Content uploaded by Jussi Palomäki
Author content
All content in this area was uploaded by Jussi Palomäki on Aug 11, 2022
Content may be subject to copyright.
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 1
This article has been accepted for publication in Psychology of Addictive Behaviors. American
Psychological Association (APA) holds the copyright to the final version. This article may not
exactly replicate the authoritative document published in the APA journal. It is not the copy of
record. DOI: https://doi.org/10.1037/adb0000875 .
Predicting online problem gambling treatment discontinuation: New evidence from cross-
validated models
Jussi Palomäki1*, Kalle Lind2, Maria Heiskanen2, Sari Castrén234
1 Gambling Clinic, Helsinki University Hospital, Finland
2 Finnish Institute for Health and Welfare, Health and Well-being Promotion Unit, Finland
3 University of Turku, Department of Psychology and Speech-Language Pathology, Turku, Finland
4 University of Helsinki, Department of Medicine, Helsinki, Finland
*Correspondence: jussi.palomaki@hus.fi
Keywords: Problem gambling; treatment; discontinuation; dropout; cognitive-behavioral therapy
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 2
Abstract
Objective. There are tens of millions of problem gamblers in the world, many of whom either do
not seek treatment or fail to commit to it. Dropout rates are high, and not enough is known about
factors predicting treatment adherence. We focus on an online cognitive behavioral therapy
program for severe problem gambling to determine the likelihood of treatment discontinuation at
three different treatment phases: pre-treatment, before half-way, and before the end of the program.
Method. Participants were Finnish adults (N = 1139, 670 males, Mage = 34.5) registered in the
program between 2019 and 2021. Using logistic regression and 5-fold cross-validated naïve Bayes
classification, we predicted discontinuation with demographic-, psychometric-, and other gambling-
related variables, including the quality of one’s social relations, time spent on the waiting list, and
experienced readiness to behavioral change. Results. The models had acceptable predictive ability
(AUCs from .69 to .745; cross-validated balanced classification accuracy = 63.2%). In logistic
regressions, dropping out was prominently associated with younger age (p = .008), lower education
(p < .001), not being ready to change gambling behavior (p < .001), problem gambling severity (p <
.0001), longer time spent on the treatment waiting list (p < .0001), and fewer close social
relationships (p < .001). Conclusions. We found significant new real-world evidence on factors
statistically predicting treatment discontinuation, which is crucial when existing programs are
modified to better serve those in need.
Public health statement:
This study found that long treatment waitlist periods, few good social relationships, and not being
ready to change gambling behavior are highly significant predictors of online problem gambling
treatment discontinuation. This knowledge is crucial when existing treatment programs are
modified to better serve those in need.
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 3
Introduction
Problem gambling is a global phenomenon associated with a range of detrimental
consequences for the affected individuals, including financial losses, emotional turmoil, and
conflicts in social life (World Health Organization, 2019). Only about 10% of individuals
experiencing gambling problems seek help due to barriers such as stigma, feelings of shame,
wanting to handle problems alone, or not being ready to change one’s behavior (Hofmarcher et al.,
2020; Suurvali et al., 2009). Among those who engage in treatment, many re-schedule or cancel
their appointments frequently or simply fail to show up (Toneatto, 2005).
One way to tackle the barriers to seeking help is offering treatment online. Internet-based
therapies have gained popularity especially during the COVID-19 pandemic and they can be
delivered with or without therapist contact (Augner et al., 2022). Their advantages include easy
accessibility, flexibility, anonymity, and privacy, which help to increase treatment uptake in
populations characterized by low treatment rate and high attrition in traditional treatment settings
(van der Maas et al., 2019; Sagoe et al., 2021). Van der Maas et al. (2019) reviewed studies using
randomized controlled trials on Internet-based interventions and concluded that they are generally
effective on reducing problem gambling severity, anxiety, and depression (see also Buker et al.,
2021). Other studies without control groups have further shown that online interventions reduce
gambling-related erroneous thoughts, alcohol consumption, and distress (Castrén et al., 2013;
Carlbring et al., 2012; Rodda et al., 2018). Similarly, a recent meta-analysis (Augner et al., 2022) on
online interventions for gambling disorder, and its less severe forms, found moderate beneficial
effects, though only a minority of the studies included guided interventions. While unguided
interventions appeared to be the trend, no differences were reported between the outcomes of
guided and unguided interventions.
Peli poikki is a Finnish Internet-based 8-week cognitive behavioral therapy program
supported by therapist involvement via weekly telephone sessions. The program has previously
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 4
been assessed with encouraging results (Castrén et al., 2013; Palomäki et al., 2022), though the
treatment seems to work better for people without symptoms of depression or a lack of financial
control (Palomäki et al., 2022). In this paper we extend our focus to predictors of not completing the
program. Recovery from gambling problems is a non-linear process with periods of progress and
setbacks (Nilsson et al., 2021; Reith & Dobbie, 2013), and thus more knowledge is needed on the
social, psychological, and demographic factors contributing to successful treatment outcomes and
good retention rates (Sagoe et al., 2021).
In a recent meta-analysis, problem- and disordered gambling face-to-face treatment
dropout rates ranged between 0% and 72% across 24 studies with an overall weighted rate of 39.1%
(Pfund et al., 2021), which is in line with earlier meta-analytic findings (Melville et al., 2007) but
lower than the rates reported for Internet-based treatments, which, in turn, ranged from 38% to 83%
(van der Maas et al., 2019). Several demographic and psychological factors predict dropping out of
treatment, although findings across studies are inconsistent due to small sample sizes and different
methodological definitions of dropout (Melville et al., 2007; Pfund et al., 2021). For example,
dropping out may be defined as the discontinuation of treatment before finishing all or just some of
the prescribed sessions of a treatment protocol (Pfund et al., 2021; Pickering et al., 2018), or based
on therapist judgment (Smith et al., 2015). Pre-treatment dropout refers to those who sign up for
treatment but do not start it (Dowling, 2009). Definitional inconsistencies notwithstanding,
dropping out has been linked to demographic factors such as young age, low education, male
gender, living alone, not being married, being unemployed, having gambling debts and high
gambling frequency, and psychological factors such as impulsivity, sensation-seeking, prominent
feelings of guilt or shame, poor stress management, mental health problems, and substance use, as
well as to changing life circumstances (Alvarez-Aragay et al., 2015; Jimenez-Murcia et al., 2012;
Maniaci et al., 2017; Melville et al., 2007; Merkouris et al., 2016; Nilsson et al., 2021; Pfund et al.,
2021).
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 5
The term “dropout” often implies a failure in treatment commitment and thus has a
negative connotation. Not completing the Peli poikki program might reflect commitment difficulties
but may also be caused by no longer needing professional help, or other beneficial changes in one’s
life situation. Thus, we refrain from using the term “dropout” in relation to the Peli poikki program,
and instead use the more neutral term treatment “discontinuation”.
Internet-based treatments with therapist-guided interventions may induce high demand,
which leads to long waiting lists (Palomäki et al., 2022). Having to wait a long time for treatment
may negatively impact commitment and reduce the person’s readiness for making a behavioral
change (Redko et al., 2006), though to our best knowledge no studies have statistically evaluated
the predictive effects of waiting list times on the likelihood of treatment discontinuation. According
to the transtheoretical model of change (TMC; DiClemente & Prochanska, 1998), an individual’s
readiness in the process of behavioral change can affect treatment outcomes and adherence,
depending on which stage of the process the person is in (Legerwood et al., 2012; Johansen et al.,
2019; Rodda et al., 2015). Among these stages are pre-contemplation, contemplation, and action –
each describing the individual’s level of readiness to change their behavior – and measuring them
whilst in treatment or on the waiting list may provide new information on how to improve treatment
adherence.
Furthermore, the influence of an individual’s social network, such as the number of family
members and friends with whom they have a good relationship, as well as openness and support
from concerned significant others and peers, are significant factors contributing to recovery
(Nilsson et al., 2021; Pettersen et al., 2019). Concerned significant others are included into
treatment in couple and family interventions (Kourgiantakis et al., 2021). For women, relapse
frequencies were found to be higher if the person was divorced (Baño et al., 2021). Nonetheless, the
effect of close relationships on recovery or the likelihood of dropping out of problem gambling
treatment has not been thoroughly investigated (see Nilsson et al., 2021 for an exception).
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 6
In the current study we examine the likelihood of treatment discontinuation at three
different phases of the Peli poikki program: pre-treatment, before half-way, and before the end of
the treatment. The available data (from 2019 to 2021) provide a demographically representative
sample of individuals in a natural and recent “real world” treatment setting for severe gambling
problems, both pre- and post-COVID-19 pandemic. We perform exploratory cross-validated
predictions on treatment discontinuation using demographic-, psychometric- and other gambling-
related variables, including the quality of one’s social relations, time spent on the waiting list, and
experienced readiness to behavioral change. Our aim is to gain new knowledge on the statistical
predictors of treatment discontinuation, which can be leveraged to modify and tailor existing
programs to better serve those in need.
