0 (1.8) 2.9 (1.5) 2.6 (1.7) 0.028d
Family 1.8 (1.7) 2.8 (1.4) 2.3 (1.6) 0.009d
Financial situation 0.9 (1.4) 1.5 (1.4) 0.7 (1.3) <0.001e
Living conditions 0.9 (1.4) 1.5 (1.4) 0.7 (1.3) 0.012f
Social support post-remission
Family 42.1 (61.5) 63.7 (81.6) 74.1 (84.7) <0.001g
Friends 37.7 (54.0) 53.9 (68.5) 63.7 (76.1) <0.001g
Others 64.7 (75.8) 67.2 (77.9) 76.2 (83.5) <0.007c
Satisfaction with life 16.6 (4.3) 17.1 (3.5) 18.8 (4.3) 0.005c

a
LPHS differed (P < 0.05) from LPLS and HPMS; bLPHS differed (P < 0.05) from LPLS differed from HPMS; cLPHS differed (P < 0.05) from LPLS; dHPMS differed
(P < 0.05) from LPLS; eHPMS differed (P < 0.05) from LPHS; fHPMS differed (P < 0.05) from LPLSand LPHS; gHPMS and LPHS differed (P < 0.05) from LPLS.

dent upon the chosen method and the variables used to
form the clusters, the groups represented in our data set
have proven to be stable, independently of the chosen
method.
Consistent with our hypothesis, findings show that
models of natural recovery from alcohol dependence need
to distinguish between alcohol-related stressors, on one
hand, and psychosocial resources on the other hand. One
cluster is characterized by many resources with few prob-
lems. This group might be characterized by the term
‘social capital’, especially if we focus on socio-
demographic resources in addition to social support, i.e.
low unemployment rate, high degree of married subjects
and more years of schooling. Within both the other clus-
ters, an interaction between problem severity and social
support is clear: subjects with low support manage to
recover on their own as long as the severity of alcohol-
related problems is low, while subjects with high alcohol-
related problems appear to be in need of a minimum of
social support in order to overcome their dependence.
The data clearly confirm the hypothesis that social
capital is an important enabling factor of processes of nat-
ural recovery, with social capital being characterized by
few social problems and a high degree of social support.
This constellation of variables, however, is found in less
than half of all subjects investigated in the present study.
Therefore, it is not surprising that studies using treated
control groups had been unable so far to identify group
differences according to the construct of social capital.
According to our data, this lack of findings might have
been caused by a reciprocal interaction of resources and
stressors in the remaining subgroups; while subjects with
high alcohol-related problems also revealed high social
support, another group with low social support revealed
a moderate severity of dependence in terms of all subjects
investigated.
Interestingly, although in this study social support
increased in all groups after remission, the relative differ-
ences between all groups remained stable, with subjects
from the medium severity group revealing the lowest level
of social support. This finding strenghtens the assump-
tion that the contribution of psychosocial factors as trig-
gering variables and maintenance factors for unassisted
recoveries differs consistently between groups. However,

© 2003 Society for the Study of Addiction to Alcohol and Other Drugs Addiction, 98, 1737–1746
1744 Gallus Bischof et al.
the relative importance of social resources differs between
groups, with our HPMS group relying specifically on
social support while our LPHS group gave less emphasis
to social factors in order to enable remission. The latter
can be seen in their relatively low extent of seeking social
support as coping behaviour and their ratings of impor-
tant contributing life-domains.
Taken together, the data suggest strongly that a lack of
identifying specific variables of natural recovery in previ-
ous research might be due to heterogeneous subgroups of
natural remitters. In order to build a conceptual frame-
work for understanding processes of natural recovery,
constellations of different factors have to be considered.
