· alcohol consumption at pre-treatment, neuroticism (echeburua et al. 2001), and gambling related...
TRANSCRIPT
1 23
Journal of Gambling Studies e-ISSN 1573-3602 J Gambl StudDOI 10.1007/s10899-013-9408-3
Predictors of Relapse in Problem Gambling:A Prospective Cohort Study
David P. Smith, Malcolm W. Battersby,Rene G. Pols, Peter W. Harvey, JaneE. Oakes & Michael F. Baigent
1 23
Your article is protected by copyright and all
rights are held exclusively by Springer Science
+Business Media New York. This e-offprint is
for personal use only and shall not be self-
archived in electronic repositories. If you wish
to self-archive your article, please use the
accepted manuscript version for posting on
your own website. You may further deposit
the accepted manuscript version in any
repository, provided it is only made publicly
available 12 months after official publication
or later and provided acknowledgement is
given to the original source of publication
and a link is inserted to the published article
on Springer's website. The link must be
accompanied by the following text: "The final
publication is available at link.springer.com”.
ORI GIN AL PA PER
Predictors of Relapse in Problem Gambling:A Prospective Cohort Study
David P. Smith • Malcolm W. Battersby • Rene G. Pols •
Peter W. Harvey • Jane E. Oakes • Michael F. Baigent
� Springer Science+Business Media New York 2013
Abstract To explore the variation of predictors of relapse in treatment and support
seeking gamblers. A prospective cohort study with 158 treatment and support seeking
problem gamblers in South Australia. Key measures were selected using a consensus
process with international experts in problem gambling and related addictions. The out-
come measures were Victorian Gambling Screen (VGS) and behaviours related to gam-
bling. Potential predictors were gambling related cognitions and urge, emotional
disturbance, social support, sensation seeking traits, and levels of work and social func-
tioning. Mean age of participants was 44 years (SD = 12.92 years) and 85 (54 %) were
male. Median time for participants enrolment in the study was 8.38 months
(IQR = 2.57 months). Patterns of completed measures for points in time included 116
(73.4 %) with at least a 3 month follow-up. Using generalised mixed-effects regression
models we found gambling related urge was significantly associated with relapse in
problem gambling as measured by VGS (OR 1.29; 95 % CI 1.12–1.49) and gambling
behaviours (OR 1.16; 95 % CI 1.06–1.27). Gambling related cognitions were also sig-
nificantly associated with VGS (OR 1.06; 95 % CI 1.01–1.12). There is consistent asso-
ciation between urge to gamble and relapse in problem gambling but estimates for other
D. P. Smith (&) � M. W. Battersby � R. G. Pols � P. W. Harvey � J. E. Oakes � M. F. BaigentFlinders Human Behaviour and Health Research Unit, Department of Psychiatry, Flinders University,GPO Box 2100, Adelaide, SA 2001, Australiae-mail: [email protected]
M. W. Battersbye-mail: [email protected]
R. G. Polse-mail: [email protected]
P. W. Harveye-mail: [email protected]
J. E. Oakese-mail: [email protected]
M. F. Baigente-mail: [email protected]
123
J Gambl StudDOI 10.1007/s10899-013-9408-3
Author's personal copy
potential predictors may have been attenuated because of methodological limitations. This
study also highlighted the challenges presented from a cohort study of treatment and
support seeking problem gamblers.
Keywords Problem gambling � Relapse � Predictors � Cohort study
Introduction
Problem gambling is a serious public health concern at an international level (e.g. Ger-
many, Holland, Spain, Switzerland, Australia, United States, Britain and Hong Kong) and
occurs more frequently in younger populations (Becona 1996; Bondolfi et al. 2000; Del-
fabbro 2009; Shaffer and Hall 2001; Wardle et al. 2007; Wong and Ernest 2003). Co-
morbid mental disorders such as depression and anxiety are common in both treatment and
non-treatment seeking problem gamblers (Lorains et al. 2011).
The availability of treatments for gambling disorders has increased in the past 10 years
and are similar to those in other addictions, for example brief interventions (Hodgins
2009), psychotherapies (Gooding and Tarrier 2009), and peer-support (Oei and Gordon
2008). However, a discernible threat to treatment efficacies exist due to high rates of
relapse (Hodgins et al. 2007; Hodgins and el-Guebaly 2004) as in other addictive
behaviours and psychological disorders [e.g. alcohol use, anxiety disorder, and obsessive–
compulsive disorder (Brandon et al. 2007)].
In the effort to identify predictors of relapse, a range of outcome measures have been
used to quantify the behaviour. In two studies conducted in The Netherlands, relapsers
were classified as either those that responded in the affirmative to ‘‘Do you think that you
have a gambling problem again?’’ (Goudriaan et al. 2007) or as ‘‘…the presence of any
gambling behaviour’’ in a one year follow-up time period (Wilde et al. 2013). In a United
States study that examined factors associated with length of gambling abstinence attempt,
relapsers were defined as either ‘‘delayed’’ if they were ‘‘current PG (pathological gam-
bling) and had at least one sustained period of gambling abstinence C3 months, or as
‘‘immediate’’ according to ‘‘PG’s who had never remained abstinent for a period
[2 weeks’’ (Daughters et al. 2005). In a Canadian study on retrospective and prospective
precipitants to relapse, relapse was defined as ‘‘gambling after 2 weeks of abstinence’’
where all participants in the study had a goal of abstinence (Hodgins and el-Guebaly 2004).
In Spain, an investigation of relapse following behavioural treatment for slot-machine
gambling defined relapsers as those that had ‘‘more than two isolated episodes of gambling
in the 12 months follow-up or a total expense higher than a week of gambling before the
treatment’’ (Echeburua et al. 2001). These studies highlight the disparity in operational-
izing relapse and the challenge to homogenizing a gambling related taxonomy at a
transnational level.