Methods
Design
The Peli poikki program is an 8-week, 8-module online cognitive behavioral therapy that
employs motivational interviewing to meet the clients’ needs. The modules start with a focus on
psychoeducation and motivation for treatment, followed by recognizing triggers, identifying
erroneous thoughts and the social consequences of gambling, and end with a focus on relapse
prevention (see Castrén et al., 2013 and Palomäki et al., 2022 for a full description of the treatment
modules). Peli poikki is easily accessible with a low threshold for entry, allowing and encouraging
all individuals in Finland to attend, and governed by the national helpline Peluuri. Participants
complete the weekly modules, receive weekly phone calls from a therapist, and have the option to
participate in an elective online discussion forum. Due to therapist involvement, Peli poikki has a
waitlist period. Thus, it differs from non-therapist-guided online interventions, which are often used
in research and for which a waitlist would not be needed (Augner et al., 2022).
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 7
The program is free, anonymous
1
, and open to all Finnish- or Swedish-speaking people
over 18 years of age, without any other exclusion criteria (it was originally adapted from the
program described in Carlbring & Smit, 2008). However, people with severe symptoms of
depression (MADRS >=20; see the Independent variables section) are encouraged to contact a
specialized physician
2
. The participants are allowed to have additional therapeutic support for other
mental health problems and to attend mutual support groups during the program, but simultaneous
treatment for gambling problems is discouraged. As an anonymous program, an official referral is
not needed. Clients are typically told about Peli poikki by the national helpline Peluuri (or its
website), or by social and health care professionals, as not all areas in Finland have outpatient
treatment available for gambling disorder and its less severe forms.
Participants completed pre- and post-treatment questionnaires and follow-ups 6 and 12
months after the treatment. The full details and contents of the program, including descriptions of
the post-treatment and follow-up measures, and treatment effectiveness, have been described in
Palomäki et al. (2022) and Castren et al. (2013). Since our focus in the current paper is on treatment
discontinuation, we analyze only data from the screening (upon entering the waiting list) and
baseline (right before beginning treatment, i.e., the moment of registering in) measurement periods,
thus, the post-treatment and follow-up measures are not analyzed. The questionnaires included
questions from 9 themes (including demographics, gambling history, current gambling behavior and
the games played, life situation, health, previous help seeking, motivation for change, and the
consequences of gambling; see the Measures-section for the variables selected for the current
study). In the current analyses, there is only one measurement per participant for each variable. The
1
The participants may choose to give their name or a pseudonym. They need a valid e-mail address and a phone
number (prepaid numbers are accepted). The Peli poikki program is not part of public health or social services; thus, it
is not linked with health registers, and it is managed by a non-governmental organization.
2
Those who score >= 20 on the MADRS self-report will be told about depression associated with gambling problems,
advised to contact a specialized physician, and to discuss the matter with their therapist in the Peli poikki program. The
therapists are trained to systematically probe participants’ suicidal thoughts and monitor for acute suicide risk, and, if
needed, guide people towards appropriate treatment.
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 8
demographic variables and previous help for gambling problems were measured during the
screening phase (upon entering the waiting list); other variables were measured during the baseline
phase (just before starting treatment). It took 30 minutes on average to complete a questionnaire. In
addition to the measures collected in the screening and baseline phases, we obtained data on
whether the participants had discontinued the treatment, and at which point (see the Measures
section for details).
Participants and therapists
The analyses cover the Peli poikki program data from 1.1.2019 to 15.12.2021. During this
time, 1139 individuals registered in the program (670 males, 459 females, 10 unreported; Mage =
34.5, SDage = 10.2). The median self-reported income category was €25000–€34999, with 24% of
participants having at least a bachelor's degree and 69.4% being employed. The mean reported
gambling start age was 23.1 (SD = 9.68). See table 1 for more details. The Peli poikki program was
delivered by five trained therapists, who were social workers with additional training in treating
addiction (e.g., training in motivational interviewing techniques or previous experience in substance
abuse treatment).
Measures
Dependent variables: Treatment discontinuation
There were three dichotomous dependent variables (DVs) indexing whether a participant
discontinued their treatment (0 = did not discontinue, 1 = discontinued) either 1) before the first
treatment module (did not discontinue [N=1075] vs. discontinued [N=64]), 2) before the 4th module
(did not discontinue half-way [N=880] vs. discontinued half-way [N=259]), and 3) before the final
8th module (completed whole treatment [N=782] vs. did not complete whole treatment [N=357]).
The discontinuation rates before the first session, before finishing the 4th module, and before
finishing the final 8th module were thus 5.6%, 22.7%, and 31.3%, respectively. Note that these rates
are cumulative, that is, people who discontinued treatment before the first module also (by
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 9
definition) discontinued it before the 4th and 8th modules, and people who discontinued before the
4th module also discontinued before the 8th module. Figure 1 illustrates the flow of participants from
the screening phase (i.e., entering the waitlist) to registering in the treatment and completing
modules 1–8. The currently analyzed data does not cover those individuals (N = 868) who entered
the waitlist but never registered in.
Figure 1
Participant flow through the Peli poikki program, from first entering the treatment waitlist, to
registering in the treatment and completing treatment modules 1-8.
Note. Participants could not proceed to subsequent treatment modules without having completed the
previous one. Here we do not analyze data from the 868 individuals who entered the treatment
waitlist but did not register in the program. The red negative numbers represent the number of
participants who discontinued treatment (cumulative numbers in brackets relative to those who had
registered in).
Independent variables
Psychometric scales. The self-report National Opinion Research Center DSM Screen for
Gambling Problems (NODS; Gerstein et al., 1999; Matheson et al., 2021) is a 17-item scale that
sums dichotomous items under conditional response options (score range: 0–10 points). Following
Palomäki et al. (2022), we used NODS as a continuous variable (Cronbach’s alpha = 0.69). When
responding to NODS, participants reflected on their past 2 months of gambling. The self-
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 10
administered Montgomery-Åsberg Depression Rating Scale (MADRS; Montgomery & Åsberg,
1979) was used to measure severity of depression, and the brief Alcohol Use Identification Test
(AUDIT-C; Bush et al., 1998) was used to measure alcohol consumption. MADRS has 9 items,
each evaluated from 0 to 6, and an overall score ranging from 0 to 54 points where higher points
indicate higher severity of depression (Cronbach’s alpha = 0.88). AUDIT-C has 3 items evaluated
on a scale from 0 to 4, overall scores range from 0 to 12 points, and higher points indicate higher
alcohol consumption (Cronbach’s alpha = 0.56). The Readiness to Change Questionnaire (RCQ;
Budd & Rollnick, 1996) is a 12-item scale with three subscales (Pre-contemplation, Contemplation,
and Action, with 4 items each) based on three stages of the transtheoretical model of change (TMC;
DiClemente & Prochaska, 1998). All RCQ items were evaluated on a Likert 1-4 agreement scale.
We averaged all items across the subscales (the pre-contemplation subscale was reverse -coded; see
Budd & Rollnick, 1996), which resulted in a uniform RCQ scale where higher scores reflect higher
overall readiness to behavioral change (Cronbach’s alpha = 0.64).
Demographic variables. Education (ranging from 1 = Primary education to 5 =
Bachelor’s degree or higher) and Income (ranging from 1 = €6500–€9999 to 7 = over €50000) were
measured as ordinal variables but analyzed as continuous predictors. Age and Gambling start age
were measured as continuous numerical variables and Gender as a dichotomous variable (0 = male,
1 = female). Finally, Employed was a dichotomous variable measuring whether the participant was
employed or not (0 = not employed, 1 = employed).
Other variables. Sense of financial control was a three-level categorical variable: “Do you
consider your financial situation as (1) Good, (2) Bad but in control, (3) Bad and not in control.”
Presence of gambling debt was a dichotomous variable (0 = no gambling debt, 1 = gambling debt),
and Physical health and Psychological health were evaluated by single Likert items (“What is your
current physical/psychological health situation?”; 1 = Very poor, 5 = Very good). Number of years
with gambling problems was measured by a single item (“How many years have you suffered from
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 11
gambling problems?”) and analyzed as a continuous variable. Sought help was a dichotomous
variable measuring whether the participant had previously sought help for their gambling problems
(0 = no, 1 = yes).
Participants were also asked whether specific Game types had been a cause of gambling
problems for them in the past year and responded either “yes” or “no” to a list of games (we
excluded games that were mentioned by fewer than 3% of the participants): 1) online slots or
videopoker, 2) other online casino games, 3) online scratch cards, 4) online slow lotteries, 5) online
fast lotteries (e.g., Bingo), 6) online sports betting, 7) online poker, 8) online computer, console or
mobile games, 9) offline slots or videopoker, 10) other offline casino games, 11) offline scratch
cards, 12) offline slow lotteries, 13) offline sports betting.
To measure quality of social relations, we calculated a Social score variable that indexed
the number of good social relationships the participants reported having. Participants were asked to
evaluate whether they had a good or bad relationship (or no relationship at all) with their 1) mother,
2) father, 3) siblings, 4) spouse, 5) children, and 6) friends. We then summed the number of good
relationships the participants reported having, which yielded an ordinal variable ranging from 0 to
6, where higher scores indicate a higher number of good social relationships (M = 4.01, SD = 1.36).
Cronbach’s alpha for the Social score measure was .707
3
. Social score is a novel measure and thus
we evaluate its construct validity via its correlations with other predictor variables (see figure 2).