We therefore assume that, consistent with various psy-
chological models on health behaviour, such as the the-
ory of salutogenesis (Antonovsky 1987; Antonovsky
1996), both resources and stressors play an important
role for processes of remission without formal help. This
assumption is strengthened by findings of Humphreys,
Moos & Finney (1995) in California, although their study
was not restricted to alcohol-dependent subjects and
included subjects with regular participation in Alcoholics
Anonymous (AA). Their findings revealed that subjects
with lower levels of drinking problems had more social
support available to them at the beginning of their recov-
ery attempt, relied less on AA and became moderate
drinkers more often after recovery.
Analysis of ethnic minority groups in the United King-
dom suggests that the relative impact of variables used to
form the clusters in the present analysis might differ
between ethnic subgroups. In a study comparing ‘spon-
taneous remitters’ of ethnic minorities from the Indian
subcontinent with white indigenous remitters, the ethnic
minority groups revealed a lower problem severity and a
higher impact of social support on their resolution (Cam-
eron et al. 2002). This finding might indicate the percent-
age of natural remitters with high support and low
problems to be even higher if ethnic minorities are taken
into consideration.
Some limitations to our data need to be considered.
The present analysis is restricted to media-recruited
subjects, who are characterized by a high severity of
alcohol dependence compared to representative sam-
ples of natural remitters (Rumpf et al. 2000). Because
severity of dependence is one important factor in differ-
entiating clusters in the present analysis, effects might
differ in representative samples. Due to the retrospec-
tive nature of the present study, a possible recall bias
inherent in this approach has to be considered. There-
fore, future research on natural recovery should focus
on representative samples of alcohol-dependent sub-
jects that should be analysed prospectively in order to
determine the relative impact of resources and stressors
in various subgroups.
© 2003 Society for the Study of Addiction to Alcohol and Other Drugs
ACKNOWLEDGEMENTS
This study is part of the German research network Ana-
lytical Epidemiology of Substance Abuse (ANEPSA
Research Group 1998) and was supported by the German
Federal Ministry of Education, Science, Research and
Technology grant no. 01 EB 9406. The authors would
like to thank Dipl. Stat. Bernd-Wolfgang Igl, Institute for
Medical Biometry and Statistics, University of Lübeck, for
statistical support and Dipl. Psych. Janina Grothues,
Department of Psychiatry and Psychotherapy, University
of Lübeck, for manuscript editing.
REFERENCES
American Psychiatric Association (1995) Diagnostic and Statis-
tical Manual of Mental Disorders, 4th edn, International ver-
sion. Washington, DC: American Psychiatric Association.
ANEPSA Research Group (1998) German Research Network
‘Analytical Epidemiology of Substance Abuse’ (ANEPSA).
European Addiction Research, 4, 203–204.
Antonovsky, A. (1987) Unraveling the Mystery of Health. San
Francisco.
Antonovsky, A. (1996) The salutogenetic model as theory to
guide health promotion. Health Promotion International, 11,
11–18.
Berwick, D. M., Murphy, J. M., Goldman, P. A., Ware, J. E.,
Barsky, A. J. & Weinstein, M. C. (1991) Performance of a
five-item mental health screening test. Medical Care, 29, 169–
176.
Bischof, G., Rumpf, H. J., Hapke, U., Meyer, C. & John, U. (2000)
Maintenance factors of recovery from alcohol dependence in
treated and untreated individuals. Alcoholism: Clinical and
Experimental Research, 24, 1773–1777.
Bischof, G., Rumpf, H. J., Hapke, U., Meyer, C. & John, U. (2001)
Factors influencing remission without formal help from alco-
hol dependence in a representative population sample. Addic-
tion, 96, 1327–1336.
Bischof, G., Rumpf, H. J., Hapke, U., Meyer, C. & John, U. (2002)
Natural recovery from alcohol dependence: how restrictive
should our definition of treatment be ? Journal of Studies on
Alcohol, 63, 229–236.
