This is an openaccess article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on http://www.jmir.org/, as well as this copyright and license information must be included.
Webbased selfhelp interventions for problem drinking are coming of age. They have shown promising results in terms of costeffectiveness, and they offer opportunities to reach out on a broad scale to problem drinkers. The question now is whether certain groups of problem drinkers benefit more from such Webbased interventions than others.
We sought to identify baseline, clientrelated predictors of the effectiveness of Drinking Less, a 24/7, freeaccess, interactive, Webbased selfhelp intervention without therapist guidance for problem drinkers who want to reduce their alcohol consumption. The intervention is based on cognitivebehavioral and selfcontrol principles.
We conducted secondary analysis of data from a pragmatic randomized trial with followup at 6 and 12 months. Participants (N = 261) were adult problem drinkers in the Dutch general population with a weekly alcohol consumption above 210 g of ethanol for men or 140 g for women, or consumption of at least 60 g (men) or 40 g (women) one or more days a week over the past 3 months. Six baseline participant characteristics were designated as putative predictors of treatment response: (1) gender, (2) education, (3) Internet use competence (sociodemographics), (4) mean weekly alcohol consumption, (5) prior professional help for alcohol problems (level of problem drinking), and (6) participants’ expectancies of Webbased interventions for problem drinking. Intentiontotreat (ITT) analyses, using lastobservationcarriedforward (LOCF) data, and regression imputation (RI) were performed to deal with loss to followup. Statistical tests for interaction terms were conducted and linear regression analysis was performed to investigate whether the participants’ characteristics as measured at baseline predicted positive treatment responses at 6 and 12month followups.
At 6 months, prior help for alcohol problems predicted a small, marginally significant positive treatment outcome in the RI model only (beta = .18,
Although female and more highly educated users appeared slightly more likely to derive benefit from the Drinking Less intervention, none of the baseline characteristics we studied persuasively predicted a favorable treatment outcome. The Webbased intervention therefore seems well suited for a heterogeneous group of problem drinkers and could hence be offered as a firststep treatment in a steppedcare approach directed at problem drinkers in the general population.
International Standard Randomized Controlled Trial Number (ISRCTN): 47285230; http://www.controlledtrials.com/isrctn47285230 (Archived by WebCite at http://www.webcitation.org/5cSR2sMkp).
Problematic alcohol use is not only a pervasive individual problem; it also imposes serious health and social burdens on the general population [
It is well known that treatment response is not influenced by treatment alone [
We therefore investigate here whether specific baseline characteristics can be identified as predictors of a positive treatment outcome for problem drinkers in the Dutch population who completed a Webbased selfhelp intervention called Drinking Less. On the basis of predictors already reported in the literature, we hypothesized that six putative baseline characteristics—(1) female gender, (2) higher education, (3) Internet competence, (4) a moderate level of problem drinking, (5) prior professional help for problem drinking, and (6) high expectancy for positive results from a Webbased intervention—would interact with Drinking Less to predict a more favorable treatment outcome at followup. We conducted a secondary analysis of our Drinking Less trial data [
To the best of our knowledge, this is the first article that uses randomized trial data to assess predictors of short and longerterm outcomes in Webbased selfhelp for problem drinkers in the general population.
Data were retrieved from a pragmatic randomized trial with two parallel groups using block randomization stratified for gender, with followup at 6 and 12 months [
We kept our exclusion criteria to a minimum to facilitate a lowthreshold inclusion strategy consistent with the nature of selfhelp interventions without therapeutic guidance. We therefore did not conduct diagnostic interviews. After screening and baseline assessment, participants were randomly assigned to the experimental condition (the Drinking Less intervention) or to the control condition (an online psychoeducational brochure on alcohol use that could be read in 10 minutes) [
Flow of participants through the trial
Participants in the experimental condition received access to the Drinking Less intervention [
Drinking Less home page [
Our choice of baseline participant characteristics as putative predictors was based on theoretical assumptions and results from previous prediction studies [
The outcome measure was defined as the individual differences between baseline (T0) mean weekly alcohol consumption and the mean level of consumption at posttreatment (6 months, T1) and at followup (12 months, T2) in the total group. Alcohol consumption was assessed with the Dutch version of Weekly Recall (WR) [
We first used
In the third step, we created dichotomous measures for the continuous and categorical baseline variables, alongside the already dichotomous variable of gender (female gender: yes/no). Values on the WR scale were transformed into a variable distinguishing moderate problem drinking (14  35 mean weekly alcohol units for women, 21  50 for men) from severe problem drinking (> 35 or > 50 units women/men). Categorical variables with more than two categories were recoded into two meaningful categories: (1) education: high/low (university and professional degrees versus the rest); (2) Internet competence: experienced/beginner; (3) prior professional help for alcohol problems: yes/no; and (4) expectancies of Webbased intervention: high/low. We then applied regression analyses to ascertain whether these particular groups benefited more from the intervention than others. We assessed the interactions between the abovebaseline attributes and the Drinking Less intervention modality, and then the effects of those interactions on treatment outcome. In this model, the standardized individual change scores (pre to postintervention effect sizes) served as the dependent or outcome variable. The interaction terms of each participant characteristic with the intervention dummy (Drinking Less experimental condition = 1, control condition = 0) served as independent predictor variables, along with their constituent main effects.
We next calculated the product of the intervention dummy and each of the dummy variables describing the participants’ characteristics [
We subsequently repeated this procedure in completersonly analyses on those participants who completed the followup questionnaire at 6 months (n = 151) or at 12 months (n = 163) to verify whether the results of the two ITT analyses would be sustained. Finally, we used descriptive statistics to illustrate the changes in alcohol consumption over time in terms of the identified predictors. The sample size provided 24 participants per variable at 6 months and 26 per variable at 12 months [
The demographic and clinical characteristics of participants at baseline are shown in
Baseline characteristics of the 261 participants (values are numbers and percentages of participants, unless otherwise indicated)
Condition^{a}  
Experimental 
Control 