Finally, we calculated the Number of days on waiting list for each participant based on the
time difference between their waiting list entry date and the date when they filled the baseline
questionnaire (M = 102.9, SD = 35.3). Table 1 lists all analyzed independent (predictor) variables
and their mean (SD)-, median category- or percentage split values, where appropriate.
3
The scale may be considered formative instead of reflective, in which case calculating Cronbach’s alpha may not be
warranted. In reflective measurement models, indicators of a construct are caused by that construct, while in formative
measurement models the measured variables are the cause of the latent variable. In the case of Social score, it may be
more sensible to consider “social score” being caused by good or bad relationships, rather than the goodness of social
relationships being caused by a latent “social score” construct.
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 12
Table 1
Analyzed independent (predictor) variables with Mean (SD), Median category or Percentage split
values, where appropriate
Variable
Mean (SD)
Median category
Percentage split
Psychometric scales
MADRS
1.74 (1.02)1
NODS
5.79 (2.85)
AUDIT-C
2.44 (0.93)
Readiness to change
3.66 (0.27)
Demographic variables
Gender (male / female)
58.8% / 40.3%
Age
34.5 (10.25)
Education
Upper secondary
Employed (no / yes)
30.6% / 69.4%
Income
€25000–€34999
Gambling start age
23.18 (9.68)
Other variables
Number of years with gambling problems
8.58 (7.34)
Sought help (no / yes)
42.6% / 57.4%
Sense of financial control2
7.9% / 53.1% / 39%
Gambling debt (no / yes)
21.2% / 78.8%
Psychological health
3.13 (0.88)
Physical health
3.41 (0.85)
Number of days on waiting list
102.9 (35.3)
Social score
4.01 (1.36)
Problem game types3
See note.
Note. 1 Mean value of the scale, which corresponds to a raw MADRS score of 10.44 (SD = 6.12). 2
Categories include “Good”, “Bad but in control” and “Bad and not in control”. 3 Problem game
types includes no/yes -questions for 1) online slots or videopoker (37.6% / 62.4%), 2) other online
casino games (42.6% / 57.4%), 3) online scratch cards (95.1% / 4.9%), 4) online slow lotteries
(95.8% / 4.2%), 5) online fast lotteries (e.g., Bingo; 96.1% / 3.9%), 6) online sports betting (83.2% /
16.8%), 7) online poker (95.1% / 4.9%), 8) online computer, console or mobile games (92% / 8%),
9) offline slots or videopoker (60.1% / 39.9%), 10) other offline casino games (91.4% / 8.6%), 11)
offline scratch cards (96.2% / 3.8%), 12) offline slow lotteries (95.6% / 4.4%), 13) offline sports
betting (91% / 9%).
Statistical analyses
Data were analyzed within the R platform for statistical computing (v. 4.0.5; R Core Team,
2013), first using logistic regression, and then 5-fold cross-validation to determine the predictive
ability of the models. Logistic regression models were initially fit separately for the three
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 13
dichotomous DVs, using all predictors. Thereafter, clearly non-significant predictors (ps > .1) were
removed and 5-fold cross-validation was performed on models with the remaining predictors. In k-
fold cross-validation (k = 5 in the current study), the dataset is first shuffled and then split into k
unique groups. Each group is used, in turn, as the test (or hold-out) set, while the remaining groups
are used collectively as the training set. The model is fit on the training set and its performance then
evaluated on the test set – in 5-fold cross-validation this process is repeated 5 times (such that each
unique group is used as the test set once), and the results are summarized. We used the naïve Bayes
and logistic regression methods to perform classification (i.e., to predict whether a specific
individual discontinued treatment or not). Naïve Bayes makes a strong assumption of statistical
independence between the included variables, which is almost always incorrect (hence the name
"naïve"), but it often still outperforms other modeling approaches. Model performance was
evaluated using odds ratios for individual predictors, Nagelkerke’s pseudo r2, area under the curve
(AUC) values, and average sensitivity-, specificity- and balanced accuracy values from the cross-
validated predictions (using the “caret” package in R; Kuhn, 2008). We also present zero-order
Pearson correlations for our predictor variables, and point-biserial correlations for dichotomous
variables (see figure 2).
There were no strongly influential outliers, and the continuous variables had linear
associations with the predicted logit-values. Age and Gambling start age had high multicollinearity
(VIFs > 5). Since Gambling start age had no statistically significant association with any of the
DVs (ps > .15), we dropped it from further analyses, which resulted in reduced and acceptable
multicollinearity (VIFs < 2.47). As an additional robustness check, we reran all logistic regression
analyses using bootstrapping (using the “boot” package in R; Canty & Ripley, 2021; Davison &
Hinkley, 1997) to obtain model estimates, but there was no change in the pattern of the results (thus
we do not report the bootstrapped results).
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 14
Given our relatively large sample size, all analyses were run by removing missing values
(listwise deletion). Therefore, in our final analyses the sample size ranged between 803 and 885
participants (depending on the analysis and variables included). The missing values are primarily
due to 153 participants not reporting their age, and 103 not reporting the number of years gambling
had been a problem for them. Little’s test (Little, 1988) indicated that the data were not missing
completely at random (MCAR), so in an additional analysis we replaced Age and Number of years
with gambling problems with two dichotomous variables (both coded 0 = value not missing, 1 =
value missing): having missing values in either variable was not significantly associated with the
likelihood of discontinuing treatment (all ps > .1). Thus, we concluded that the data were likely
missing at random (MAR), justifying listwise deletion. On the psychometric scales, only ~1% of the
responses to the MADRS, AUDIT-C and RCQ questions were missing, allowing us to calculate
participant-wise scale means despite the missing values. The missing values were also ignored
when calculating NODS scores. However, the results remained essentially unchanged even after
omitting all participants with any missing values.
As a robustness check we also analyzed our data by imputing all missing values using
predictive mean matching (PMM) with the mice R package (van Buuren & Groothuis-Oudshoorn,
2011). PMM is widely used, and it typically imputes more plausible values than other imputation
methods, since it draws real values sampled from the data as replacements for the missing values.
The results largely mirrored the results from the analyses using listwise deletion, confirming their
robustness (in tables 3 and 5 we report the imputed analysis results alongside the main analyses
without imputation).
Finally, as an exploratory analysis, we evaluated whether the COVID-19 pandemic period
affects the likelihood of treatment discontinuation, and whether the main results are robust to
controlling for periods of COVID lockdown. This was done in two parts: first, we created a
dichotomous variable indexing pre- and post-COVID periods (before [=0] and after [=1] first of
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 15
March 2020) based on the date when they registered in the program. Then, we created a
dichotomous variable indexing a specific lock-down period in Finland (which occurred from
17.3.20 to 15.6.20), again based on the date participants registered in the program (registered during
the lockdown period = 1, else = 0). Both dichotomous variables were used separately as additional
covariates predicting treatment discontinuation at the three program stages.
Transparency and Openness
Our sample size was fully determined by the availability of data: we used all available Peli
poikki program data from the years 2019 through 2021. All analyzed variables are reported, and
details on all measured variables are presented in Palomäki et al. (2022) and Castrén et al. (2013).
The data cannot be made available due to their sensitive nature and legal constraints. This study’s
design and its analysis were not pre-registered. The study procedures were carried out in accordance
with the Declaration of Helsinki. The study was approved by the Ethics committee of the Finnish
Institute for Health and Welfare.
Results
Zero-order correlations
We calculated zero-order Pearson correlations (and point-biserial correlations for
dichotomous measures) for all predictor variables (excluding Sense of financial control, which is
categorical, and the dichotomous Game type variables), which are presented in figure 2. Statistically
significant (p <.001) correlations are colored according to their direction (shades of red = positive,
shades of blue = negative). Since Social score was a measure developed for the current study, these
correlations also serve as evidence of the measure’s construct validity. The Social score measure
was negatively correlated (all ps < .001) with MADRS scores (r = -.26), NODS scores (r = -.13),
gambling start age (r = -.12), and positively with psychological (r = .23) and physical health (r =
.13), income (r = .18), and the likelihood of being employed (r = .17). Thus, higher social score,
which ostensibly reflects a stronger social support network, was linked with several self-reported
benefits in terms of health and socio-economic status. See figure 2 for further details.
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 16
Figure 2
Pearson- and point biserial correlations across all non-categorical predictor variables.
Note. Correlations at p < .001 are colored. Gender is coded as male = 0, female = 1; Employed as 0
= unemployed, 1 = employed; Sought help as 0 = has not previously sought help, 1 = has previously
sought help for gambling problems; Debt as 0 = no gambling debt, 1 = gambling debt.
DV: Discontinuation before first treatment module
The significant predictors of discontinuing treatment before the first module were Age
(OR = 0.93, 95% CI [0.87, 0.98], Wald = -2.19, p = .02), RCQ scores (OR = 0.28, 95% CI [0.08,
1.00], Wald = -2.02, p = .04), and Psychological health (OR = 2.08, 95% CI [1.05, 4.21], Wald =
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 17
2.09, p = .03). All other predictors were non-significant (ps > .08). Participants were more likely to
discontinue treatment if they were younger and not ready make a behavioral change. Participants
were also more likely to discontinue treatment if they reported higher psychological well-being.