On the flip side of the relapse equation only a few studies have examined risk factors
that may precipitate relapse and these are mostly limited to one or two predictor variables
(Ledgerwood and Petry 2006). A broad range of precipitants to relapse in gambling have
been proposed from psychological, psychobiological, social, and environmental domains
(Ledgerwood and Petry 2006). The studies that have investigated specific factors include
neurocognitive indicators of disinhibition and decision making (Goudriaan et al. 2007),
affective states and stress reactivity (Daughters et al. 2005), satisfaction with treatment,
J Gambl Stud
123
Author's personal copy
alcohol consumption at pre-treatment, neuroticism (Echeburua et al. 2001), and gambling
related cognitions and urge (Hodgins and el-Guebaly 2004; Oei and Gordon 2008). The
findings are generally mixed and associations have been modest or null.
In substance use disorders there is an established evidence-base to conceptualizing and
preventing relapse (Brandon et al. 2007). A significant influence in this field has been
Marlatt’s cognitive-behavioural model of relapse (Marlatt and Gordon 1985). This model
has underpinned the development of relapse prevention (RP) interventions in addiction
treatment centers across the world including those for problem gambling (Brandon et al.
2007; Donovan and Witkiewitz 2012). Since inception, it has evolved from a linear con-
figuration to a non-linear dynamic system where the complex interplay between multiple
risk factors is integral to predicting a relapse event (Witkiewitz and Marlatt 2004). The risk
factors are clustered as distal or stable factors and proximal or time-varying factors.
Examples of distal factors in alcohol use disorder include comorbid conditions and severity
of alcoholism. Proximal factors can involve psychological disturbance such as stress,
rapidly shrinking social support, and craving (Donovan 1996). It has been proposed that
distal factors may predict those individuals that are vulnerable to relapse whilst proximal
factors may account for more temporal characteristics of a relapse event (Shiffman 1989).
In problem gambling, data are lacking to clarify the association between potential
predictors and relapse and if associations are consistent across time. A better understanding
of distal and proximal factors associated with relapse in treatment seeking problem
gamblers will help predict those vulnerable to relapse, improve relapse prevention strat-
egies, and reduce the risk of relapse.
In this study, we aimed to identify potential predictors of relapse using an expert
consensus process and then based on these findings to address the research question:
among a population of treatment and support seeking problem gamblers what are the
demographic, behavioural, clinical and social characteristics that may consistently predict
relapse in problem gambling across time
Methods
The study involved a two-staged process to identify predictors of relapse in problem
gamblers. Firstly, a Delphi inquiry was used to obtain expert consensus on definitions,
potential elements and predictors of relapse. This process is briefly described below with
full details available at http://www.gamblingresearch.org.au/home/research/ (Battersby
et al. 2010). Secondly, these findings were used to empirically test potential predictors
using a prospective cohort design and comprise the main focus of this paper.
Delphi
Due to a paucity of research literature into relapse in problem gambling we utilised a
Delphi study to identify and quantitatively measure uncertainty on definitions, elements,
and predictors of relapse. The Delphi is an iterative process that provides the opportunity
for individual experts to change their opinions based on feedback of summary measures
from preceding rounds (Mullen 2003). In this study the Delphi was conducted in four
rounds. Round one involved a project advisory group comprising six expert clinicians (4
psychiatrists, 1 social worker and 1 psychologist) with at least 15 years clinical experience
each in problem gambling and substance use disorders. Of these, four were widely pub-
lished international researchers in problem gambling and related disorders. Tasks included
J Gambl Stud
123
Author's personal copy
creating definitions of relapse in problem gambling and listing potential elements and
predictors of relapse. A list of international experts in gambling disorders and related
addictions was also generated by the advisory group to invite for participation in sub-
sequent rounds.
Round two comprised the entire list of elements and predictors and alternative defini-
tions of relapse sent to potential participants for rating. There was a response rate of 46 %
with twenty two participants returning completed questionnaires. In round three a revised
questionnaire was developed to include median scores from the preceding round and sent
to participants who were asked to once again rate each item in light of the new information.
The principal investigator then convened a meeting of an expert panel of six international
researchers who attended the Auckland 2008 International Think Tank on Gambling at the
Auckland University of Technology to discuss findings from the Delphi iterations.
Final predictors selected for the observational study were required to be measurable in
terms of validated self-rating scales and parsimony where there was item overlap within
domains of psychological, psychobiological, and social and environmental factors. Each
measure is described in the following section.
Cohort
Setting and Participants
The study took place in Adelaide, South Australia from March 2008 to March 2009.
Adelaide has a population of 1.2 million and is the capital city of South Australia (total
population 1.6 million) where problem gambling is mainly a result of the widespread
availability of 12,688 live gaming machines in venues in nearly all towns and cities across
the state (Government of South Australia: Consumer and Business Services 2012).
Participants were adults who, at baseline, were either seeking or engaged in treatment or
support for problem gambling. Recruitment was initiated by research staff contacting
gambling help services in South Australia with information about the study. All services
contacted agreed to participate. Participant numbers recruited by agency were influenced
by factors that included service size and duration of agency engagement with the study.
Participating services were the Statewide Gambling Therapy Service (SGTS) that
provides individual cognitive behavioural therapy in outpatient settings; Pokies Anony-
mous (PA), a self-help peer support organisation based on similar principles to Alcoholics
Anonymous; Gambling Helpline (GHL), a free 24-h counselling, information and referral
service; Offenders Aid and Rehabilitation Service (OARS), a support organisation for
people affected by problem gambling who have been, or are at risk of entering the criminal
justice system in South Australia; and Relationships Australia South Australia (RASA)
which provides specialised counselling services for problem gamblers.
The study received approval from the Southern Adelaide Health Service/Flinders
University Human Research Ethics Committee. Participants were given an information
statement regarding the study and asked to provide written informed consent before data
collection began.