Female gender^{b}  64 (49.2)  64 (48.9) 
Education^{b}  
Low  41 (31.5)  38 (29.0) 
High (academic/professional)^{b}  89 (68.5)  93 (71.0) 
High Internet competence^{b}  104 (80.0)  100 (76.3) 
High treatment expectancy^{b}  61 (46.9)  66 (49.6) 
Weekly alcohol intake in standard units^{c}

43.7 (21.0)  43.5 (22.3) 
Moderate problem drinking^{b}

74 (56.9)  74 (56.5) 
Severe problem drinking 
56 (43.1)  57 (43.5) 
Prior professional help for problem drinking^{b}  18 (13.8)  15 (11.5) 
Contemplation stage^{d}  116 (89.2)  115 (87.8) 
Alcohol moderation as goal  120 (92.3)  123 (93.9) 
Age (mean, SD)  45.9 (8.9)  46.2 (9.2) 
Living with a partner  75 (57.7)  71 (54.2) 
Paid employment  94 (72.3)  96 (73.3) 
^{a}All differences between conditions were nonsignificant (tested at
^{b}Indicates putative predictor of favorable treatment response.
^{c}A standard unit contains 10 g of ethanol.
^{d}Assessed with validated Dutch version of Readiness to Change Questionnaire [
Participants who did not return the questionnaire 6 months after baseline did not differ from posttreatment responders in terms of the characteristics assessed at baseline (
Analyses of predictorbytreatment interaction effects in terms of a successful reduction of mean weekly alcohol use at 6 and 12 months showed similar results for the lastobservationcarriedforward (LOCF) and the completersonly model. We therefore present here only the intentiontotreat models. Results of the completersonly analysis are available from the first author.
Analyses of predictorbytreatment interaction effects in terms of a successful reduction of mean weekly alcohol use found no significant effects for the putative predictors at 6 months (
Predictorbytreatment interaction regressed individually using lastobservationcarriedforward (LOCF) imputation at 6 and 12month followup
Interaction term: participant characteristic by condition (Drinking Less = 1)  Effect on mean weekly alcohol consumption^{a} at 
Effect on mean weekly alcohol consumption^{a}


beta^{b} 


Beta^{b} 



Female  .003  .98  .03  .22  .045  .02 
High educational level  .17  .17  .03  .33  .01  .03 
High Internet competence  .13  .39  .03  .11  .44  .00 
High treatment expectancy  .09  .37  .03  .09  .37  .00 
Moderate problem drinking (female/male 1435 or 2150 units a week^{a})  .02  .86  .03  .04  .70  .06 
Prior help for drinking  .07  .48  .03  .05  .60  .00 
^{a}measured in standard units containing 10 g of ethanol
^{b}beta: standardized regression coefficient
^{c}
Predictorbytreatment interaction regressed individually using regression imputation (RI) at 6 and 12month followup
Interaction term: participant characteristic by condition (Drinking Less = 1)  Effect on mean weekly alcohol consumption^{a} at 6 months (N = 261)  Effect on mean weekly alcohol consumption^{a} at 12 months (N = 261)  
beta^{b} 