However, due to a very low number of individuals who discontinued before the first treatment
module (5.6% of the individuals in the whole sample, and 3.9% in the statistical model with all
predictors), we were unable to perform reliable model cross-validations, and these results need to be
interpreted with caution (notably, the association with psychological health was not replicated in
subsequent analyses with more statistical power).
DV: Discontinuation before half-way of treatment
The predictors of treatment discontinuation before half-way of the treatment (before
finishing the 4th module) that were selected for the subsequent 5-fold cross-validation modelling
(based on the inclusion criterion of p < .1) were Age, Education, MADRS scores, NODS scores,
RCQ scores, Number of days on waiting list, Social score, and reporting that slow lotteries had been
a cause for gambling problems (see Table 2 for the full model details).
Table 2
Logistic regression model predicting treatment discontinuation before half-way of the Peli poikki program
Dependent variable =
Discontinued before half-way of treatment (0=No [N=642], 1=Yes [N=159])
Predictor
OR
95% CI
Wald
p-value
Gender (reference: "Male")
1.35
0.85–2.15
1.21
.193
Age
0.97
0.95–0.99
-1.98
.047
Education
0.77
0.63–0.93
-2.61
.009
Employed (reference: "No")
0.86
0.50–1.48
-0.55
.584
Income
0.94
0.82–1.07
-0.89
.375
Number of years with problems
1.01
0.98–1.04
1.07
.282
Sought help (reference: "No")
0.80
0.54–1.18
-1.09
.274
MADRS
1.37
1.03–1.84
2.18
.029
NODS
1.09
1.00–1.18
2.09
.036
AUDIT
0.95
0.78–1.17
-0.39
.692
Readiness to change
0.31
0.16–0.62
-3.31
<.001
Sense of financial control
(reference: “Good”)
Bad but in control
1.43
0.62–3.66
0.81
.419
Bad and not in control
1.25
0.50–3.42
0.48
.634
Gambling debt (reference: "No")
1.05
0.64–1.77
0.21
.836
Psychological health
1.25
0.91–1.73
1.40
.160
Physical health
1.04
0.80–1.34
0.31
.755
Number of days on waiting list
1.01
1.00–1.01
2.46
.013
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 18
Table 2 Continued
Predictor
OR
95% CI
Wald
p-value
Social score
0.88
0.78–1.05
-1.76
.077
Problem game types online
(reference: "No" for all)
Slots, videopoker
0.92
0.61–1.40
-0.36
.716
Other casino games
1.04
0.68–1.59
0.22
.825
Scratch cards
1.23
0.44–3.22
0.42
.678
Lotteries, slow
2.63
0.87–7.59
1.76
.077
Lotteries, fast (e.g. Bingo)
0.98
0.35–2.45
-0.03
.978
Sports betting
1.14
0.60–2.09
0.42
.676
Poker
1.12
0.42–2.66
0.25
.803
Computer, console and mobile
0.68
0.32–1.32
-1.08
.279
Problem game types offline
(reference: "No" for all)
Slots, videopoker
0.76
0.50–1.17
-1.21
.226
Other casino games
0.68
0.30–1.41
-0.96
.335
Scratch cards
0.49
0.11–1.73
-1.03
.304
Lotteries, slow
0.54
0.15–1.61
-1.03
.304
Sports betting
1.17
0.52–2.53
0.41
.678
Model performance
Nagelkerke r2 = .147
AUC = .717
Note. Highlighted cells (ps < .1) were retained in the subsequent predictive cross-validation model (Table 3).
In the final cross-validated model, statistically significant predictors were Age (OR = 0.98,
95% CI [0.96, 0.99], Wald = -2.22, p = .025), Education (OR = 0.73, 95% CI [0.61, 0.87], Wald = -
3.46, p = <.001), NODS scores (OR = 1.01, 95% CI [1.03, 1.18], Wald = 2.78, p = .005), RCQ
scores (OR = 0.35, 95% CI [0.19, 0.66], Wald = -3.31, p = <.001), Number of days on waiting list
(OR = 1.01, 95% CI [1.00, 1.01], Wald = 2.8, p = .005), and Social score (OR = 0.86, 95% CI
[0.75, 0.98], Wald = -2.24, p = .024). The model AUC was 0.692 and Nagelkerke’s r2 was 0.123.
Participants were more likely to discontinue before half-way of treatment if they were younger, had
lower education, reported high gambling severity, were not ready to make a behavioral change, had
fewer good social relationships, and spent a longer time in the waiting list to treatment.
In the 5-fold cross-validation analysis, we contrasted a logistic regression method with a
naïve Bayes method, both of which yielded higher-than chance detection accuracy, with the naïve
Bayes method performing better (53.2% [logistic regression] vs. 55.5% [naïve Bayes] of individuals
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 19
correctly identified as having or not having discontinued treatment when 50% is the chance-level;
see Table 3 for details).
Table 3
5-fold cross-validated model predicting treatment discontinuation before half-way of the Peli poikki program
Dependent variable =
Discontinued before half-way of treatment (0=No [N=701], 1=Yes [N=184])
Missing values
imputed (PMM)
Predictor
OR
95% CI
Wald
p-value
Wald
p-value
Age
0.98
0.96–0.99
-2.22
.025
-2.63
.008
Education
0.73
0.61–0.87
-3.46
<.001
-4.48
<.0001
MADRS
1.12
0.92–1.35
1.16
.243
1.30
.191
NODS
1.10
1.03–1.18
2.78
.005
2.05
.040
Readiness to change
0.35
0.19–0.66
-3.31
<.001
-3.64
<.001
Number of days on waiting list
1.01
1.00–1.01
2.80
.005
2.66
.007
Social score
0.86
0.75–0.98
-2.24
.024
-2.05
.040
PGT: Lotteries, slow (reference: “No”)
2.35
1.03–5.09
2.15
.031
1.00
.315
Model performance
Nagelkerke r2 = .123
AUC = .692
5-fold cross-validation method
Sensitivity
Specificity
Balanced accuracy*
Logistic regression
.081
.984
.532
Naïve Bayes
.185
.924
.555
Note. * Chance-level is .5. Balanced accuracy = (Sensitivity + Specificity) / 2. PMM = Predictive mean
matching imputation. PGT = Problem gambling type.
DV: Discontinuation before finishing treatment
The predictors of treatment discontinuation before finishing the entire treatment (before
finishing the 8th and final module) that were selected for the subsequent 5-fold cross-validation
modelling (based on the inclusion criterion of p < .1) were Age, Education, MADRS scores, NODS
scores, the RCQ subscales of pre-contemplation and contemplation, Psychological health, Number
of days on waiting list, Social score, and reporting that online computer, console and mobile games
had been a cause for gambling problems (see Table 4 for full model details).
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 20
Table 4
Logistic regression model predicting treatment discontinuation before finishing the Peli poikki program
Dependent variable =
Discontinuation before finishing treatment (0=No [N=574], 1=Yes [N=227])
Predictor
OR
95% CI
Wald
p-value
Gender (reference: “Male”)
1.20
0.79–1.84
0.89
.353
Age
0.98
0.96–1.00
-1.61
.100
Education
0.80
0.67–0.95
-2.43
.014
Employed (reference: “No”)
0.69
0.42–1.15
-1.41
.158
Income
0.96
0.85–1.10
-0.51
.612
Number of years with gambling problems
0.99
0.96–1.02
-0.12
.900
Sought help (reference: “No”)
0.78
0.54–1.12
-1.31
.188
MADRS
1.44
1.11–1.88
2.72
.006
NODS
1.12
1.04–1.21
3.09
.002
AUDIT
0.95
0.78–1.14
-0.52
.600
Readiness to change
0.34
0.17–0.64
-3.28
.001
Sense of financial control
(reference: “Good”)
Bad but in control
1.36
0.64–3.06
0.80
.424
Bad and not in control
1.33
0.58–3.16
0.67
.502
Gambling debt (reference: “No”)
0.92
0.57–1.47
-0.34
.730
Psychological health
1.22
0.91–1.64
1.39
.165
Physical health
1.08
0.85–1.36
0.51
.519
Number of days on waiting list
1.01
1.00–1.01
5.17
< .0001
Social score
0.86
0.75–0.98
-2.20
.027
Problem game types online
(reference: “No” for all)
Slots, videopoker
0.97
0.66–1.42
-0.14
.891
Other casino games
0.98
0.66–1.44
-0.08
.935
Scratch cards
0.92
0.35–2.29
-0.17
.864
Lotteries, slow
1.93
0.64–5.52
1.21
.225
Lotteries, fast (e.g., Bingo)
0.90
0.35–2.14
-0.21
.831
Sports betting
0.91
0.50–1.61
-0.30
.765
Poker
1.58
0.70–3.43
1.15
.249
Computer, console and mobile
0.41
0.20–0.79
-2.52
.011
Problem game types offline
(reference: “No” for all)
Slots, videopoker
0.83
0.56–1.22
-0.92
.358
Other casino games
1.26
0.66–2.36
0.73
.466
Scratch cards
0.68
0.19–2.11
-0.63
.529
Lotteries, slow
0.46
0.14–1.30
-1.38
.168
Sports betting
1.48
0.72–3.04
1.09
.276
Model performance
Nagelkerke r2 = .213
AUC = .744
Note. Highlighted cells (ps < .1) were retained in the subsequent predictive cross-validation model, see Table
5).