Design and Procedure
A prospective cohort design was used to investigate predictors of relapse in problem
gamblers, which followed participants over a 6–12 month time period. Participants were
recruited in the period March to September 2008. Baseline measures were collected
J Gambl Stud
123
Author's personal copy
following consent to participate in the study. Follow-up time points were one and 3 months
for all participants. Depending on the date of recruitment participants either completed a
6 month measure as their final follow-up, or a 6 month and a final 12 month measure by
study completion in March 2009.
To improve follow-up response rates, self-report questionnaires were administered by
mail to reduce burden of visits to the study site (National Research Council 2010). Mail-
outs included an unconditional gift voucher of $10 with 3, 6, and 12 month questionnaires,
and a gift voucher of $20 on return of the final 6 or 12 month questionnaire (Edwards et al.
2002). A follow-up letter was sent to participants 1 week after mailing out of each
questionnaire. The purpose of this letter was to thank those who had returned their
questionnaires and act as a reminder for those who had not. At 3 weeks a further letter was
sent out to participants with unreturned questionnaires. A final follow-up occurred at
5 weeks with a telephone call for unreturned questionnaires to offer the mailing out of a
further set of questionnaires if required (Dillman 2007).
Outcome Measures
Outcome status was represented by three unordered items (remission, relapse, and
continuing to problem gamble). All participants were measured to be either in remission
(non-problem gambling), or continuing to problem gamble at baseline, the vast majority
being in the continuing category as they were treatment and support seeking. To assess
predictors of relapse, participants needed to achieve at least some period of remission
during the study so subsequent relapse predictors could be observed and measured. At
follow-up time points of 1, 3, 6, and 12 months, outcome status of participants were
classified as either in remission, continuing to problem gamble, or relapsed if the individual
had returned to problem gambling following a remission period. To compare alternative
measures of outcome derived from the Delphi process, assessments were conducted at each
point using two approaches; Victorian Gambling Screen (VGS) self-harm subscale and
self-reported gambling behaviours as quantified by the Delphi definition of relapse.
For the purpose of this study, selection of the VGS was based on the scales following
properties: (i) items on the self-harm subscale relate to the person’s experiences in the
previous 4 weeks, and therefore an enhanced sensitivity to relapse and temporal associa-
tions, (ii) representation of all domains of elements in the final Delphi list (behavioural,
cognitive, and interpersonal factors), and (iii) a validated cut score indicative of problem
gambling. Of other validated instruments that classify problem or probable pathological
gambling, neither the Canadian Problem Gambling Index (CPGI) (Ferris and Wynne 2001)
or the South Oaks Gambling Screen (SOGS) (Lesieur and Blume 1987) satisfied both
criteria (i) and (ii).
The VGS is a self-reported questionnaire measuring the extent to which a gambler’s
behaviour has caused harm to themselves, their family or the community in the previous
4 weeks. The harm to self sub-scale has been validated for use in Australia in community
and clinical populations (Tolchard and Battersby 2010) where a cut score of 21 or higher is
indicative of problem gambling. Concurrent validity indicates the scale correlates very
highly with the SOGS (r = 0.97), but extends the score range. The VGS has also shown
similar properties in construct validity as the CPGI on a number of problem gambling
correlates (e.g. ‘self-rating of problem’; ‘wanted help’; and ‘suicidal tendencies’)
(McMillen and Wenzel 2006).
J Gambl Stud
123
Author's personal copy
For gambling behaviours participants self-reported their gambling activities during the
previous 4 weeks and were classified according to their aims of abstinence or controlled
gambling. Three questions pertaining to outcome status were asked:
(i) Are you currently aiming to be abstinent from gambling? (Yes/No);
When answering ‘‘No’’ the person was asked:
(ii) To what level are you aiming to limit your gambling? (Once a week/Twice a week/
Three or more times a week/daily).
(iii) How often did you gamble during the last 4 weeks? (Never/Once/Twice/Once a
week/Twice a week/Three or more times a week).
Participants who answered ‘‘Yes’’ to question (i) and reported gambling activity in that
time period were categorised as having either relapsed or continuing to problem gamble
(depending on their previous assessment). Participants answering ‘‘No’’ to question (i) and
reporting gambling activity in question (iii) that exceeded their aims in question (ii) were
also categorised as either having relapsed or continuing to problem gamble. Otherwise
participants were categorised as in remission.
Predictors
A range of self-report measures were selected a priori on the basis of Delphi findings for
potential predictor variables and were administered at all study time points. To assess
internal consistency of each measure completed by participants at baseline, Cronbach alpha
(a) coefficients were calculated and are reported in the following section.
The experience of depression, state anxiety and stress was measured using the 21 item
self-report Depression Anxiety Stress Scale (DASS-21) that has been validated against
other depression and anxiety inventories (a = 0.96) (Lovibond and Lovibond 1995).
Levels of trait anxiety were recorded using the 20 item self-report Trait Anxiety Inventory
(TAI) Y-20 self-report measure shown to have good reliability measured by test–retest
coefficients and sound validity (a = 0.92) (Spielberger et al. 1983). Gambling Urge Scale
(GUS) is a 6 item self-report questionnaire measuring the extent of gambling urge and has
been validated in community, university and clinical samples of gamblers (a = 0.93)
(Raylu and Oei 2004b; Ashrafioun et al. 2013; Smith et al. 2012). Gambling Related
Cognition Scale (GRCS) is a 23 item self-report questionnaire that records common
thoughts associated with PG and a comparison with the South Oakes Gambling screen
indicated the scale has good psycho-metric properties (a = 0.91) (Raylu and Oei 2004a).
Alcohol Use Disorders Identification Test (AUDIT) is a non-diagnostic ten item ques-
tionnaire indicating hazardous alcohol and psycho-metric properties include superior
specificity and sensitivity to those of other self-report screening measures and good test–
retest reliability and internal consistency (a = 0.91) (Reinert and Allen 2002). Arnett
Inventory of Sensation Seeking (AISS) is a 20 item self-report questionnaire that measures
sensation seeking personality traits and has been shown to be free from social desirability
bias (a = 0.69) (Roth 2003). Multidimensional Scale of Perceived Social Support
(MSPSS) is a 12 item self-report questionnaire containing three sub-scales (significant
other, family and friends sub-scales) and is psychometrically sound, with good reliability,
factorial validity, and adequate construct validity (a = 0.94) (Zimet et al. 1988). Work and
Social Adjustment Scale is a self-report questionnaire used to measure a patient’s per-
spective of their functional ability/impairment on a seven-point scale (a = 0.84) (Mundt
et al. 2002).
