As of January 1, 2014, the legal age to purchase alcohol in the Netherlands increased from 16 to 18 years [33], which had some implications for our study design. Originally, the baseline assessment was planned for October 2013 and the follow-up assessment for April 2014. However, to avoid the law change (and its potential impact on drinking behavior) taking place between the baseline and follow-up assessments, we delayed the baseline assessment until after the law change was in effect in January 2014. Furthermore, we decided on a 4-month follow-up assessment instead of a 6-month follow-up because a 6-month follow-up assessment would have fallen in the summer vacation period.

We conducted a cluster randomized controlled trial (RCT) (trial registration number: NTR4048), randomizing Dutch schools of either lower secondary education and lower vocational training or higher secondary education into an experimental and a control condition. The experimental condition received the online intervention in the form of a game that contained computer-tailored feedback. The control condition only filled in the online baseline questionnaire. Both groups were given an online follow-up assessment after 4 months, responding to the same questionnaire as used in the baseline assessment. The study took place in the Netherlands between January and June 2014.

Adolescents were recruited in schools. Information letters addressed to teachers and coordinators of the highest grades at secondary education schools (grades 4-6) and at vocational training schools were sent via postal mail. These information letters informed the teachers about the intervention and provided contact details and the address of the study website, so schools could obtain more information and subscribe to the study. All eligible schools in the Netherlands (approximately 600 schools) received an invitation. If schools did not respond, they were called 2 to 3 weeks later. Schools were randomly assigned to either the experimental or control condition after their consent to participate in the study. Schools were not blind to their condition because experimental schools had to plan 3 lessons for the intervention (one lesson for the baseline assessment and first game session, a second lesson for the second and third game session, and a third lesson for the 4-month follow-up assessment) and control schools had to plan 2 lessons (the baseline assessment and the 4-month follow-up).

Approximately 3 weeks before the intervention started, teachers were provided with a letter containing more information about the procedure, privacy, and confidentiality. All adolescents in the classes were provided with a letter on the day of the intervention to avoid them starting the intervention prematurely. The letter informed the adolescents that all their answers in this study would never be shared with teachers, parents, or any other third person; would be used for research purposes only; would be analyzed anonymously; and that they could end participation at any point in time. Adolescents were also made aware that, at the end of the study, they would participate in a lottery for 300 gift vouchers worth €25 each. Furthermore, adolescents were provided with a letter for their parents. In this letter, the parents were informed that their child participated in an online alcohol intervention at school and the parents were invited to visit a separate website specifically for parents, where they could take part in the parental component of the intervention. When starting with the intervention, teachers asked the adolescents to visit the study website and create an account. They were routed to the according condition (either control or experimental) based on the school they attended. Before starting with the baseline questionnaire, all adolescents had to check a box on the first page of the website that contained informed consent information confirming their informed consent. If they did not wish to participate or refused to provide informed consent, they could check an “I do not wish to participate in this study” box; they were thanked and could then close the intervention website (see Multimedia Appendix 1 for the CONSORT eHealth checklist filled out for this study).

Our main target group was adolescents aged 16 to 18 years. Because we were recruiting the adolescents in schools, we also included younger (15 years) and older (19 years) adolescents because they are often in the same class. Schools had to provide the adolescents with individual access to a computer with Internet connection.

The idea of a game instead of a purely text-based computer-tailored intervention was first brought up by adolescents during focus group interviews [32]. During the development of the intervention, all materials and questions concerning the game (eg, its name, screenshots and characters of the game, realistic scenarios after drinking too much alcohol, realistic advice for adolescents who are trying to drink less in tempting situations, layout and design of the first version of the game) were presented to a Facebook panel. This panel consisted of a convenience sample of 24 adolescents aged 16 to 18 years, who provided us with feedback on those materials. The feedback was used to adapt the game to match the desires of the target group as closely as possible. After the development was completed, the game was pilot-tested at 5 schools to test the feasibility of the recruitment strategy, the design, and the content of the intervention. In total, 481 adolescents played the first game session and provided us with feedback about appreciation, comprehension, attractiveness, and level of personalization of the game. They were also asked about what they liked and disliked about the game. Based on this pilot, we shortened the game and the feedback messages were shortened and rewritten by a professional writer to make them more appealing to our target group. Originally, only the first game session was offered in the school and the adolescents were asked to continue with the game at home. After reviewing the feedback and the pilot data, we decided to make some changes to the design and to offer all 3 game sessions within the school setting.