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 21
In the final cross-validated model, largely reflecting the earlier analysis on discontinuing
half-way, statistically significant predictors were Age (OR = 0.97, 95% CI [0.96, 0.99], Wald = -
2.64, p = .008), Education (OR = 0.75, 95% CI [0.64, 0.88], Wald = -3.46, p = <.001), NODS
scores (OR = 1.14, 95% CI [1.07, 1.22], Wald = 4.02, p < .0001), RCQ scores (OR = 0.36, 95% CI
[0.20, 0.65], Wald = -3.40, p = <.001). Number of days on waiting list (OR = 1.01, 95% CI [1.00,
1.01], Wald = 5.39, p < .0001), and Social score (OR = 0.80, 95% CI [0.71, 0.90], Wald = -3.54, p
< .001). The model AUC was 0.721 and Nagelkerke’s r2 was 0.186. Participants were more likely to
discontinue before finishing the treatment if they were younger, had lower education, reported
severe gambling problems, were not ready to make a behavioral change, had fewer good social
relationships, and spent a longer time in the waiting list to treatment.
In the 5-fold cross-validation analysis, we again contrasted a logistic regression method
with a naïve Bayes method, both of which yielded higher-than chance detection accuracy, with the
naïve Bayes method again performing better (61.3% [logistic regression] vs. 63.2% [naïve Bayes]
of individuals correctly identified as having or not having discontinued treatment when 50% is the
chance-level; see Table 5 for details).
Table 5
5-fold cross-validated model predicting treatment discontinuation before finishing the Peli poikki program
Dependent variable =
Discontinued before finishing treatment (0=No [N=625], 1=Yes [N=260])
Missing values
imputed (PMM)
Predictor
OR
95% CI
Wald
p-value
Wald
p-value
Age
0.97
0.96–0.99
-2.64
.008
-2.76
.006
Education
0.75
0.64–0.88
-3.46
<.001
-4.35
<.0001
MADRS
1.16
0.97–1.39
1.70
.088
1.26
.206
NODS
1.14
1.07–1.22
4.02
<.0001
3.41
<.001
Readiness to change
0.36
0.20–0.65
-3.40
<.001
-4.20
<.001
Number of days on waiting list
1.01
1.00–1.01
5.39
<.0001
5.68
<.0001
Social score
0.80
0.71–0.90
-3.54
<.001
-3.18
.001
PGT: OCCM (reference: “No”)
0.59
0.31–1.06
-1.69
.089
-1.80
.070
Model performance
Nagelkerke r2 = .186
AUC = .721
5-fold cross-validation method
Sensitivity
Specificity
Balanced accuracy*
Logistic regression
.296
.931
.613
Naïve Bayes
.392
.873
.632
Note. OCCM = Online computer, console, and mobile games. * Chance-level is .5. Balanced accuracy =
(Sensitivity + Specificity) / 2. PMM = Predictive mean matching imputation. PGT = Problem gambling type.
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 22
Effect of COVID time-period on dropout
Participants were less likely to discontinue treatment before half-way and before finishing
it, if they had registered in the program after March 1, 2020 (discontinuing before half-way: OR =
0.48, 95% CI [0.30, 0.78], Wald = -2.92, p = .003; discontinuing before finishing the program: OR
= 0.59, 95% CI [0.38, 0.91], Wald = -2.33, p = .019). However, there was no association between
the specific lockdown period with treatment discontinuation (occurring between 17.3.20 and
15.6.20; ps > .33). There was no change in the pattern of the main results by including the
dichotomous COVID-variables as exploratory covariates.
Summary of results
What emerge as the clearest pattern of results are the effects of Number of days on waiting
list, Social score, Readiness to change, NODS scores, Age, and Education on the probability of
treatment discontinuation, especially before half-way of treatment and before finishing the whole
treatment. These effects are visualized in Figure 3.
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 23
Figure 3
Probability of discontinuation across different treatment phases and predictor variables
Note. The probability of discontinuing treatment either before starting the treatment (yellow line),
before half-way of the treatment (blue line), and before finishing the treatment (red line), separately
for the six most significant predictors (Number of days on waiting list, Social score [i.e. number of
good social relationships], Age, Education, Readiness to change, and NODS score) and with 95%
confidence bands shaded.
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 24
Discussion
In the current study we performed cross-validated predictions for not completing the
Finnish Peli poikki therapist-guided online treatment for severe problem gambling, focusing on
demographic-, psychometric- and previously unexplored variables such as time spent on the waiting
list, and the number of good social relationships. We examined treatment discontinuation at three
cumulative time points: 1) before the first treatment module, 2) before the 4th module (half-way)
and 3) before the final 8th module. The overall discontinuation-rate of 31.3% was in line with the
rates of 30–40% reported in two meta-analyses (Melville et al., 2007; Pfund et al., 2021), but
slightly lower than the range of 38% – 83% reported in a review on Internet-based treatments (van
der Maas et al., 2019). Our results show that the six most significant predictors of treatment
discontinuation were younger age, low education, severity of problem gambling, fewer close social
relationships, not being ready to change gambling behavior, and longer time spent on the waiting
list (as shown in Figure 2). Notably, the observed effects of close social relationships, readiness to
change, and time spent on the waiting list have not previously been recognized. Thus, our results
have implications in developing problem gambling treatment contents in Finland, and in tailoring
treatment programs for online interventions more generally.
We found that the number of good social relationships was a strong individual predictor of
continuing treatment, which resonates well with earlier research: Having meaningful social
relationships is beneficial not only for protection against problem gambling but also for better
recovery from it, as well as adherence to treatment programs (Oksanen et al., 2021; Petry & Weiss,
2009). In the meta-analysis of Pfund et al. (2021), being married was associated with lower dropout
rates. Problem gambling can be detrimental for social relations by leading to arguments, mistrust,
feelings of shame, and divorces (Castrén et al., 2021; Hing et al., 2020; Jeffrey et al., 2019). Several
studies have shown that social relationships have both long-term and short-term impacts on physical
and mental health, as well as mortality risk (Ohrberger et al., 2017; Vuorinen et al., 2021; Yang et
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 25
al., 2016). Weak social networks and lack of meaningful social relationships have adverse effects
on psychiatric treatment outcomes (Kumar et al., 2021), including gambling treatment (Petry &
Weiss 2009), and stronger social support predicts successful gambling abstinence (Gomes &
Pascual-Leone, 2009; Oei & Gordon 2008). Lastly, both emotional- and instrumental social support
have positive effects on motivation and commitment to quit gambling, self-esteem, and depression,
all of which contribute to better problem gambling treatment outcomes (Gomes & Pascual-Leone
2009, Magnusson et al., 2019; Nilsson et al., 2020).
The participants were more likely to discontinue treatment the lower they scored on the
readiness to change measure. Readiness to change may vary during treatment, preventing some
from completing it (Miller & Rollnick, 2013). Those who are not ready to change or re-evaluate
their behavior may not be fully aware of its hazardous long-term effects (DiClemente, 1999;
DiClemente et al., 1991). On the other hand, people may be aware of the problem but not yet ready
to do anything about it (“I’ll do something someday” vs. “I’m doing something now”; Prochaska et
al., 2013). Individuals’ readiness to change reflects their mode of action and overall functioning and
is connected to treatment engagement as well as the severity of the problem being treated
(Merkouris et al., 2016).
The Peli poikki program is a low threshold treatment by design (it is free, anonymous, and
requires no referrals) and provided by a non-governmental organization (NGO). In Finland, public
services are generally responsible for problem gambling treatment, but despite improved
availability of treatment services in the past years, their accessibility still varies, and professionals’
awareness of the various treatment options is lacking (Heiskanen et al., 2020). Much of the support
and treatment for problem gambling is offered by NGOs. The Peli poikki program is well-known
and popular, and thus has a long waitlist period – a problem that is shared across the Finnish health
care services (e.g., Keskimäki et al., 2019). We found that spending more time on the treatment
waitlist significantly increased the probability of treatment discontinuation. The effects of waiting
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 26
list times on treatment adherence or effectiveness are not well understood. People may enter a
waitlist for treatment because they feel ready to make a behavioral change, but get discouraged by
the prolonged waiting (e.g., Müssener et al., 2019; Redko et al., 2006). On the other hand, waiting
time may increase motivation to change by giving people hope that help is coming in due course.
People may also already have received other support during the waitlist period: The Peli poikki
program guides people on the waiting list to other forms of support, such as professionally led
online mutual support groups. Alternatively, some individuals may have started to naturally
recover, and consequently they may not complete the whole program because they feel they no
longer need help in the remaining sessions.
Younger age, symptoms of depression and lower education have previously been linked
with the likelihood of discontinuing problem gambling treatment, which is in line with our current
findings (e.g., Pickering et al., 2018; Ronzitti et al. 2017; Araguay et al. 2015; Jiménez-Murcia et al.