J Gambl Stud
123
Author's personal copy
At baseline participant’s reported their gender, age, marital status, highest education
level, employment status, and living arrangement. Self-reported data for duration of
gambling problem and primary form of PG was also collected.
Baseline Characteristics
Demographic and clinical characteristics of the study cohort are presented in Table 1. When
compared to previous normal population scores baseline DASS-21 means where higher for
depression, anxiety, and stress scales, and in the moderate severity range (Lovibond and
Lovibond 1995). Mean scores on the TAI were higher than previous normative data in
working adults, college and high school students by at least one standard deviation (Spiel-
berger et al. 1983). Stratifying the VGS self- harm subscale at cut score 21 found 94.9 %
(n = 150) participants were classified as problem gamblers at study commencement.
Statistical Methods
Statistical analyses were conducted using Stata 10.0 (StataCorp. 2008). Baseline character-
istics were compared between participants completing baseline measures only and partici-
pants with at least one follow-up measure using t tests for continuous variables, and v2 tests of
independence for categorical variables. A significance level of 5 % was used. For statistical
modelling of predictors of relapse we used mixed-effects multinomial logistic regression to
handle correlated observations from repeated measures within subjects, and the unordered or
nominal response categories of remission, relapse, and continuing to problem gamble (Long
and Freese 2006). Mixed-effects models do not assume equal time intervals for all partici-
pants, and therefore incomplete data is not excluded. Missing data was assumed to be missing
at random (MAR) i.e. probability of missing was independent of the unobserved data and
conditional on observed data. Models were fitted using the user-written program gllamm
(generalised linear latent and mixed models) (Skrondal and Rabe-Hesketh 2003).
Variable selection for regression models commenced with univariate analyses and were
selected for model advancement based on p \ .25 (Hosmer and Lemeshow 2000). Pairwise
correlations among the significant variables were assessed for collinearity. An initial full
model was created with variables significant at p \ .25 and not collinear. Using backward
manual elimination methods, variables with the least significant Wald statistic were
removed from the model. A comparison of log likelihood values between the fitted model
and the full model was conducted for each variable removed. The goal was to construct a
model with the fewest number of variables without compromising a good fit of the data. To
interpret effect sizes, odds ratios were calculated to represent the probability of experi-
encing one outcome category (relapse or continuing to gamble) over the probability of
experiencing the reference category of remission.
Results
Participant Flow
Overall, 352 treatment and support seeking problem gamblers were potentially eligible to
participate in this study. Reasons for study exclusion were unstable mental state (n = 50)
such as mania or suicidality following clinician assessment that indicated that the problem
J Gambl Stud
123
Author's personal copy
Table 1 Demographic and clin-ical characteristics of 158 prob-lem gamblers
Data are presented as number(percentage) unless otherwiseindicated. Percentages not alwaysbased on 158 participants owingto missing data
GUS Gambling Urge Scale,WSAS Work and SocialAdjustment Scale, DASS-21Depression Anxiety and StressScale, MSPSS MultidimensionalScale of Perceived SocialSupport, VGS VictorianGambling Screen, AUDITAlcohol Use DisordersIdentification Test, AISS ArnettInventory of Sensation Seeking,GRCS Gambling RelatedCognition Scale, TAI TraitAnxiety Inventorya Refers to scores collected atbaseline
Variable Value
Age, mean ± SD, year 44 ± 12.92
Male sex 85 (54)
Marital status
Married/defacto 63 (40)
Single 55 (35)
Separated 33 (21)
Widowed 7 (4)
Highest education level
Primary school 2 (1)
High school 88 (56)
TAFE/trade qualification 43 (28)
University degree 24 (15)
Employment
Full-time 58 (37)
Part-time 28 (18)
Not working 50 (31)
Retired 17 (11)
Student 5 (3)
Living arrangement
Alone 36 (24)
Couple with dependent children 29 (19)
Couple without dependent children 38 (25)
Single parent 15 (10)
Living with parents 24 (16)
Sharing 11 (7)
Primary form of gambling
Gaming machines 138 (87)
Duration of gambling problem
\2 year 31 (20)
2–5 year 36 (23)
[5 year 87 (57)
Clinical measures, mean (SD)a
GUS 14.16 (11.52)
WSAS 16.03 (9.73)
DASS-21: depression 10.76 (6.35)
DASS-21: anxiety 6.63 (5.65)
DASS-21: stress 10.60 (6.01)
MSPSS 48.88 (22.08)
VGS: self-harm subscale 40.09 (11.46)
AUDIT 6.07 (7.46)
AISS 46.72 (7.65)
GRCS 65.50 (25.03)
TAI 53.86 (11.01)
J Gambl Stud
123
Author's personal copy
gambler would not be able to participate fully in the study, confidentiality concerns
(n = 15), limited English (n = 7), and no fixed postal address (n = 3). Reasons for non-
participation included refusal (n = 68), inconsistent application of research protocol
(n = 25), and no reason provided (n = 26). Of the 158 consented participants at baseline,
106 (67.1 %) were referred from SGTS, 31 (19.6 %) from GHL, 11 (7 %) from PA, 5
(3.2 %) from RA, and 5 (3.2 %) from OARS. Overall, median time for participants
enrolment in the study was 8.38 months with 50 % of participants having times between 7
and 9.57 months (IQR = 2.57 months) and 25 % less than 7 months. Patterns of com-
pleted measures for points in time included 116 (73.4 %) with at least a 3 month follow-
up, and 99 (62.7 %) with at least a 6 month follow-up.