Bischof, G., Rumpf, H. J., Hapke, U., Meyer, C. & John, U. (2003a)
Skala zur Erfassung sozialen Drucks (SSD) [Social Pressure
Scale, German version]. In: Glöckner-Rist, A., Rist, F. &
Küfner, H., eds. Elektronisches Handbuch Zu Erhebungsinstru-
menten Im Suchtbereich (EHES) 3.00. [Electronic Handbook
on Assessment Instruments in the Addiction Field]. Mannheim:
Zentrum für Umfragen, Methoden und Analysen. Avail-
able online: http://www.psy.uni-muenster.de/institut1/ehes/
startseite.htm.
Bischof, G., Rumpf, H. J., Hapke, U., Meyer, C. & John, U. (2003b)
Deutsches Coping-Behaviours Inventory (CBI-G) [German
Coping Behaviors Inventory]. In: Glöckner-Rist, A., Rist, F. &
Küfner, H., eds. Elektronisches Handbuch Zu Erhebungsinstru-
menten Im Suchtbereich (EHES) 3.00. [Electronic Handbook
on Assessment Instruments in the Addiction Field]. Mannheim:
Zentrum für Umfragen, Methoden und Analysen. Avail-
able online: http://www.psy.uni-muenster.de/institut1/ehes/
startseite.htm.
Addiction, 98, 1737–1746

Blomquist, J. (1999) Treated and untreated recovery from alco-
hol misuse: environmental influences and perceived reasons
for change. Substance Use and Misuse, 34, 1371–1406.
Booth, B. M., Russel, D. W., Soucek, S. & Laughilin, P. R. (1992)
Social support and outcome of alcoholism treatment: an
exploratory analysis. American Journal of Drug and Alcohol
Abuse, 18, 87–101.
Bott, K. E., Rumpf, H. J., Bischof, G., Meyer, C., Hannöver, W.,
Hapke, U. & John, U. (2003) Alkoholabstinenz-Selbstwirk-
samkeitsfragebogen; deutsche version (AASE-G): Deutsche
Version der Alcohol Abstinence Self-Efficacy (AASE) Scale
[Alcohol Abstinence Self-Efficacy-Scale, German version]. In:
Glöckner-Rist, A., Rist, F. & Küfner, H., eds. Elektronisches
Handbuch Zu Erhebungsinstrumenten Im Suchtbereich (EHES)
3.00. [Electronic Handbook on Assessment Instruments in the
Addiction Field]. Mannheim: Zentrum für Umfragen, Meth-
oden und Analysen. Available online: http://www.psy.uni-
muenster.de/institut1/ehes/startseite.htm.
British Medical Association (1995) Guidelines on Sensible Drink-
ing. London: British Medical Association.
Cameron, D., Manik, G., Bird, R. & Sinorwalia, A. (2002) What
may we be learning from so-called spontaneous remission in
ethnic minorities ? Addiction Research and Theory, 10, 175–
182.
Cloninger, C. R., Bohman, M. & Sigvardsson, S. (1981) Inherit-
ance of alcohol abuse: cross-fostering analysis of adopted men.
Archives of General Psychiatry, 38, 861–868.
Cunningham, J. A. (1999) Resolving alcohol-related problems
with and without treatment: the effects of different problem
criteria. Journal of Studies on Alcohol, 60, 463–466.
Davidson, R. & Raistrick, D. (1986) The validity of the Short
Alcohol Dependence Data (SADD) questionnaire: a short self-
report questionnaire for the assessment of alcohol depen-
dence. British Journal of Addiction, 81, 217–222.
DiClemente, C. C., Carbonari, J. P., Montgomery, R. P. G. &
Hughes, S. O. (1994) The Alcohol Abstinence Self-Efficacy
Scale. Journal of Studies on Alcohol, 55, 141–148.
Diener, E., Emmons, R. A., Larsen, R. J. & Griffin, S. (1985) The
Satisfaction With Life Scale. Journal of Personality Assessment,
49, 71–75.
Dlugosch, G. E. & Krieger, W. (1995) Fragebogen Zur Erfassung
Des Gesundheitsverhaltens (FEG) [Questionnaire to Assess Health
Behaviour]. Göttingen: Hogrefe.