2015). Symptom comorbidity is strongly linked with overall psychological well-being, both of
which hinder treatment commitment and effectiveness (Carlbring et al., 2012; Castrén et al., 2013;
Cunningham et al., 2019; Dowling et al., 2015; Palomäki et al., 2022). This is also reflected in the
current finding that high severity of problem gambling itself significantly increased the likelihood
of treatment discontinuation.
In further exploratory analyses on the COVID-19 pandemic period, we found that
participants were less likely to discontinue treatment if they had registered in the program after
March 1, 2020. This finding is tentative, as we did not observe it over a shorter specific lockdown
period in Finland. However, it is supported by a recent report by Baenas et al. (2021), who reported
a similar, albeit statistically nonsignificant, trend in an observational treatment setting, whereby
treatment dropout was less likely during a pandemic lockdown period. Further research may be
warranted on the potentially beneficial effects of the COVID-19 pandemic on treatment adherence.
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 27
Implications and suggestions for online treatment improvements
Our findings underscore the need for a better understanding on how social relations,
readiness to change, and waitlist periods impact online treatment adherence. Ideally, low threshold
online treatments for severe gambling problems should be suitable for many different individuals
regardless of their social situations. In the Peli poikki program, social relations are discussed in
module four, halfway of the program, where people are encouraged to discuss their gambling
problems with their significant others. Introducing this module earlier in the treatment could be
beneficial to the recovering gambler. However, due to the sensitive nature of the problem itself,
some clients may need more time to be ready to discuss the topic with their close others.
Should the clients think their social relations are scarce and not supportive, which may
undermine treatment commitment, the program could be adjusted to support them in other ways. In
addition to motivating the clients to strengthen their social relations with family members or
friends, they could be encouraged to contact people with lived experience from gambling problems
(Nilsson et al., 2021; Ortiz et al., 2021). The clients’ significant others can also be invited to be
involved in the treatment (Nilsson et al., 2018). There are specific treatment programs for
significant others alone or with their partner, and therefore encouraging them to support the gambler
should be recommended (Edgren et al., 2022). It is also possible that clients with few good social
relationships have difficulties connecting with people in general, including their therapist. Such
individuals may benefit more from a self-directed treatment option, though more work is needed to
better understand the role of social relationships in treatment adherence.
Identifying an individual’s level of readiness to change plays an integral role in treating
addictive disorders. Our results suggest that treatment programs should more carefully consider the
individual’s level of readiness and tailor the treatment length and contents to better suit them. For
example, some clients could be referred to a lighter (fewer than 8 modules) treatment option. This
may work well particularly in online treatment modalities where individuals’ readiness to change is
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 28
relatively high (Rodda et al., 2014). The waiting period before commencing the program is also
important because readiness to change and waiting period length are often linked in addiction
treatment. Our finding that longer waiting times increase the likelihood of treatment discontinuation
further underscores the need to tailor the treatment program according to clients’ varying situations.
Importantly, people’s readiness to change may shift during the waiting period, which, in turn,
affects their motivation to engage in treatment and may change their initial decision to seek help
(Wieczorek & Dąbrowska, 2018). A possible solution is to offer (and document) self-directed
interventions for well-motivated clients to shorten the waiting list queues. Another option is to offer
stepped care approaches, where the number of treatment modules offered is matched to the clients’
needs (i.e., a treatment protocol where clients do not need to finish all modules if they recover
earlier).
The contents and structure of the Peli poikki program or its means of communication
(written exercises, online platform, phone calls) may not be suitable for all ages and levels of
education. More carefully tailored interventions, perhaps with fewer written exercises and more
therapist contacts, might be needed for younger people and those with lower education, who may
find the program’s current components too cognitively demanding (no one-size-fits-all treatment
program exists). This may be partly explained by impulsivity, which is linked with both low
education and young age. Help-seeking individuals could be given more detailed descriptions of the
treatment content and the level of commitment required to motivate also younger and less educated
individuals to continue in the program. Reoccurring depression should also be assessed more
thoroughly and clinically before the treatment program, with follow-ups on the patients’ status
throughout the treatment should they need further care for depression. Due to the high comorbidity
between depression symptoms and problem gambling severity, better care for depression may
improve treatment adherence among those with high baseline problem gambling severity (which
was a significant predictor of treatment discontinuation). One way forward is to combine qualitative
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 29
and quantitative methods in future work to understand the complex reasons and life situations
behind the discontinuation of problem gambling treatment.
Limitations
While the Peli poikki program offers many benefits for scientific research (e.g., large
sample size and ecological validity) there are some noteworthy limitations. We were unable to
properly evaluate predictors of discontinuing treatment before the first module due to a low number
of observations. This issue can be solved in the future by re-analyzing the data when more
individuals have registered in the program
4
. Nonetheless, our current findings on treatment
discontinuation before the first module remain preliminary. There was no documentation of
simultaneous treatment or support received during the waiting list period, which may have been a
reason for participants to discontinue their treatment. However, the clients’ own regional public
health care providers may not have easily accessible and good-quality treatment for severe problem
gambling, which makes concurrent treatment unlikely (applying to the Peli poikki program in the
first place indicates that regional treatment was not available). The long waitlist period is largely
caused by the public sector’s inability to address the service needs in severe problem gambling.
Self-report measures entail well-known limitations such as participants’ unwillingness to reveal
personal details, exaggeration, or other biasing factors like responding according to social
desirability. Combining statistical and qualitative methods (e.g., semi-structured interviews) would
enrich the interpretation of the reasons for dropping out of treatment. Finally, while all therapists
delivering the Peli poikki program used the same 8-module treatment manual, no integrity checking
(such as a coding tool, Rodda et al., 2018) was used to assess a potential therapist effect.
4
While there are data on the Peli poikki program from the year 2007 onwards, changes were made to the program in
2018 (these changes concerned structural details of the measured variables and treatment contents), which prevent
combining the old and new data. Data on treatment effectiveness and moderators from the years 2007–2018 are
reported in Palomäki et al. (2022).
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 30
Conclusions
We identified several predictors of treatment adherence, three of which have been
previously under-explored yet statistically highly significant: time spent on the waiting list, the
number of good social relations, and self-reported readiness to behavioral change. Knowledge of
statistical predictors of treatment dropout is crucial for more informed decisions on how to modify
the existing programs so that they better serve those in need. We discussed the implications of our
findings and presented several concrete ways to modify existing online intervention programs to
better serve individuals with varying psychological and demographic characteristics.
References
Aragay, N., Jimenez-Murcia, S., Granero, R., Fernandez-Aranda, F., Ramos-Grille, I., Cardona, S.,
et al. (2015). Pathological gambling: Understanding relapses and dropouts. Comprehensive
Psychiatry, 57, 58–64. https://doi.org/10.1016/j.comppsych.2014.10.009
Augner, C., Vlasak, T., Aichhorn, W & Barth, A. (2022). Psychological online interventions for
problem gambling and gambling disorder – A meta-analytic approach. Journal of
Psychiatric Research, 151, 86–94.
Baenas, I., Etxandi, M., Codina, E., Granero, R., Fernández-Aranda, F., Gómez-Peña, M., ... &
Jiménez-Murcia, S. (2021). Does confinement affect treatment dropout rates in patients
with gambling disorder? A Nine-Month Observational Study. Frontiers in Psychology, 12.
Baño, M., Mestre-Bach, G., Granero, R., Fernández-Aranda, F., Gómez-Peña, M., Moragas, L., et
al. (2021). Women and gambling disorder: Assessing dropouts and relapses in cognitive
behavioral group therapy. Addictive Behaviors, 123, 107085.
https://doi.org/10.1016/j.addbeh.2021.107085
Budd, R. J., & Rollnick, S. (1996). The structure of the Readiness to Change Questionnaire: A test
of Prochaska & DiClemente's transtheoretical model. British Journal of Health
Psychology, 1(4), 365–376. https://doi.org/10.1111/j.2044-8287.1996.tb00517.x
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 31
van Buuren, S., & Groothuis-Oudshoorn, K. (2011). mice: Multivariate Imputation by Chained
Equations in R. Journal of Statistical Software, 45(3), 1–67.
https://doi.org/10.18637/jss.v045.i03
Bush, K., Kivlahan, D.R., McDonell, M.B., Fihn, S.D., Bradley, K.A., & Ambulatory Care
Quality Improvement Project (ACQUIP; 1998). The AUDIT alcohol consumption
questions (AUDIT-C): an effective brief screening test for problem drinking. Archives of
Internal Medicine, 158(16), 1789–1795. https://doi.org/10.1001/archinte.158.16.1789
Bücker, L., Gehlenborg, J., Moritz, S. et al. A randomized controlled trial on a self-guided Internet-
based intervention for gambling problems. Sci Rep 11, 13033 (2021).
https://doi.org/10.1038/s41598-021-92242-8
Calado, F., & Griffiths, M.D. (2016). Problem gambling worldwide: An update and systematic
review of empirical research (2000–2015). Journal of Behavioral Addictions, 5(4), 592–
613. https://doi.org/10.1556/2006.5.2016.073
Canty, A., Ripley, B.D. (2021). boot: Bootstrap R (S-Plus) Functions. R package version 1.3-28.