Overall, there were 34 baseline assessment completers only (21.5 %), with 124 par-
ticipants (78.5 %) completing at least one follow-up measure. There were significantly
more males with baseline scores only (28 males = 82.4 %) compared to females (6
females = 17.6 %), (v12 = 14.21; p \ .01). There were also significant differences on
mean age between baseline completers only and those that had at least one follow-up
assessment (38.5 ± 11.42 vs. 45.51 ± 12.94, p \ .01), AISS (49.37 ± 7.50 vs.
45.99 ± 7.56, p = .02), and TAI (50.62 ± 10.26 vs. 54.75 ± 11.08, p = .05). There were
no significant differences on remaining measures.
Regression Models
The observed sample sizes and response proportions by outcome category of remission,
relapse, or continuing to problem gamble for outcome measures VGS and gambling
behaviours are presented in Table 2. These observed proportions indicate an increase in
remission and relapse rates across time, and a corresponding decrease in the proportion of
participants continuing to gamble. Figure 1 shows change in proportions across time for
VGS. Results of the univariate mixed-effects multinomial regression analysis describing
each variable’s association with gambling status are presented in Table 3.
Results of the final regression models are shown in Table 4. For GUS scores the odds of
participants experiencing a relapse over remission for each increase of one unit increased
by 29 % while holding all other variables constant. The odds of participants continuing to
Table 2 Gambling status as measured with the Victorian Gambling Screen (VGS) and gambling behav-iours across time: response proportions and sample sizes (n)
Outcome Time-point
Baseline 1 month 3 months 6 months 12 months
VGS
Remission 0.051 0.375 0.418 0.552 0.617
Relapsed 0 0.013 0.055 0.115 0.170
Continuing 0.949 0.612 0.527 0.333 0.213
n 158 80 91 87 47
Gambling behaviours
Remission 0.051 0.488 0.468 0.500 0.533
Relapsed 0 0.012 0.117 0.211 0.289
Continuing 0.949 0.500 0.415 0.289 0.178
n 158 86 94 90 45
J Gambl Stud
123
Author's personal copy
gamble over remission increased by 20 %. For each one unit increase on GRCS the odds of
participants experiencing a relapse over remission increased by 6 %. The odds of
continuing to gamble over remission increased by 8 %. For each one unit increase on the
Work and Social Adjustment Scale the odds of a participant continuing to gamble over
remission increased by 13 %. Whilst AUDIT was significant in the overall regression
model (p = .02) there was no significant effect in either the relapse or continuing to
gamble categories when compared to the remission category.
For outcome measure gambling behaviours i.e. gambling once or more than either
planned abstinence or controlled gambling, each one unit increase on GUS the odds of
participants experiencing a relapse over remission increased by 16 %. The odds of par-
ticipants continuing to gamble over remission increased by 9 %. The odds of participants
continuing to gamble over remission increased by 18 % with each one unit increase on the
WSAS.
Discussion
The primary aim of this study was to model the probability of a study participant expe-
riencing a relapse in problem gambling following a period of remission as a function of
potential predicting variables. Outcome events were determined by two independent
methods for comparative purposes: scores on the VGS with a cut score of 21 and changes
in gambling behaviours. The candidate predictor variables covered a range of proximal and
distal factors including general mental health status, levels of social support and func-
tionality, sensation seeking traits, gambling specific measures of urge and cognitions, and
socio-demographic variables.
When the outcome status of relapse, remission, or continuing to gamble at each time
point was determined with scores on the VGS we found that gambling urge and gambling
related cognitions were significant predictors of relapse. Predictive factors of continuing to
gamble over remission were gambling urge, gambling related cognitions, and work and
social functionality. For outcome gambling behaviour, urge was a significant predictor of
both relapse and continuing to gamble. Work and social functionality was also a significant
predictor of continuing to gamble.
0.2
.4.6
.81
1 3 6 12time (months)
remission relapsedcontinuing
Fig. 1 Proportion versus time for outcome status using the Victorian Gambling Screen
J Gambl Stud
123
Author's personal copy
Similar to craving in substance use disorders, the urge to gamble has been designated a
conspicuous symptom in gambling disorders based on brain imaging studies and is an
antecedent to relapse (van Holst et al. 2010). Cognitions such as inability to stop gambling
have also been linked to relapse (Raylu and Oei 2002)) and a qualitative investigation has
identified both gambling related urges and erroneous cognitions as precipitants to relapse
(Oakes et al. 2011). Similarly, a prospective study of gambling relapse found that cog-
nitions relating to winning and giving into urges were associated with relapse (Hodgins and
el-Guebaly 2004). These findings indicate that relapse prevention should include tech-
niques targeting urge and cognitions with similar priority to ‘‘producing initial behaviour
change’’ (Brandon et al. 2007).