Finney, W. & Moos, R. (1995) Entering treatment for alcohol
abuse: a stress and coping model. Addiction, 90, 1223–1240.
Granfield, R. & Cloud, W. (1996) The elephant that no one sees:
natural recovery among middle-class addicts. Journal of Drug
Issues, 26, 45–61.
Hasin, D. S. (1994) Treatment/self-help for alcohol-related prob-
lems: Relationship to social pressure and alcohol dependence.
Journal of Studies on Alcohol, 55, 660–666.
Hingson, R., Mangione, T., Meyers, A. & Scotch, N. (1982) Seek-
ing help for drinking problems. Journal of Studies on Alcohol,
43, 273–288.
Humphreys, K., Moos, R. H. & Finney, J. W. (1995) Two path-
ways out of drinking problems without professional treat-
ment. Addictive Behaviors, 20, 427–441.
John, U., Rumpf, H. J. & Hapke, U. (2003) A new measure of
the Alcohol Dependence Syndrome: the SEverity Scale of
Alcohol dependence (SESA). European Addiction Research, 9,
87–93.
Klingemann, H. K. (1991) The motivation for change from prob-
lem alcohol and heroin use. British Journal of Addiction, 86,
727–744.
© 2003 Society for the Study of Addiction to Alcohol and Other Drugs
Types of natural recovery from alcohol dependence 1745
Klingemann, H., Sobell, L., Barker, J., Blomquist, J., Cloud, W.,
Ellinstead, T., Finfgeld, D., Granfield, R., Hodgings, D., Hunt,
G., Junker, C., Moggi, F., Peele, S., Smart, R., Sobell, M. &
Tucker, J. (2001) Promoting self-change from problem sub-
stance use: practical implications for policy, prevention and
treatment. Dordrecht: Kluwer Academic Publishers.
Laireiter, A. R. (1996) Skalen Sozialer Unterstützung [Scales of
Social Support]. Mödling: Dr G. Schuhfried GmbH.
Litman, G. K., Stapelton, J., Oppenheim, A. M. & Peleg, M.
(1983) An instrument for measuring coping behaviours in
hospitalized alcoholics: implications for relapse prevention
treatment. British Journal of Addiction, 78, 269–276.
Lohaus, A. & Schmitt, G. M. (1989) Fragebogen Zur Erhebung Von
Kontrollüberzeugungen Zu Krankheit und Gesundheit (KKG),
[Questionnaire to Assess Locus of Control with Respect to Illness
and Health]. Göttingen: Hogrefe.
Lundberg, O. & Nyström Peck, M. (1995) A simplified way of
measuring sense of coherence. European Journal of Public
Health, 5, 56–59.
Michael, A., Rumpf, H.-J., Meyer, C., Hapke, U., Bischof, G. &
John, U. (2003) Gründe für Nichtinanspruchnahme suchts-
pezifischer Hilfen (GNSH) [Reasons for not seeking help for
substance-related disorders]. In: Glöckner-Rist, A., Rist, F. &
Küfner, H., eds. Elektronisches Handbuch Zu Erhebungsinstru-
menten Im Suchtbereich (EHES) 3.00. [Electronic Handbook
on Assessment Instruments in the Addiction Field]. Mannheim:
Zentrum für Umfragen, Methoden und Analysen. Avail-
able online: http://www.psy.uni-muenster.de/institut1/ehes/
startseite.htm.
Miller, W. R., Westerberg, V. S., Harris, R. J. & Tonigan, J. S.
(1996) What predicts relapse? Prospective testing of anteced-
ent models. Addiction, 91, 155–172.
Milligan, G. W. & Cooper, M. C. (1987) Methodology review:
clustering methods. Applied Psychological Measures, 11, 329–
354.
Moos, R. H., Cronkite, R. C., Billings, A. G. & Finney, J. W. (1985)
Health and Daily Living Form Manual. Stanford University Med-
ical Centers. Stanford, USA.
