The proposed research will evaluate a full Internet-based self-directed intervention for problem gamblers using a single blind, randomized controlled trial comparing participants who are provided access to the SCTs as compared to those who are only provided the CYG personalized feedback screener (the control condition). There is one primary hypothesis and four secondary hypotheses:
Hypothesis 1: Respondents will display significant reductions in their gambling behaviour in the twelve months after being provided access to the online SCTs as compared to participants only provided with the normative feedback control intervention (CYG).
In addition, two potential moderators and two potential mediators of the impact of the full intervention, program involvement and perceived self-efficacy, will be tested as secondary hypotheses:
Hypothesis 2: Respondents with greater baseline self-efficacy will display significant reductions in their gambling behaviour in the twelve months after being provided access to the online SCTs or the CYG compared to participants with lower baseline self-efficacy.
Hypothesis 3: Respondents who have never accessed gambling or mental health treatment will display significant reductions in their gambling behaviour in the twelve months after being provided access to the online SCTs or the CYG compared to participants with previous treatment involvement.
Hypothesis 4: Respondents in the SCTs intervention condition who have more involvement with the SCTs intervention between baseline and three-month follow-up will demonstrate more improvement in gambling outcomes at six- and twelve-month follow-up, compared to respondents who have less involvement with the SCTs intervention.
Hypothesis 5: Respondents in the SCTs intervention condition will display significant increases in their perceived self-efficacy to deal with their gambling concerns in the three months after being provided access to the online SCTs as compared to participants in the control condition. Participants with greater improvement in self-efficacy will show greater improvement in gambling behaviour.
The research methods to be used in this study have been approved by the University of Calgary Conjoint Faculties Research Ethics Board (certificate 7279).
Participant recruitment and randomization
Following procedures we have used in our earlier studies, media announcements (newspapers, radio, and websites) will be used to recruit individuals concerned about their gambling and interested in web-based self-directed treatment. Inclusion criteria will be identical to Hodgins et al. (2009): 18 years of age or older; perception of a gambling problem and scoring a 3 or greater on the Problem Gambling Severity Index of the Canadian Problem Gambling Index (PGSI- CPGI; ); gambled at least once in the past month; not involved in treatment at present (includes Gamblers Anonymous and any medical or psychological treatment where gambling problems are addressed); willingness and ability to access a website in English (to ensure reading ability); willingness to have telephone contacts recorded, willing to provide follow-up data on gambling; willingness to provide the name of a collateral (family or friend) to help locate them for follow-up interviews and the name of the same or a different collateral for data validation. Use of psychiatric medications for other mental health disorders will not be an exclusion criterion, although use will be assessed and monitored as a potential treatment moderator.
To ensure an adequate number of participants are assessed within the timeframe, recruitment will occur across Canada. Both urban and rural settings will be targeted. Interested individuals will be provided with a website address with information about the study, including eligibility criteria and a toll-free number to call or email address to use if interested. Having participants access a website at this point in the recruitment process is designed to minimize the number of participants who are randomly assigned but never access the intervention or control website (30% in one of our recent alcohol trials).
Individuals will be contacted by telephone and if they meet eligibility criteria and provide informed consent, they will receive a brief telephone assessment (gambling history and behaviour, self-efficacy) and then will be randomly assigned to one of the two intervention conditions, stratified on sex, gambling problem severity and treatment history (yes or no)  using MINIM, a computer program which uses the method of minimization. Problem severity, based on the NODS described below, will be defined as low-moderate (6 or fewer DSM-IV criteria) or high (7–10 criteria).
Self-change Tools (SCTs)
SCTs, as described above, integrate the self-directed written materials  that were evaluated in the three brief treatment trials [13, 15, 46] and a trial of a self-directed relapse prevention for gambling problems . A major focus is to provide individuals with clear and concise behavioural and cognitive strategies for meeting the goal of reducing or quitting gambling. The materials are presented as a series of options from which participants will choose what strategies seem most relevant to them. Participants will have ongoing access to the site over the follow-up period to make it as similar as possible to other web resources in terms of accessibility. The enhancements to the workbook content include three features. First, whereas the workbook encourages individuals to self-monitor their gambling, the website will provide a more structured option for individuals to maintain an online log of gambling and gambling urges. A further option will be to use a smartphone application to collect this self-monitoring information. A second enhancement is the expansion of the self-assessment component of the workbook to include more structured personalized normative feedback of gambling involvement. Again, the workbook provides some of this feedback but in a less comprehensive way. The third enhancement is the option for individuals to receive motivational email or text reminders of their progress and goals. This feature is somewhat similar to the motivational telephone booster calls provided in Hodgins et al. (2009).
CYG will be used as the control intervention. Although we hypothesize that the SCTs will be a more powerful intervention than the CYG, CYG is a credible and ethically appropriate control comparison. In the CYG, the participant completes a brief assessment and then receives a personalized feedback report. The personalized feedback materials start out with a brief statement of the purpose of the report (“help to give you a picture of your gambling and let you know how your gambling compares with other Canadians”). The person is then provided with a summary of the types of gambling engaged in, along with a comparison of how this relates to other Canadians of their sex, a summary of their Problem Gambling Severity Index score and interpretation and a description of the types of gambling cognitions that the person endorsed. The final element of the feedback is a list of techniques that the person could adopt to lower the risk associated with their gambling.