Carlbring, P., & Smit, F. (2008). Randomized trial of internet-delivered self-help with telephone
support for pathological gamblers. Journal of Consulting and Clinical Psychology, 76(6),
1090–1094. https://doi.org/10.1037/a0013603
Castrén, S., Lind, K., Hagfors, H., & Salonen, A.H. (2021): Gambling-related harms for affected
others: findings from a Finnish population-based survey. International Journal of
Environmental Research and Public Health, 18, 9564.
https://doi.org/10.3390/ijerph18189564
Castrén, S., Pankakoski, M., Tamminen, M., Lipsanen, J., Ladouceur, R., & Lahti, T. (2013).
Internet‐based CBT intervention for gamblers in Finland: experiences from the field.
Scandinavian Journal of Psychology, 54(3), 230–235. https://doi.org/10.1111/sjop.12034
Carlbring, P., Degerman, N., Jonsson, J., & Andersson, G. (2012). Internet-based treatment of
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 32
pathological gambling with a three-year follow-up. Cognitive Behaviour Therapy, 41(4),
321–334. https://doi.org/10.1080/16506073.2012.689323
Cunningham, J.A., Hodgins, D.C., Mackenzie, C.S., Godinho, A., Schell, C., Kushnir, V., &
Hendershot, C.S. (2019). Randomized controlled trial of an Internet intervention for
problem gambling provided with or without access to an Internet intervention for co-
occurring mental health distress. Internet Interventions, 17, 100239.
https://doi.org/10.1016/j.invent.2019.100239
DiClemente, C.C., & Prochaska, J.O. (1998). Toward a comprehensive, transtheoretical model of
change: Stages of change and addictive behaviors. In W.R. Miller & N. Heather
(Eds.), Treating addictive behaviors (pp. 3–24). Plenum Press. https://doi.org/10.1007/978-
1-4899-1934-2_1
Davison, A.C., Hinkley, D.V. (1997). Bootstrap methods and their applications. Cambridge
University Press: Cambridge.
DiClemente, C.C. (1999). Motivation for change: Implications for substance abuse treatment.
Psychological Science, 10(3), 209–213. https://doi.org/10.1111/1467-9280.00137
DiClemente, C.C., Prochaska, J.O., Fairhurst, S.K., Velicer, W.F., Velasquez, M.M., & Rossi, J.S.
(1991). The process of smoking cessation: an analysis of precontemplation, contemplation,
and preparation stages of change. Journal of Consulting and Clinical Psychology, 59(2),
295. https://doi.org/10.1037/0022-006X.59.2.295
Dowling, N. (2009). Client characteristics associated with treatment attrition and outcome in female
pathological gambling. Addiction Research & Theory, 17, 205–219.
https://doi.org/10.1080/16066350802346193
Dowling, N. A., Cowlishaw, S., Jackson, A. C., Merkouris, S. S., Francis, K. L., & Christensen, D.
R. (2015). Prevalence of psychiatric co-morbidity in treatment-seeking problem gamblers:
A systematic review and meta-analysis. Australian & New Zealand Journal of Psychiatry,
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 33
49(6), 519–539. https://doi.org/10.1177/0004867415575774
Edgren, R., Pörtfors, P., Raisamo, S., & Castrén, S. (2022). Treatment for the concerned significant
others of gamblers: A systematic review. Journal of Behavioral Addictions. Advance
online publication. https://doi.org/10.1556/2006.2021.00088
Reith, G., & Dobbie, F. (2013) Gambling careers: A longitudinal, qualitative study of gambling
behaviour. Addiction Research & Theory, 21(5), 376–390.
https://doi.org/10.3109/16066359.2012.731116
Gerstein, D., Hoffmann, J., Larison, C., Engelman, L., Murphy, S., Palmer, A., et al. (1999).
Gambling impact and behavior study. Report to the National Gambling Impact Study
Commission. National Opinion Research Center at the University of Chicago, Chicago.
Gomes, K., & Pascual-Leone, A. (2009). Primed for change: Facilitating factors in problem
gambling treatment. Journal of Gambling Studies, 25, 1–17.
https://doi.org/10.1007/s10899-008-9111-y
Hannöver, W., Thyrian, J. R., Hapke, U., Rumpf, H. J., Meyer, C., & John, U. (2002). The
readiness to change questionnaire in subjects with hazardous alcohol consumption, alcohol
misuse and dependence in a general population survey. Alcohol and Alcoholism, 37(4),
362–369. https://doi.org/10.1093/alcalc/37.4.362
Heiskanen, M., Kesänen, M. & Tenkanen, O. (2021) Rahapeliongelman hoidon saatavuus
Suomessa – kuntakyselyn tuloksia [Availability of gambling treatment in Finland – results
from a questionnaire to municipalities]. Tutkimuksesta tiiviisti 23/2021. Terveyden ja
hyvinvoinnin laitos, Helsinki. Available in Finnish only.
Hofmarcher, T., Romild, U., Spångberg, J., Persson, U. & Håkansson, A. (2020). The societal costs
of problem gambling in Sweden. BMC Public Health, 20, 1921.
https://doi.org/10.1186/s12889-020-10008-9
Hing, N., O’Mullan, C. , Nuske, E. , Breen, H. , Mainey, L. , Taylor, A., et al. (2020). The
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 34
relationship between gambling and intimate partner violence against women [Research
report]. Australia’s National Research Organisation for Women’s Safety Limited
(ANROWS).
Jimenez-Murcia, S., Aymami, N., Gomez-Pena, M., Santamaria, J. J., Alvarez-Moya, E.,
Fernandez-Aranda, F., et al. (2012). Does exposure and response prevention improve the
results of group cognitive-behavioural therapy for male slot machine pathological
gamblers? British Journal of Clinical Psychology, 51, 54–71.
https://doi.org/10.1111/j.2044-8260.2011.02012.x
Jeffrey, L., Browne, M., Rawat, V., Langham, E., Li, E., & Rockloff, M. (2019). Til debt do us part:
Comparing gambling harms between gamblers and their spouses. Journal of Gambling
Studies, 35(3), 1015–1034. https://doi.org/10.1007/s10899-019-09826-3
Johansen, A. B., Helland, P. F., Wennesland, D. K., Henden, E., & Brendryen, H. (2019).
Exploring online problem gamblers' motivation to change. Addictive Behaviors Reports,
10, 100187. https://doi.org/10.1016/j.abrep.2019.100187
Keskimäki, I., Tynkkynen, L., Reissell, E., Koivusalo, M., Syrjä, V., Vuorenkoski, L., Rechel, B. &
Karanikolos, M. (2019). Finland. Health system review. Health Systems in Transition, 21(2).
Kourgiantakis, T., Ashcroft, R., Mohamud, F., Fearing, G., & Sanders, J. (2021). Family-focused
practices in addictions: A scoping review. Journal of Social Work Practice in the
Addictions, 21(1), 18–53. https://doi.org/10.1080/1533256X.2020.1870287
Kuhn, M. (2008). Building Predictive Models in R Using the caret Package. Journal of Statistical
Software, 28(5), 1–26. https://doi.org/10.18637/jss.v028.i05
Kumar, N., Oles, W., Howell, B. A., Janmohamed, K., Lee, S. T., Funaro, M. C., et al. (2021). The
role of social network support in treatment outcomes for medication for opioid use
disorder: A systematic review. Journal of Substance Abuse Treatment, 127, 108367.
https://doi.org/10.1016/j.jsat.2021.108367
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 35
Ledgerwood, D., M., Wiedemann, A., A., Moore, J., & Arfken, C., L. (2012). Clinical
characteristics and treatment readiness of male and female problem gamblers calling a state
gambling helpline, Addiction Research & Theory, 20(2), 162–171, https://doi:
10.3109/16066359.2011.588815
Little, R.J.A (1988). A test of missing completely at random for multivariate data with missing
values. Journal of the American Statistical Association, 83(404), 1198–1202.
https://doi.org/10.1080/01621459.1988.10478722.
van der Maas, M., Shi, J., Elton-Marshall, T., Hodgins, D.C., Sanchez, S., Lobo, D.S., Hagopian,
S., & Turner, N.E. (2019). Internet-Based Interventions for Problem Gambling: Scoping
Review. JMIR mental health, 6(1), e65. https://doi.org/10.2196/mental.9419
Maniaci, G., La Cascia, C., Picone, F., Lipari, A., Cannizzaro, C., & La Barbera, D. (2017).