Problem gamblers may experience similar trajectories to relapse as in substance use
disorders where proximal and distal risk factors interact (Katie Witkiewitz and Marlatt
2007). For example, an individual with a baseline co-morbid disorder of depression (distal
risk) may be more vulnerable to return to previous gambling behaviours through changes
in proximal effects of gambling related cognitions such as ‘‘gambling makes things seem
better’’ (Raylu and Oei 2004a). Similarly, the urge to gamble may be translated by the
Table 3 Univariate mixed-effects multinomial logistic regression analyses describing each variable’sassociation with outcome patterns of response
Variables Victorian Gambling Screen Gambling behaviours
Wald p value Advanced to finalmodel?a
Wald p value Advanced to finalmodel?a
Time 80.47 \.001 Yes 82.65 \.001 Yes
Demographic
Age, years 5.12 .077 Yes 18.12 \.001 Yes
Gender 4.39 .111 Yes 10.64 .005 Yes
Marital status 7.36 .499 4.77 .782
Highest education level 2.06 .358 3.82 .148 Yes
Employment 0.46 .794 1.51 .470
Living arrangement 3.35 .188 Yes 0.20 .906
Duration of gamblingproblem, years
1.98 .372 1.60 .450
Measures
GUS 53.75 \.001 Yes 60.51 \.001 Yes
WSAS 62.02 \.001 Yes 72.89 \.001 Yes
DASS-21 45.26 \.001 Yes 41.10 \.001 Yes
MSPSS 11.72 .003 Yes 6.16 .046 Yes
AUDIT 9.19 .010 Yes 0.38 .826
AISS 4.50 .150 Yes 9.03 .011 Yes
GRCS 60.58 \.001 Yes 59.95 \.001 Yes
TAIb 36.29 \.001 22.92 \.001
GUS Gambling Urge Scale, WSAS Work and Social Adjustment Scale, DASS-21 Depression Anxiety andStress Scale, MSPSS Multidimensional Scale of Perceived Social Support, AUDIT Alcohol Use DisordersIdentification Test, AISS Arnett Inventory of Sensation Seeking, GRCS Gambling Related Cognition Scale,TAI Trait Anxiety Inventorya Advanced to final model if p \ .25b Collinearity with DASS-21, therefore did not advance to final model
J Gambl Stud
123
Author's personal copy
expectation of a more tolerable mood state and result in a relapse (Anton 1999; Raylu and
Oei 2004b). In this study we did not identify any distal factors which may have been a
result of the use of more global self-report measures such as DASS-21. Further studies may
consider more formal assessments to identify comorbid conditions such as clinical
depression and the interactions with more immediate precipitants such as gambling related
thoughts.
We accounted for the temporal associations between gambling behaviour and predictors
by using a repeated measure design or equivalently, a cross-sectional time series. For
example, it was found that gambling urge was a significant predictor of both relapse and
maintenance of problem gambling across time on two possible outcome measures. Mod-
elling of this trend focused on the individual deviation from the overall population trend
using an iterative method based on all available data. This provided a more precise esti-
mate of the probability of relapse from differing amounts of information prior to and
following behaviour at the individual level. This approach has been used previously in a
longitudinal study of relapse in cigarette smoking (Hedeker and Mermelstein 1996).
Other strategies to predicting relapse include non-linear methods such as cusp catas-
trophe modelling as previously used in alcohol use disorders (Katie Witkiewitz and Marlatt
2007). To increase sensitivity to minor changes in risk the catastrophe model comprises of
two possible values of control parameters (proximal and distal risks) that may result in
more than one value of the behaviour parameter (relapse). This approach has shown to
provide an overall better fit of the data when compared to linear models of drinking
behaviour. A drawback of this study was that cases having missing data were excluded
from analyses and model estimates were based on a complete case analysis. Alternative
Table 4 Mixed-effects multinomial logistic regression model (variables significant overall at p \ .05)describing the association between variables and patterns of response measured with the Victorian GamblingScreen and gambling behaviours
Variable Victorian Gambling Screen OR (95 % CI)a Gambling behaviours OR (95 % CI)a
Relapse versusremission
Continuing versusremission
Relapse versusremission
Continuing versusremission
Time 1.23(1.03–1.47)* 0.76(0.66–0.88)* 1.26(1.11–1.43)* 0.78(0.69–0.88)*
Gender – – 0.40(0.11–1.46) 0.37(0.14–1.00)
Livingarrangement
1.98(0.41–9.65) 0.49(0.17–1.40) – –
GUS 1.29(1.12–1.49)* 1.20(1.07–1.35)* 1.16(1.06–1.27)* 1.09(1.02–1.17)*
WSAS 0.93(0.78–1.10) 1.13(1.03–1.24)* 0.94(0.83–1.07) 1.18(1.09–1.27)*
DASS–21 – – 0.99(0.95–1.03) 0.98(0.95–1.01)
MSPSS 1.00(0.96–1.04) 1.02(0.99–1.05) – –
AUDIT 1.09(0.98–1.22) 0.96(0.88–1.04) – –
AISS – – 1.01(0.93–1.09) 1.05(0.99–1.11)
GRCS 1.06(1.01–1.12)* 1.08(1.04–1.13)* 1.02(0.99–1.05) 1.02(1.00–1.04)
GUS Gambling Urge Scale, WSAS Work and Social Adjustment Scale, DASS-21 Depression Anxiety andStress Scale, MSPSS Multidimensional Scale of Perceived Social Support, AUDIT Alcohol Use DisordersIdentification Test, AISS Arnett Inventory of Sensation Seeking, GRCS Gambling Related Cognition Scale
* 95 % confidence interval significant at p \ .05a Odds ratio = eb
J Gambl Stud
123
Author's personal copy
methods such as mixed-effects models or multiple imputation may have enhanced external
validity of findings by including all available data.
Our study has several limitations. Approximately 67 % of participant’s were recruited
from a single treatment service, limiting the generalisability of results to the wider pop-
ulation of problem gamblers. However, strengths of this homogenous sample included
minimisation of confounding that may result from a number of treatment types in a cohort
study. These findings would be useful for developing a model of relapse with treatment
specificity to problem gambling. For other addictions it has been suggested that ‘‘different
treatments may be associated with different relapse processes’’ that requires further
research (Brandon et al. 2007).
Approximately 22 % of participants did not complete a follow-up assessment beyond
baseline who were generally younger (p \ .01), mostly male (p \ .01), and had higher
AISS scores for sensation seeking traits (p = .02). Therefore, conclusions from this study
are drawn from a partially observed dataset. Although this loss of power could not be
reversed we used maximum likelihood estimation for all available data to help minimise
the impact of missing data. Our strategy for improving response rates included mailing
out of questionnaires to limit participant burden in attending appointments and offering
unconditional incentives. The plausibility of data missing at random (MAR), a necessary
assumption of mixed-models, may have been value-added in this study by collecting
additional secondary information on the ease of obtaining data. Future studies on gam-
bling relapse may consider this as part of a handling missing data strategy. (White et al.
2011).