The brief telephone baseline assessment (adapted from Hodgins et al. 2009) will include a demographic profile (age, sex, education, marital status, income, ethnicity and racial status, employment status) and a gambling, mental health and treatment history including a timeline interview of types of gambling (online, land-based), frequency, and money spent for the past three months [48, 49]. Problem gambling severity will be assessed using the past year Problem Gambling Severity Index and the lifetime and past three month version of the NORC DSM-IV Screen for Gambling Problems (NODS) [50, 51] which indicates DSM-IV severity. Hodgins  administered the NODS to problem gamblers as part of a 1-year follow-up after a brief treatment to assess its utility as a treatment outcome measure. Internal reliability was fair to good and the factor structure and item-total correlations supported the existence of a single higher order construct that correlated moderately with gambling behavior and outcome. Self-efficacy will be administered using the Gambling Abstinence Self-efficacy Scale (GASS, , a 21 item self-report scale with evidence of concurrent and predictive validity in problem gambling treatment samples. In addition, participants will be asked to identify a treatment goal (quit or reduce gambling) and how successful they think they would be (0 “not at all” to 10 “extremely”) in the next 6 months and in the next 12 months. The Kessler 10 (K10) questionnaire will be included to provide a continuous measure of general psychological distress that is responsive to change over time. The K10 has been well validated and its brevity and simple response format are attractive features. It also produces a summary measure indicating probability of currently experiencing an anxiety or depressive disorder [54, 55]. Quality of life will be assessed by the WHOQoL-8, an eight item version of a widely used measure. This short form has been used in a number of countries, is robust psychometrically, and overall performance is strongly correlated with scores from the original WHOQoL .
After three, six and twelve months, a follow-up assessment of gambling behaviour (timeline method), problem gambling severity (NODS), self-rated improvement, self-efficacy (GASS), psychiatric distress, quality of life, use of other treatment resources, and impressions of site features and tools will be conducted. Although it would appear efficient to conduct follow-up assessments via the Internet, previous experience has shown that attrition is extremely high with this type of design. In our previous research, follow-up interviews were conducted via telephone with good follow-up rates. As a result, our follow-up interviews in this study will be conducted via telephone. We will collect extensive contact information for participants to minimize losing contact (e.g., contact information for family and friends, work contact, email contact, etc.).
Participants will also be asked to provide the name of a collateral (family or friend) who can confirm their self-reported gambling behaviour through a brief telephone call. In Hodgins et al., (2001, 2009) agreement between participants and collaterals was r = .61 and ICC = .69 for dollars lost and r = .66 and ICC = .53 for days of gambling, indicating fair agreement. As in previous research [49, 57], collaterals who described themselves as more confident showed better agreement and in general collaterals tended to report less gambling than participants suggesting that participant reports are not attenuated by minimization.
We will have access to a complete record of the amount and type of use participants make of the SCTs (and the CYG). Following methods used by Danaher (2008) and Strecher (2008), we will operationalize degree of involvement with SCTs by recording the number of times the participant accesses the site as well as the number of tools the participant uses (as assessed by page views, form completions, etc.) and length of involvement with the site (e.g. use of the site over time). This information will be used to test the mediation hypothesis that degree of involvement is related to success at overcoming gambling problems.
Primary and secondary outcome variables
Two major outcome variables for testing the primary and secondary hypotheses will be: mean days per month gambling and NODS scores. Mean dollars lost per gambling day, total dollars lost and self-rated improvement will be used as secondary outcome variables. Days and dollars spent gambling will be calculated for the three months pre-treatment and for each follow-up period. The data will be inspected for approximate normality or symmetry and, if necessary, subjected to appropriate transformation.
Participants will be aware that they will be assigned to one of two intervention options, although neither one will be described as potentially superior. Baseline assessment occurs prior to randomization. Follow-up assessments are conducted by interviewers who will be blind to participant assignment.
Primary analyses for Hypothesis 1, comparing outcomes for the two groups, will be based on the appropriate random effects model to properly account for the longitudinal nature of this data. Condition (2) and Time (0,3,6,12) will be modeled as fixed factors while the participants will be modeled as random factors. Separate analyses will be conducted for each primary and secondary outcome variable. This same analytic approach will be used for Hypothesis 2 and 3, examining moderators. Baseline self-efficacy, past use of treatment, and baseline NODS scores will be included as additional factors. For Hypothesis 3 and 4, the mediation hypotheses, the procedures recommended by Preacher and Hayes  will be conducted using a series of random effect models. Missing data should not be a major issue but techniques for these matters may be used depending on the nature, location and type of missingness.
Sample size and power analysis
We propose to collect a sample of 180 participants and we estimate (based upon Hodgins et al., 2009) that we will successfully follow about 153 participants at three months. This number will ensure a heterogeneous sample of individuals that will provide a valid assessment of the perceived value of different components of the SCTs. Based upon previous experience, about half will be of each sex. This number will also provide sufficient power to conduct the proposed statistical tests comparing the two conditions, based upon gambling frequency and NODS data from Hodgins et al. (2001, 2009), assuming a correlation of .5 between baseline and follow-up values, power = 0.80 and a Bonferroni corrected α = .025 (.05/2 outcome variables). This sample size is estimated so as to be sufficient to detect a difference of about 2 gambling days per month between conditions at each follow-up interval (medium effect size). This degree of difference is clinically meaningful in terms of gambling involvement. Similarly, this sample size will detect a 1 point difference on the NODS at 12 months. Given the complexity of estimating power for HLM models, these calculations are based upon a more simple repeated measures ANOVA model , with an attrition rate (i.e., not followed after baseline assessment) estimated to be 15%. The proposed hlm analysis will likely have greater statistical power because all observed data are included.