Predictors of early dropout in treatment for gambling disorder: The role of personality
disorders and clinical syndromes. Psychiatry Research, 257, 540–545.
https://doi.org/10.1016/j.psychres.2017.08.003
Magnusson, K., Nilsson, A., Andersson, G., Hellner, C., & Carlbring, P. (2019). Internet-delivered
cognitive-behavioral therapy for significant others of treatment-refusing problem gamblers:
A randomized wait-list controlled trial. Journal of Consulting and Clinical Psychology,
87(9), 802–814. https://doi.org/10.1037/ccp0000425
Matheson, F.I., Dastoori, P., Hahmann, T., Woodhall-Melnik, J., Guilcher, S.J., & Hamilton-
Wright, S. (2021). Prevalence of Problem Gambling Among Women Using Shelter and
Drop-in Services. International Journal of Mental Health and Addiction, 1–12.
https://doi.org/10.1007/s11469-021-00524-z
Maynard, B.R., Wilson, A.N., Labuzienski, E., Whiting, S.W. (2018). Mindfulness-based
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 36
approaches in the treatment of disordered gambling: A systematic review and meta-
analysis. Research on Social Work Practice, 28(3), 348–362.
https://doi.org/10.1177/1049731515606977
Melville, K.M., Casey, L.M., & Kavanagh, D.J. (2007). Psychological treatment dropout among
pathological gamblers. Clinical Psychology Review, 27(8), 944–958.
https://doi.org/10.1016/j.cpr.2007.02.004
Merkouris, S.S., Thomas, S.A., Browning, C.J., & Dowling, N.A. (2016). Predictors of outcomes of
psychological treatments for disordered gambling: A systematic review. Clinical
Psychology Review, 48, 7–31. https://doi.org/10.1016/j.cpr.2016.06.004
Miller, W. R., & Rollnick, S. (2013). Motivational interviewing: Helping people change (3rd
ed.). Guilford Press.
Montgomery, S.A., & Åsberg, M. (1979). A new depression scale designed to be sensitive to
change. The British Journal of Psychiatry, 134(4), 382–389.
https://doi.org/10.1192/bjp.134.4.382
Müssener, U., Linderoth, C., & Bendtsen, M. (2019). Exploring the experiences of individuals
allocated to a control setting: findings from a mobile health smoking cessation trial. JMIR
Human Factors, 6(2), e12139. https://doi.org/10.2196/12139
Nilsson, A., Simonsson, O. & Hellner, C. (2021). Reasons for dropping out of internet-based
problem gambling treatment, and the process of recovery – a qualitative assessment.
Current Psychology, 1–12. https://doi.org/10.1007/s12144-021-02368-1
Nilsson, A., Magnusson, K., Carlbring, P., Andersson, G., & Hellner, C. (2020). Behavioral couples
therapy versus cognitive behavioral therapy for problem gambling: A randomized
controlled trial. Addiction, 115(7), 1330–1342. https://doi.org/10.1111/add.14900
Nilsson, A., Magnusson, K., Carlbring, P. (2018). The Development of an internet-based treatment
for problem gamblers and concerned significant others: A pilot randomized controlled trial.
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 37
Journal of Gambling Studies, 34, 539–559. https://doi.org/10.1007/s10899-017-9704-4.
Oei, T.P.S., & Gordon, L.M. (2008). Psychosocial factors related to gambling abstinence and
relapse in members of Gamblers Anonymous. Journal of Gambling Studies, 24(1), 91–105.
https://doi.org/10.1007/s10899-007-9071-7
Ohrnberger, J., Fichera, E., & Sutton, M. (2017). The relationship between physical and mental
health: A mediation analysis. Social Science & Medicine, 195, 42–49.
https://doi.org/10.1016/j.socscimed.2017.11.008
Oksanen, A., Sirola, A., Savolainen, I., Koivula, A., Kaakinen, M., Vuorinen, I., et al. (2021).
Social ecological model of problem gambling: A cross-national survey study of young
people in the United States, South Korea, Spain, and Finland. International Journal of
Environmental Research and Public Health, 18(6), 3220.
https://doi.org/10.3390/ijerph18063220
Ortiz, V., Cain, R., Formica, S.W., et al. (2021). Our voices matter: Using lived experience to
promote equity in problem hambling prevention. Current Addiction Reports, 8, 255–262.
https://doi.org/10.1007/s40429-021-00369-5.
Palomäki, J., Heiskanen, M., & Castrén, S. (2022). Online 8-week cognitive therapy for problem
gamblers: The moderating effects of depression symptoms and perceived financial
control. Journal of Behavioral Addictions, 11(1), 75–87.
https://doi.org/10.1556/2006.2021.00091
Pettersen, H., Landheim, A., Skeie, I., Biong, S., Brodahl, M., Oute, J., & Davidson, L. (2019).
How social relationships influence substance use disorder recovery: A collaborative
narrative study. Substance Abuse: Research and Treatment, 13, 1178221819833379.
https://doi.org/10.1177/1178221819833379
Petry, N. M., & Weiss, L. (2009). Social support is associated with gambling treatment outcomes in
pathological gamblers. The American Journal on Addictions, 18(5), 402–408.
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 38
https://doi.org/10.3109/10550490903077861
Pfund, R.A., Peter, S.C., McAfee, N.W., Ginley, M.K., Whelan, J.P., & Meyers, A.W. (2021).
Dropout from face-to-face, multi-session psychological treatments for problem and
disordered gambling: A systematic review and meta-analysis. Psychology of Addictive
Behaviors. Advance online publication. https://doi.org/10.1037/adb0000710
Pickering, D., Keen, B., Entwistle, G., & Blaszczynski, A. (2018). Measuring treatment outcomes
in gambling disorders: A systematic review. Addiction, 113, 411–426.
https://doi.org/10.1111/add.13968
Prochaska, J.O., Norcross, J.O., & DiClemente, C.C. (2013). Applying the stages of change.
Psychotherapy in Australia, 19 (2), 10–15.
Redko, C., Rapp, R.C., & Carlson, R.G. (2006). Waiting Time as a Barrier to Treatment Entry:
Perceptions of Substance Users. Journal of Drug Issues, 36(4), 831–852.
https://doi.org/10.1177/002204260603600404
Rodda, S., Merkouris, S.S., Abraham, C., Hodgins, D C., Cowlishaw, S., & Dowling, N.A. (2018).
Therapist-delivered and self-help interventions for gambling problems: A review of
contents. Journal of Behavioral Addictions, 7(2), 211–226.
https://doi.org/10.1556/2006.7.2018.44
Rodda, S. N., Lubman, D. I., Iyer, R., Gao, C. X., and Dowling, N. A. (2015). Subtyping based on
readiness and confidence: the identification of help-seeking profiles for gamblers
accessing web-based counselling, Addiction, 110, 494–501.
https://doi.org/10.1111/add.12796
Sagoe, D., Griffiths, M.D., Erevik, E. K., Høyland, T., Leino, T., Lande, I.A., Sigurdsson, M.E., &
Pallesen, S. (2021). Internet-based treatment of gambling problems: A systematic review
and meta-analysis of randomized controlled trials. Journal of Behavioral Addictions, 10(3),
546–565. https://doi.org/10.1556/2006.2021.00062
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 39
Salonen, A.H., Hagfors, H., Lind, K., & Kontto, J. (2020). Gambling and problem gambling:
Finnish Gambling 2019: Prevalence of at-risk gambling has decreased. THL Statistical
report 9/2020. Retrieved on 10.3.22 from:
https://www.julkari.fi/bitstream/handle/10024/139651/SR2019_tilastoraportti%20englanni
ksi%2020042020_final%20%282%29.pdf?sequence=1&isAllowed=y
Smith D.P., Battersby M.W., Harvey P.W., Pols R.G., Ladouceur R. (2015). Cognitive versus
exposure therapy for problem gambling: Randomised controlled trial. Behaviour Research
and Therapy, 69, 100–110. https://doi.org/10.1016/j.brat.2015.04.008
Suurvali, H., Cordingley, J., Hodgins, D., & Cunningham, J. (2009). Barriers to seeking help for
gambling problems: A review of the empirical literature. Journal of Gambling Studies,
25(3), 407–424. https://doi.org/10.1007/s10899-009-9129-9
Team, R. C. (2013). R: A language and environment for statistical computing.
Toneatto, T. (2005). A Perspective on Problem Gambling Treatment: Issues and Challenges.
Journal of Gambling Studies, 21(1), 75–80. https://doi.org/10.1007/s10899-004-1925-7
Vuorinen, I., Oksanen, A., Savolainen, I., Sirola, A., Kaakinen, M., Paek, H.J., & Zych, I. (2021).
The Mediating Role of Psychological Distress in Excessive Gambling among Young
People: A Four-Country Study. International Journal of Environmental Research and
Public Health, 18(13), 6973. https://doi.org/10.3390/ijerph18136973
World Health Organization (WHO) (2019). ICD-10: International Statistical Classification of
Diseases and Related Health Problems (11th ed.). Author.
Wieczorek, Ł. & Dąbrowska, K. (2018). What makes people with gambling disorder undergo
treatment? Patient and professional perspectives. Nordic Studies on Alcohol and Drugs 35
(3), 196–214.
Williams, R.J., Volberg, R.A., & Stevens, R.M.G. (2012). The population prevalence of problem
gambling: Methodological influences, standardized rates, jurisdictional differences, and
PREDICTING PROBLEM GAMBLING TREATMENT DISCONTINUATION 40
worldwide trends. Report prepared for the Ontario Problem Gambling Research Centre
and the Ontario Ministry of Health and Long Term Care. Retrieved on 10.3.22 from:
https://opus.uleth.ca/bitstream/handle/10133/3068/2012-PREVALENCE-
OPGRC%20(2).pdf
Yang, Y.C., Boen, C., Gerken, K., Li, T., Schorpp, K., & Harris, K.M. (2016). Social relationships
and physiological determinants of longevity across the human life span. Proceedings of the
National Academy of Sciences, 113(3), 578–583. https://doi.org/10.1073/pnas.1511085112