A randomized controlled trial (RCT) was conducted to compare the 2 versions of Change Your Drinking with follow-up surveys 6 weeks and 3 months after registration. Participants were invited by email and reimbursed with a €10 shopping voucher for their participation. The study was approved by the ethics committee of the Department of Applied Human Sciences at the University of Magdeburg-Stendal (Ref 4973-15) and was registered with Current Controlled Trials (ISRCTN: 31586428).

The intervention and the study were purely Web-based on the addiction prevention website Drugcom [23] and the alcohol prevention website Kenn dein Limit (Know Your Limit) [24] both run by the BZgA. Recruitment of participants started in December 2010 and ended in March 2012. All users of the freely accessible Check Your Drinking self-assessment were invited for the study if they met the eligibility criteria described subsequently. The results of the self-assessment were used both for the Change Your Drinking program and as baseline data for the trial.

If the eligibility criteria (see Study Criteria) were met, users were informed about the study. A portable document format (PDF) file containing all relevant study details was offered for download (see Multimedia Appendix 1). Users who were willing to participate were then asked to register and to provide their informed consent by clicking an “I agree to participate” button. After successfully confirming their email addresses, participants were randomly assigned either to the original (version 1) or the revised program (version 2) by random number generator software. Researchers could not influence nor predict the randomization result. The participants were blind to the results of the randomization because they only received detailed information about the program version they were allocated to.

Those who opted not to participate in the study or who did not meet the eligibility criteria had full access to the original version of the program and were not included in any follow-up surveys.

Trial data were collected via self-assessment in the baseline survey, as well as 6 weeks and 3 months afterwards. The past 7 days were used as reference period for alcohol consumption. In order to quantify the alcohol intake, participants were first asked to indicate their number of alcohol use days in the previous 7 days. Afterwards, they were requested to specify the number and type of alcoholic beverages consumed on each drinking day. Using these details, the amount of pure alcohol and the number of standard glasses (SG) per day was calculated. According to the BZgA, 1 SG in Germany corresponds to 10 grams of pure alcohol [25]. If 5 or more SG were consumed on any of the previous 7 days, this was classified as binge drinking.

Another outcome was risky consumption in the past 7 days, as defined by the following factors: (1) an average of more than 24 or 12 g (for males and females, respectively) of pure alcohol per day (these are the tolerable upper alcohol intake levels for males and females in Germany according to Burger and colleagues [26]), or (2) more than 5 days of consumption, or (3) at least 1 incident of binge drinking in the reference period. The definition of risky alcohol use has been derived from the guidelines for low-risk consumption as defined by the BZgA [27].

Alcohol-related problems were measured by a German version of McGee and Kypri’s [28] Alcohol Problems Scale (APS). The scale consists of 14 items describing negative consequences of alcohol consumption, such as vomiting, unprotected sexual intercourse, or blackouts, during the last 30 days. Each item is to be answered with yes, no, or “no answer.”

The AUDIT was used as part of the baseline survey to test for the study criteria. A cut-off of 8 points as suggested by Babor et al [29] was used to define risky alcohol consumption. As a measure of the program usage we tracked the diary usage (used at least once: yes/no).

To be invited for the study, participants had to be Internet literate, at least 18 years old, and had either reached the cut-off of 8 points in AUDIT or have consumed more than 24/12 g (male/female) of pure alcohol per day on average in the past week. We did not use stricter study criteria (eg, exclusion of individuals currently in other treatment) because we wanted to increase the generalizability of the results to the regular users of Change Your Drinking.

Change Your Drinking is an Internet-based self-help program based entirely on automated tailored feedbacks. A comparison of both versions of the intervention is displayed in Table 1. Exemplary feedbacks of each version in German language are included in Multimedia Appendix 2.

The comparison of both program versions was conducted with generalized estimating equations (GEE) using Stata 11 (StataCorp LP, College Station, TX, USA), accounting for all 3 data collection points. The GEE analyses were modeled with unstructured correlation matrices between the data collection points. We assumed to have binary distributed data (binge drinking yes/no), Poisson-distributed data (alcohol use days and alcohol-related problems), and a negative binomial distribution with log link in the alcohol intake data (intake in grams).