The intervention, Alcohol Alert, consisted of an online baseline questionnaire, after which the adolescents played 3 sessions of the game “What happened?!”. In these game sessions, the adolescent wakes up after a night of partying and does not remember what happened the night before. The goal of this 2-dimensional game was to find out what happened. Each of the game sessions depicted one of the most common drinking situations for adolescents (ie, drinking at home, drinking in a bar, drinking at a party). The sequence of the 3 game sessions was tailored and dependent on how many glasses of alcohol the adolescent indicated to typically drink in each of these situations. The adolescent started with the drinking situation in which he or she indicated drinking the most alcohol. Thus, if the adolescent indicated typically drinking 3 glasses at home, 5 glasses at a party, and 6 glasses in a bar, he or she would start with the bar scenario first, followed by the party scenario, and finally the home scenario. Each session started in the bedroom where the adolescent wakes up. The adolescent quickly discovers that something is wrong (eg, the wallet is missing in one session). The adolescent then navigates through different places in the game and talks to people he or she meets and gets clues about what happened the previous night (Figure 1).

During the game sessions, the adolescent received questions and feedback on an in-game cell phone (Figure 2). These questions and feedback were based on the I-Change Model [34], an integrated model based on theories such as the Attitude-Social Influence-Self-Efficacy Model [35], the Theory of Reasoned Action [36], the Theory of Planned Behavior [37], Social Cognitive Theory [38], the Health Believe Model [39], the Precaution Adoption Model [40], and the Transtheoretical Model [41]. The I-Change Model attempts to explain motivational and behavioral change and has been successfully used to design and evaluate health interventions previously [19,42,43]. The questions and computer-tailored feedback were based on the relevant concepts of the I-Change Model (ie, attitude, modeling, social norm, perceived pressure, and self-efficacy). Within the game, this was operationalized by presenting the in-game cell phone twice during every game session. During the first presentation of the first game session, the adolescents were asked questions about their attitude toward binge drinking and received immediate feedback to try to shift their attitude about binge drinking to be less positive. The first time the in-game cell phone was presented in the second scenario, questions about modeling of alcohol use and binge drinking were asked (ie, who in their family and friends engaged in binge drinking), and feedback was provided to help the adolescents to choose the right role models. The first time the cell phone was presented in the third scenario, questions concerning social norms (ie, if parents and friends approved of drinking) and perceived pressure (ie, whether the adolescents perceived pressure to binge drink from family or friends) were posed and the feedback messages tried to encourage adolescents to resist that pressure. The second time the cell phone was presented during each scenario, questions about situation-specific self-efficacy were posed (eg, in the bar scenario adolescents were asked how difficult it is for them not to binge drink in a bar). Feedback was provided to enhance self-efficacy and the adolescent was provided with action plans that he or she could use in that particular situation. We decided on this sequence based on the Ø pattern [44], which describes that people shift toward behavior change through first developing a favorable attitude, experiencing positive social influences, and finally developing high self-efficacy toward the behavior.

The content and methods used in the feedback messages varied depending on the message, but usually the answer of the respondent was repeated to enhance self-monitoring, correct assumptions were confirmed with positive feedback, and wrong assumptions were corrected with new information. All messages had a personal tone to show sympathy and to enhance commitment [45]. For example, the attitude questions assessed the pros (eg, “binge drinking helps me relax and connect easily with other people”) and cons (eg, “binge drinking makes me feel like I am losing control”) of binge drinking. The adolescent immediately received feedback on his or her overall attitude and for every pro and con specifically. In these feedback messages, the focus was on providing the participant with general (eg, “alcohol inhibits your brain’s natural inhibition system”) and personal (eg, “you might say or do things that you regret later”) consequences of alcohol to change attitude toward binge drinking to a more negative one. For more information about the content of the feedback messages, we refer to our study protocol [46]. Adolescents received 2 reminder emails to finish the game sessions if they did not do so at school; the first after 1 week and the second after 2 weeks. A week after the third game session, the adolescents were invited to revisit the intervention and received 2 reminder emails; the first after 1 week and the second after 2 weeks if they did not return. In this fourth session, which was not part of the game, alcohol use during the last week was assessed and the adolescents were provided with feedback about their use compared to Dutch drinking guidelines. Following this, the adolescents were asked if they had an event in the upcoming 30 days (eg, a party, wedding) where they usually drink 4 (for girls) / 5 (for boys) or more glasses of alcohol on such an occasion. If they responded positively, they were asked if they wanted to challenge themselves to drink less than they usually would. If they responded positively again, they were asked to indicate the date of the event and how many glasses they wanted to drink at most. They could then make their own action plans for how to achieve their goal or they could indicate from a list of action plans which one they would most likely follow to achieve their goal. If adolescents indicated that they had no event in the upcoming 30 days or that they did not wish to participate in the challenge, they only received advice on how action plans could help them to prevent binge drinking in the future. At the end of the fourth session, all adolescents were provided with the feedback they received during the game to boost their memory. One day before the drinking event, they were reminded by email that they accepted a challenge to drink less alcohol than they usually would for the event the next day. Two days after the event, they were invited to come back to the website to indicate if they met their goal. Reminder emails were sent after 1 and 2 weeks if they did not return. If they indicated they drank more than they had planned, they received feedback on how to do better next time and were given the opportunity to repeat the challenge. If they indicated that they had not exceeded their drinking maximum, they received congratulations and the intervention was over. For a detailed description of the development and the content of the intervention, we refer to the study protocol of this study [46].