Finally, there are other potentially important risk factors that were not investigated in
the current study. For example, self-efficacy where a person’s diminished confidence in
their perceived ability to execute control of gambling behaviours may mediate the effects
of an urge to gamble on relapse (Sharpe 2002). In terms of a multifaceted approach, low
self-efficacy may also interact with distal traits such as impulsivity or the related construct
of sensation seeking traits. We examined sensation seeking traits as an independent pre-
dictor whereas further research could explore alternative causal pathways for coping
strategies (Ledgerwood and Petry 2006).
Conclusion
This study identified predictive factors of relapse in problem gambling from a broad range
of candidate variables. Utilising alternative outcome measurements provided the oppor-
tunity to compare explanatory models and consider relapse processes beyond a singular
definition of relapse. A more dynamic approach to modelling of relapse will enable better
matching of aetiological factors to gambling interventions and reduce risk of relapse. This
study also highlighted the challenges presented from a cohort study of treatment seeking
problem gamblers.
Acknowledgments We would like to acknowledge the assistance of Statewide Gambling Therapy Ser-vice, Gambling Helpline, Relationships Australia South Australia, Pokies Anonymous, and Offenders Aidand Rehabilitation Service in recruitment and ongoing support for the follow-up of participants. This projectwas funded by Gambling Research Australia (Grant No. 084/06).
Conflict of interest The authors have no other conflicts of interest to declare.
J Gambl Stud
123
Author's personal copy
References
Anton, R. F. (1999). What is craving? Alcohol Research and Health, 23(3), 165–173.Ashrafioun, L., Kostek, J., & Ziegelmeyer, E. (2013). Assessing post-cue exposure craving and its asso-
ciation with amount wagered in an optional betting task. Journal of Behavioral Addictions, 1–5.Battersby, M., Pols, R., Oakes, J., Smith, D., Mclaughlin, K., & Baigent, M. (2010). The Definition and
Predictors of Relapse in Problem Gambling: From: http://www.gamblingresearch.org.au/home/research/gra?research?reports/the?definition?and?predictors?of?relapse?in?problem?gambling?(2010).
Becona, E. (1996). Prevalence surveys of problem and pathological gambling in Europe: The cases ofGermany, Holland and Spain. Journal of Gambling Studies, 12(2), 179–192.
Bondolfi, G., Osiek, C., & Ferrero, F. (2000). Prevalence estimates of pathological gambling in Switzerland.Acta Psychiatrica Scandinavica, 101(6), 473–475.
Brandon, T. H., Vidrine, J. I., & Litvin, E. B. (2007). Relapse and relapse prevention. Annual Review ofClinical Psychology, 3(1), 257–284.
Daughters, S. B., Lejuez, C. W., Strong, D. R., Brown, R. A., Breen, R. B., & Lesieur, H. R. (2005). Therelationship among negative affect, distress tolerance, and length of gambling abstinence attempt.Journal of Gambling Studies, 21(4), 363–378.
Delfabbro, P. (2009). Australasian Gambling Review: Independent Gambling Authority of South Australia.Dillman, D. A. (2007). Mail and internet surveys: The total design method (2nd ed.). New York: Wiley.Donovan, D. M. (1996). Marlatt’s classification of relapse precipitants: Is the Emperor still wearing clothes?
Addiction, 91(12s1), 131–138.Donovan, D., & Witkiewitz, K. (2012). Relapse prevention: From radical idea to common practice.
Addiction research & theory, 20(3), 204–217.Echeburua, E., Fernandez-Montalvo, J., & Baez, C. (2001). Predictors of therapeutic failure in slot-machine
pathological gamblers following behavioural treatment. Behavioural and Cognitive Psychotherapy, 29,379–383.
Edwards, P., Roberts, I., Clarke, M., DiGuiseppi, C., Pratap, S., Wentz, R., et al. (2002). Increasing responserates to postal questionnaires: Systematic review. British Medical Journal, 324(7347), 1183.
Ferris, J., & Wynne, H. (2001). The Canadian problem gambling index. Final Report: Canadian Centre onSubstance Abuse.
Gooding, P., & Tarrier, N. (2009). A systematic review and meta-analysis of cognitive-behavioural inter-ventions to reduce problem gambling: Hedging our bets? Behaviour Research and Therapy, 47(7),592–607.
Goudriaan, A. E., Oosterlaan, J., De Beurs, E., & Van Den Brink, W. (2007). The role of self-reportedimpulsivity and reward sensitivity versus neurocognitive measures of disinhibition and decision-making in the prediction of relapse in pathological gamblers. Psychological Medicine, 38, 41–50.
Government of South Australia: Consumer and Business Services. (2012). Gaming Machines Act 1992,Annual Report 2011–2012.
Hedeker, D., & Mermelstein, R. (1996). Application of random-effects regression models in relapseresearch. Addiction, 91(supplement), S211–S229.
Hodgins, D. C. (2009). Randomized trial of brief motivational treatments for pathological gamblers: More isnot necessarily better. Journal of Consulting and Clinical Psychology, 77(5), 950.
Hodgins, D. C., Currie, S. R., el-Guebaly, N., & Diskin, K. M. (2007). Does providing extended relapseprevention bibliotherapy to problem gamblers improve outcome? Journal of Gambling Studies, 23(1),41–54.
Hodgins, D. C., & el-Guebaly, N. (2004). Retrospective and prospective reports of precipitants to relapse inpathological gambling. Journal of Consulting and Clinical Psychology, 72(1), 72–80.
Hosmer, D., & Lemeshow, S. (2000). Applied logistic regression (2nd ed.). New York: Wiley.Ledgerwood, D. M., & Petry, N. M. (2006). What do we know about relapse in pathological gambling?
Clinical Psychology Review, 26(2), 216–228.Lesieur, H. R., & Blume, S. B. (1987). The South Oaks Gambling Screen (SOGS): A new instrument for the
identification of pathological gamblers. American Journal of Psychiatry, 144(9), 1184–1188.Long, J., & Freese, J. (2006). Regression models for categorical dependent variables using Stata. College
Station, Texas: Stata Corporation.Lorains, F. K., Cowlishaw, S., & Thomas, S. A. (2011). Prevalence of comorbid disorders in problem and
pathological gambling: Systematic review and meta-analysis of population surveys. Addiction, 106(3),490–498.