Norusis, M. (1993) SPSS for Windows: Base User’s Guide: Release
6 0. Chicago: SPSS, Inc.
Rumpf, H. J., Bischof, G., Hapke, U., Meyer, C. & John, U. (2000)
Studies on natural recovery from alcohol dependence: sample
selection bias by media solicitaion. Addiction, 95, 765–775.
Rumpf, H.-J., Meyer, C., Hapke, U., Bischof, G. & John, U. (2000)
Inanspruchnahme suchtspezifischer Hilfen von Alkoholab-
hängigen und-mißbrauchern: Ergebnisse der TACOS Bev-
ölkerungsstudie [Utilization of professional help of individuals
with alcohol dependence or abuse: findings from the TACOS
population study]. Sucht, 46, 9–17.
Skinner, H. A. (1982) Statistical approaches to the classification
of alcohol and drug addiction. British Journal of Addiction, 77,
259–273.
Skinner, H. A. & Allen, B. A. (1982) Alcohol dependence syn-
drome: measurement and validation. Journal of Abnormal Psy-
chology, 91, 199–209.
Snow, M. G., Prochaska, J. O. & Rossi, J. S. (1994) Processes of
change in Alcoholics Anonymous: maintenance factors in
long-term sobriety. Journal of Studies on Alcohol, 55, 362–371.
Sobell, L. C., Cunningham, J. A., Sobell, M. & B. (1996) Recovery
from alcohol problems with and without treatment: preva-
lence in two population surveys. American Journal of Public
Health, 7, 966–972.
Sobell, L. C., Sobell, M. B. & Toneatto, T. (1992) Recovery from
alcohol problems without treatment. In: Greeley, J., ed. Self
Addiction, 98, 1737–1746
1746 Gallus Bischof et al.
Control and the Addictive Behaviours, pp. 198–242. New York:
Macmillan Publishing.
Stall, R. (1983) An examination of spontaneous remission from
problem drinking in the bluegrass region of Kentucky. Journal
of Drug Issues, 13, 191–206.
Stockwell, H. A., Murphy, D. & Hodgson, R. (1983) The severity
of alcohol dependence questionnaire: its use, reliability and
validity. British Journal of Addiction, 78, 145–155.
Tuchfeld, B. S. (1981) Spontaneous remission in alcoholics:
empirical observations and theoretical implications. Journal of
Studies on Alcohol, 42, 626–641.
Tucker, J. A. & Gladsjo, J. A. (1993) Help-seeking and recovery
by problem drinkers: characteristics of drinkers who attended
Alcoholics Anonymous or formal treatment or who recovered
without treatment. Addictive Behaviors, 18, 529–542.
Tucker, J. A., Vuchinich, R. E. & Gladsjo, J. A. (1994) Envi-
ronmental events surrounding natural recovery from alco-
© 2003 Society for the Study of Addiction to Alcohol and Other Drugs
hol-related problems. Journal of Studies on Alcohol, 55, 401–
411.
Tucker, J. A., Vuchinich, R. E. & Pukish, M. M. (1995)
Molar environmental contexts surrounding recovery from
alcohol problems by treated and untreated problem drink-
ers. Experimental and Clinical Psychopharmacology, 3, 195–
204.
Vaillant, G. E. (1982) Natural history of male alcoholism IV:
paths to recovery. Archives of General Psychiatry, 39, 127–
133.
Vaux, A., Phillips, J., Holly, L., Thomson, B., Williams, D. & Stew-
art, D. (1986) The Social Support Appraisal (SS-A) Scale:
studies of reliability and validity. American Journal of Commu-
nity Psychology, 14, 195–219.
Wallston, K. A. & Wallston, B. S. (1978) Development of the Mul-
tidimensional Health Locus of Control (MHLC) scales. Health
Education Monographs, 6, 160–170.
Addiction, 98, 1737–1746