After 4 months, the adolescents in both conditions responded to the online follow-up questionnaire in school. If they did not finish the follow-up assessment at school, they received 2 reminders to do so at 1 and 2 weeks after the official deadline for the schools.

A separate component was added to the intervention to involve parents. During the development, a convenience sample of 14 parents provided us with feedback on the layout, usability, and content of the parental component. At baseline, adolescents in the experimental condition were asked to enter the email address of one of their parents. Parents then received an email inviting them to a separate website, where parents responded to a short questionnaire and could also receive computer-tailored feedback on how to set appropriate rules concerning alcohol use and how to communicate with their child about alcohol use. If the adolescent did not know the email address of the parent or did not wish to send an email to the parent, they could refuse to do so. They were informed that the letter they received for their parents contained all the information about the parental component and instructions on how parents could participate. A detailed description of the parental component can be found in the study protocol [46].

This trial was reviewed by the Medical Ethics Committee of Atrium Orbis Zuyd and was classified as research that does not fall under the Medical Research Involving Human Subjects Act and needed no further approval (METC number: 12-N-104).

Figure 3 depicts a flowchart of the participating schools. In total, 44 schools were randomized into the experimental or control condition. Five schools in the control condition withdrew their participation before the baseline assessment started (2 schools of secondary higher education, 1 school of secondary lower education, 1 school of lower vocational training, and 1 school of secondary education mixed). Three schools in the control condition (all secondary higher education) and 2 schools in the experimental condition (1 lower vocational education, 1 higher secondary education) did not start with the baseline assessment and did not respond to our phone calls and emails. Most schools that dropped out before the intervention started indicated that they had logistical problems (eg, they had no computer room available to provide every adolescent with his or her own computer). Another school decided after randomization that the topic was too sensitive and they did not want to do that at school. In total, 2649 adolescents from 34 schools participated in the baseline questionnaire. The adolescents in the 2 conditions significantly differed from each other in various characteristics. Participants in the experimental condition were significantly younger, consisted of more females, had a higher educational level, more often indicated to be religious, and consisted of more participants who never drink, were less often binge and excessive drinkers, and had a lower weekly consumption than participants in the control condition (Table 2). Even though 27 schools participated in the 4-month follow-up questionnaire, only 824 of 2649 adolescents (response rate 31.11%) did so. Schools that withdrew participation at the follow-up assessment either reported trouble with finding a date due to the final exams of the classes or indicated that the adolescents were not keen to continue with the intervention and, therefore, the school decided to stop participation. Dropout analyses revealed that adolescents returning to the follow-up questionnaire were significantly younger, more often female, had a higher educational level, were more likely be religious, were more often Dutch, were less likely to be excessive drinkers, less likely to be binge drinkers, and had a lower weekly consumption (Table 3).