In the GEE analyses, a group difference was assumed in case of a significant group × time interaction. To measure the development of both groups, the time effect of each outcome was examined. In case of any statistically significant result, Cohen’s d was calculated.

In a first step of data analysis, we tested whether group differences at baseline or the program usage (operationalized by online diary used: yes/no) moderated the effects of the group assignment on the study outcomes. In case of significance, the respective measure and its interaction with the group factor were included in the GEE; otherwise, it was not considered in the effectiveness testing.

For the intention-to-treat analyses (ITT), missing data was estimated by multiple imputations with Stata’s “ICE” command. We performed 10 imputations. The results from the multiple imputations were compared with completer analyses and last observation carried forward (LOCF) analyses. In the completer analyses, missing follow-up data was not imputed, so only those cases that provided follow-up data were analyzed. In the LOCF analyses, missing data was replaced with data from the preceding data collection point.

To compare both groups at study baseline and to determine whether baseline measures were predicting follow-up participation, logistic regression analyses were performed. In all analyses, we used a 2-sided significance level of alpha=.05. The study was powered to detect a group difference of d≥0.20. Therefore, we aimed for a sample size of N=624 (alpha=.05; power=0.80).

The research is reported in accordance with the E-CONSORT checklist [36] (see Multimedia Appendix 3).

Each time the Change Your Drinking starting page was opened, it was checked whether data for the Check Your Drinking self-assessment were available and whether the study criteria were fulfilled or not. In the trial period, Check Your Drinking was completed 10,887 times. A total of 5823 cases did not meet the study criteria and 4469 users refused to participate. Thus, 595 persons were included in the trial and were randomized, resulting in 2 approximately equally sized groups (see Figure 2). Participation in the trial (N=595) compared to refusing to participate (n=4469) was predicted by female gender (odds ratio [OR] 1.68, 95% CI 1.40-2.00, P<.001) and higher education (OR 1.39, 95% CI 1.22-1.58, P<.001). Thus, in the group of participants, females (38.8%) and individuals with high school education (65.0%) were more highly represented than in the group of nonparticipants (females: 27.4%, participants with high school education: 52.9%). Significant group differences were also found for the utilization of professional treatment (OR 1.83, 95% CI 1.10-3.03, P=.02) and age (OR 0.99, 95% CI 0.99-1.00, P=.04). That is, participants in the trial tended to use professional help more often (10.3% vs 5.9%) and were slightly younger (between-group d=0.09) than individuals who refrained from participating. Baseline alcohol use was not associated with trial participation (use days: OR 1.00, 95% CI 0.96-1.04, P=.95; intake in grams: OR 1.00, 95% CI 1.00-1.00, P=.16).

In total, 345 persons participated in the first follow-up and 295 in the second, resulting in follow-up rates of 58.0% and 49.6%, respectively. Although loss to follow-up was not predicted by group allocation (OR 0.82, 95% CI 0.56-1.20, P=.31), the level of education, program usage, and alcohol use were significant predictors. Thus, those who took part in the follow-up surveys were more highly educated (OR 1.36, 95% CI 1.14-1.61, P=.001), used the diary more often (OR 6.56, 95% CI 4.39-9.81, P<.001), and consumed less alcohol (OR 0.97, 95% CI 0.94-1.00, P=.04). However, we do not expect any significant bias on that account because these measures were all included in the equations of the multiple imputations.

There were no significant group differences at baseline (see Table 2). Mean age of participants was approximately 30 years and most participants (364/595, 61.2%) were male. The education level of the participants was relatively high, with 387 of 595 (65.0%) having attended high school. A total of 534 of 595 (89.7%) of the participants were currently not using any other professional help to deal with their alcohol use.