Lovibond, S. H., & Lovibond, P. F. (1995). Manual for the depression anxiety stress scales. Sydney:Psychology Foundation.
J Gambl Stud
123
Author's personal copy
Marlatt, G. A., & Gordon, J. R. (1985). Relapse prevention: Maintenance strategies in the treatment ofaddictive behaviors. New York: Guilford.
McMillen, J., & Wenzel, M. (2006). Measuring problem gambling: Assessment of three prevalence screens.International Gambling Studies, 6(2), 147–174.
Mullen, P. M. (2003). Delphi: Myths and reality. Journal of Health Organization and Management, 17(1),37–52.
Mundt, J. C., Marks, I. M., Shear, M. K., & Greist, J. M. (2002). The Work and Social Adjustment Scale: Asimple measure of impairment in functioning. The British Journal of Psychiatry, 180(5), 461–464.
National Research Council. (2010). The prevention and treatment of missing data in clinical trials. Panel onhandling missing data in clinical trials. Committee on National Statistics, Division of Behavioral andSocial Sciences and Education. Washington, DC: The National Academies Press.
Oakes, J. E., Pols, R. G., Battersby, M. W., Lawn, S. J., Pulvirenti, M., & Smith, D. P. (2011). A focus groupstudy of predictors of relapse in electronic gaming machine problem gambling, part 2: Factors that‘pull’ the gambler away from relapse. Journal of Gambling Studies,. doi:10.1007/s10899-011-9267-8.
Oei, T. P. S., & Gordon, L. M. (2008). Psychosocial factors related to gambling abstinence and relapse inmembers of gamblers anonymous. Journal of Gambling Studies, 24(1), 91–105.
Raylu, N., & Oei, T. P. (2002). Pathological gambling: A comprehensive review. Clinical PsychologyReview, 22(7), 1009–1061.
Raylu, N., & Oei, T. (2004a). The Gambling Related Cognitions Scale (GRCS): Development, confirmatoryfactor validation and psychometric properties. Addiction, 99(6), 757–769.
Raylu, N., & Oei, T. (2004b). The Gambling Urge Scale: Development, confirmatory factor validation, andpsychometric properties. Psychology of Addictive Behaviors, 18(2), 100–105.
Reinert, D., & Allen, J. (2002). The alcohol use disorders identification test (AUDIT): A review of recentresearch. Alcoholism, Clinical and Experimental Research, 26(2), 272–279.
Roth, M. (2003). Validation of the Arnett Inventory of Sensation Seeking (AISS): Efficiency to predict thewillingness towards occupational chance, and affection by social desirability. Personality and Indi-vidual Differences, 35, 1307–1314.
Shaffer, H. J., & Hall, M. N. (2001). Updating and refining prevalence estimates of disordered gamblingbehaviour in the United States and Canada. Canadian Journal of Public Health, 92(3), 168–172.
Sharpe, L. (2002). A reformulated cognitive–behavioral model of problem gambling: A biopsychosocialperspective. Clinical Psychology Review, 22(1), 1–25.
Shiffman, S. (1989). Conceptual issues in the study of relapse. In M. Gossop (Ed.), Relapse and addictivebehavior. London: Routledge.
Skrondal, A., & Rabe-Hesketh, S. (2003). Multilevel logistic regression for polytomous data and rankings.Psychometrika, 68(2), 267–287.
Smith, D. P., Pols, R. G., Battersby, M. W., & Harvey, P. W. (2013). The Gambling Urge Scale: Reliabilityand validity in a clinical population. Addiction research & theory, 21(2), 113–122.
Spielberger, C. D., Gorsuch, R. L., Lushene, R. E., Vagg, P. R., & Jacobs, G. A. (1983). Manual for thestate-trait anxiety inventory. Palo Alto: Consulting Psychologists Press.
StataCorp. (2008). Stata Statistical Software: Release 10.0. College Station, Texas: Stata Corporation.Tolchard, B., & Battersby, M. (2010). The Victorian Gambling screen: Reliability and validity in a clinical
population. Journal of Gambling Studies, 26, 623–638.van Holst, R., van den Brink, W., Veltman, D., & Goudriaan, A. (2010). Brain imaging studies in patho-
logical gambling. Current Psychiatry Reports, 12(5), 418–425.Wardle, H., Sproston, K., Orford, J., Erens, B., Griffiths, M. D., Constantine, R., et al. (2007). The British
gambling prevalence survey. London: The Stationery Office.White, I. R., Kalaitzaki, E., & Thompson, S. G. (2011). Allowing for missing outcome data and incomplete
uptake of randomised interventions, with application to an Internet-based alcohol trial. Statistics inMedicine, 30(27), 3192–3207.
Wilde, B., Goudriaan, A., Sabbe, B., Hulstijn, W., & Dom, G. (2013). Relapse in pathological gamblers: Apilot study on the predictive value of different impulsivity measures. Journal of Behavioral Addictions,2(1), 23–30.
Witkiewitz, K., & Marlatt, G. A. (2004). Relapse prevention for alcohol and drug problems: That was Zen,this is Tao. American Psychologist, 59(4), 224–235.
Witkiewitz, K., & Marlatt, G. A. (2007). Modeling the complexity of post-treatment drinking: It’s a rockyroad to relapse. Clinical Psychology Review, 27(6), 724–738.
Wong, I. L. K., & Ernest, M. T. S. (2003). Prevalence estimates of problem and pathological gambling inHong Kong. The American Journal of Psychiatry, 160(7), 1353.
Zimet, G. D., Dahlem, N. W., Zimet, S. G., & Farley, G. K. (1988). The multidimensional scale of perceivedsocial support. Journal of Personality Assessment, 52(11), 30–41.
J Gambl Stud
123
Author's personal copy