beta^{b}  P 


Female  .06  .53  .12  .27  .01  .03 
High educational level  .11  .37  .10  .21  .10  .03 
High Internet competence  .002  .99  .10  .06  .97  .01 
High treatment expectancy  .15  .14  .11  .04  .74  .00 
Moderate problem drinking (female/male 1435 or 2150 units a week^{a}  .08  .46  .16  .09  .39  .17 
Prior help for drinking  .18  .05  .11  .02  .79  .01 
^{a}measured in standard units containing 10 g of ethanol
^{b}beta: standardized regression coefficient
^{c}
We compared the mean weekly alcohol consumption at 6 and 12 months for the two conditions as shown by the intentiontotreat and completersonly analyses. The lastobservationcarriedforward (LOCF) model appeared to be the most conservative estimation method for the total group, as it returned the highest alcohol intake in both conditions—thus suggesting less improvement. We therefore chose these more cautious LOCF results to report outcomes for the two main predictors identified in our analysis. Detailed information about the other two models can be obtained from the first author.
Reductions in mean weekly alcohol consumption (in mean weekly units containing 10 g of ethanol) in experimental and control groups 6 and 12 months after baseline, by gender (LOCF)
At 6 months, the more highly educated Drinking Less (experimental) participants had achieved the greatest reduction in both absolute and relative terms (7.74 units, 19.0%) as compared to other categories (
Reductions in mean weekly alcohol consumption (in mean weekly units containing 10 g of ethanol) in experimental and control groups 6 and 12 months after baseline, by high and low education (LOCF)
The aim of this study was to determine whether some groups would benefit more than other groups from Drinking Less, a Webbased selfhelp intervention for problem drinkers, when assessed at 6 and 12 months. We investigated six characteristics of the participants at baseline as putative predictors of treatment response: (1) female gender, (2) high level of education, (3) high Internet experience, (4) moderate as opposed to severe level of problem drinking, (5) prior professional help for alcoholrelated problems, and (6) high expectancies for Webbased intervention.
At the 6month followup, we could not convincingly establish predictive value for any of these putative predictors, with the possible exception of prior help for alcohol problems, which was only marginally significant under the regression imputation model. Some other studies have likewise identified prior professional help as a predictor of positive clientbytreatment interaction leading to successful outcomes [
At 12 months, we found a modest prognostic value for female gender and for higher education; both variables were associated with better treatment response to the Drinking Less selfhelp intervention. Women who completed the intervention were found to have reduced their alcohol consumption to a significantly greater extent than men or than control group participants. Comparable results for female gender as a predictor of a successful brief intervention outcome in general population samples were reported by SanchezCraig and colleagues [
Higher levels of education also had modest predictive power and explained a small amount of variance at 12 months in combination with Drinking Less. This finding is consistent with results from other studies that identified high education as interacting with treatment interventions to produce favorable outcomes [
The other characteristics investigated were not found to act as predictors in our study. A moderate baseline level of problem drinking (in terms of mean weekly alcohol consumption) did not predict better outcomes than a severe level. This contrasts with the many studies that assume brief interventions to be better suited to moderate problem drinkers [
We did not find any predictive value for the two remaining putative predictors, Internet experience and positive expectancies of treatment efficacy, in contrast to some other studies that did [
This study has several limitations that are important to acknowledge. We conducted secondary analysis of data from our pragmatic randomized trial [
Secondly, we conducted a prespecified subgroup analysis and hence cannot rule out falsepositive or falsenegative predictors resulting from multiple testing [
We were also limited by the data in the number of predictors we could investigate. That prevented us from studying selfefficacy, a potentially important predictor [
Our study has a number of strengths as well. The study on which the analysis is based was one of the first pragmatic randomized trials on the effectiveness of Webbased selfhelp interventions without therapeutic guidance for problem drinkers in the general population. The data also enabled us to examine short and longerterm relationships. Because we had anticipated a high overall loss to followup when we first selected the trial sample, we included enough participants to ensure the statistical power to detect differences between the experimental and control conditions and between subgroups [
Female gender and a high level of education were found to have interacted with the Drinking Less selfhelp intervention to predict a somewhat better treatment response one year after the start of the intervention. This suggests that Webbased selfhelp without therapeutic guidance may hold a special attraction for problem drinkers with greater fears of stigmatization, including women or more highly educated people—population segments that might otherwise be difficult to reach with facetoface brief interventions [
At the same time, the effects of the predictors identified here offer only a very partial explanation for how client characteristics interact with treatment to affect outcome. Other baseline attributes such as selfefficacy may also play a role [
Our findings could enhance public health strategies that use steppedcare approaches to curb problem drinking in the general population. Since none of the groups we identified stood out conspicuously against others as deriving benefit from Drinking Less, we would argue that Webbased selfhelp is well suited to a broad, heterogeneous group of problem drinkers. It may therefore serve well as an initial intervention in a steppedcare model, suitable for matching to a large and varied group of problem drinkers in the general population and not just at more individual levels [
Our results add to the knowledge already gained from prediction studies in that we tested the role played by individual baseline attributes in the effectiveness of Webbased selfhelp for problem drinkers in the general population. The scope of future prediction research now needs to be extended to include the contributions of withintreatment progress variables, such as doseresponse relationships and the time required to initiate positive behavioral change, and of posttreatment variables like social support. Replication of our study is needed in view of the novelty of Webbased interventions for problem drinkers and the related prediction research.
The authors wish to thank the participants in this study. The study was funded by the Netherlands Organisation for Health Research and Development (ZonMw), grant # 2200.0140. We are grateful to Michael Dallas for his English language edit. All authors contributed to the conception, design and analysis of the study.
None declared.
attribute treatment interaction
intention to treat
last observation carried forward
quantityfrequency variability index
regression imputation
weekly recall