In contrast to the baseline measures, the usage of the diary was significantly higher in the revised version of the program (OR 3.33, 95% CI 2.30-4.81; P<.001). According to the analysis plan, the diary usage was included in the first step of the effectiveness analyses.

Diary usage did not moderate the effects of the group assignment on any of the study outcomes (alcohol use days: beta = –0.04, 95% CI –0.22 to 0.14, P=.66; alcohol intake: beta = –0.01, 95% CI –0.13 to 0.11, P=.85; binge drinking: beta = 0.27, 95% CI –0.32 to 0.85, P=.37; alcohol-related problems: beta = 0.04, 95% CI –0.11 to 0.19, P=.56; risky drinking: beta = –0.49, 95% CI –1.02 to 0.03, P=.07); therefore, it was removed from the analysis.

There were no significant group differences in any of the primary outcomes of the study (see Table 3). Thus, the number of alcohol use days, the alcohol intake, the frequency of binge drinking, the number of use-related problems, and the indicators for risky consumption follow a similar direction in both groups.

However, significant overall reductions in alcohol use can be noted. Thus, after 3 months, participants in both program versions consumed on average 1.2 days (d=0.46) and 133.3 grams (d=0.59) less than at baseline. A significant reduction in alcohol-related problems (d=0.36), binge drinking incidents (reduction of 31.1%), and risky drinking (reduction of 23.6%) were seen. All the results were confirmed through completer and LOCF analyses.

A key limitation to this study was the reliance on self-reported data. However, this type of data has repeatedly been shown to be reliable and valid [39,40]. Moreover, verification with urine samples or clinical interviews was not a practical option considering the widely scattered sample.

Because of the financial compensation given for follow-up participation, it cannot be ruled out that several participants tried to register more than once for the study. Although technical measures were taken to prevent this from happening, the anonymous study setting allowed a participant to sign up with different email addresses. However, we do not expect any bias in favor of any group due to this reason because multiple registrations (if any) presumably were equally distributed among both study groups.

Moreover, it should be noted that a significant number of participants did not take part in the final follow-up surveys. Therefore, we included all relevant participant data in the multiple imputations to estimate their follow-up data and crosschecked those results with completer analyses and LOCF analyses which came to very similar results.

To elucidate the effects of the interventions more thoroughly, it would have been desirable to collect additional data on their acceptability and usability. However, to keep the questionnaires short and the respondent burden as low as possible, we did not include these outcomes in this survey. A future evaluation study will deal with this topic in detail.

We did not include a no-intervention control group; therefore, we cannot determine the de facto effectiveness of the intervention. A comparison with other RCTs on Web-based alcohol interventions can provide only rough hints whether the reduction of alcohol use in the Change Your Drinking intervention exceeds mere trial participation. The overall reduction of alcohol intake (d=0.59) is considerably stronger than the effects of no-intervention control groups in other trials [20,41] (the difference of weekend drinking between baseline and 30-day follow-up in the trial of Doumas et al [20] and the difference of typical weekly drinking between baseline and 3-month follow-up in the trial of Cunningham et al [41], where effect sizes below d=0.10 were achieved). However, other RCTs on Web-based alcohol interventions show that nontreated groups can attain considerable performances up to d=0.40 (the difference of drinks last week between baseline and 3-month follow-up in the trial of Blankers et al [14] and the difference of alcohol intake in the past 7 days between baseline and 3-month follow-up in the trial of Jonas et al [42]). So these comparisons only suggest that both versions of Change Your Drinking to be somewhat effective, which leads to no clear conclusion. Hence, we cannot say whether the alcohol use reductions found in both groups are a consequence of the program participation, response bias, regression to the mean, or spontaneous remission. Nevertheless, since the original version of the intervention is freely accessible on the study website, a no-intervention control group was not feasible.

The revised version of Change Your Drinking is not more effective in reducing alcohol use than the original version of the program. The extension of the program and its stronger elaboration presumably played a minor role in reducing alcohol consumption. However, analyses of the program usage suggest that it may be more attractive to participants. Acceptability and usability of both program versions will be examined in a future study.