At the baseline assessment, 758 of 1622 (46.73%) adolescents in the experimental and 585 of 1027 (56.96%) adolescents in the control condition reported binge drinking in the previous 30 days. At the follow-up assessment, 194 of 456 (42.6%) adolescents in the experimental condition and 184 of 368 (50.0%) adolescents in the control condition reported binge drinking in the previous 30 days. The returning sample did not differ on baseline drinking characteristics. They did not differ on being a drinker (control: 274/368, 74.4%; experimental: 322/456, 70.6%; χ² ₁=1.3, P=.25) on binge drinking (control: 167/368, 45.4%; experimental: 203/456, 44.5%; χ² ₁=0.4, P=.83), on excessive drinking (control: 28/368, 7.6%; experimental: 22/456, 4.8%; χ² ₁=2.7, P=.10), nor on weekly consumption (control: mean 3.2, SD 6.9; experimental: mean 2.4, SD 6.1; t ₈₁₉=1.62, P=.11). There was a significant interaction effect between condition and age (P=.08) (Table 4). Age groups were analyzed separately using the pick-a-point approach [55] by centering the age variable for each year from ages 15 to 19. This way, the whole sample could be used to determine whether the intervention was effective for one or more of the age groups. Information about binge drinking prevalence at baseline and follow-up per age group are available in Table 4. Analyses revealed a significant effect of the intervention in 15-year-old adolescents (P=.03). Adolescents in the experimental group reported a significant decrease in binge drinking in the previous 30 days at 4 months after the intervention ended compared to adolescents in the control condition. Adolescents in the experimental group aged 16 years also engaged less in binge drinking after 4 months compared to the control group. This effect was not significant (OR 0.56, 95% CI 0.30-1.05, P=.07) (Table 5), but could be considered a small effect [56]. Furthermore, although participation of parents was very low (Table 2), when parents participated in the intervention, their participating child reported less binge drinking in the previous 30 days (P=.04). A higher educational level (P<.001), a lower age (P<.001), and being Protestant (P=.03), Muslim (P<.001), or a member of another religion (P=.03) (all analyzed in a model without interaction terms) were significant protective determinants of binge drinking (Table 5).

At the baseline assessment, 116 of 1622 (7.15%) adolescents in the experimental condition and 129 of 1027 (12.56%) adolescents in the control condition engaged in excessive drinking. At the follow-up assessment, 28 of 456 (6.1%) adolescents in the experimental and 37 of 368 (10.2%) adolescents in the control condition reported excessive drinking. There was a significant interaction effect between condition and educational level (P=.08). However, further analyses revealed no significant subgroup effects for either higher- or lower-educated adolescents (Table 5). Protective determinants of excessive drinking were being female (P<.001), a higher educational level (P=.01), and being younger (P<.001).

At baseline, adolescents in the experimental condition drank a mean of 3.4 (SD 8.9) standard glasses of alcohol in the previous week. Adolescents in the control condition drank a mean of 5.1 (SD 9.9) standard glasses of alcohol in the previous week. At the follow-up assessment, adolescents in the experimental condition reported a mean consumption of 3.3 (SD 7.7) and adolescents in the control condition reported a mean consumption of 4.6 (SD 8.9) standard glasses of alcohol during the previous week. Although the effects were in the expected direction, no significant effects of the intervention were found for weekly consumption. The analysis only revealed that being female (P<.001), having a higher educational level (P=.002), and being younger (P<.001) (all analyzed in a model without interaction terms) were significant determinants with a protective effect on weekly consumption (Table 5).

After the baseline assessment, adolescents in the intervention condition were supposed to start with the first game session. Of 1622 adolescents who were randomized into the experimental condition, only 1097 (67.63%) started with the first game session. Only 467 adolescents (28.79%) returned to the second and only 347 (21.39%) adolescents to the third game session. Just 27 (1.66%) adolescents returned to the fourth session at home, and zero participated in the fifth home session.

Subsequently, to investigate the effects of adherence, we decided to rerun the analyses with the subsample of the group that completed the different game sessions. We made 3 groups: the first group consisted of all adolescents who participated in the first game session, the second group consisted of adolescents who also participated in the second game session, and the third group did all 3 game sessions. Descriptive analyses of prevalence of binge drinking per age group and per adherence group can be found in the Multimedia Appendix 2. The effects for binge drinking are summarized in Table 6.

Again, we found a significant interaction effect with age for all 3 groups (P=.007 for adolescents that participated in at least 1 session; P<.001 for adolescents that participated in at least 2 sessions; P=.002 for adolescents that engaged in all 3 game sessions). When engaging in at least 1 session, only 15-year-old adolescents reported less binge drinking (P=.01). The effect sizes increased when 15-year-olds adhered longer to the intervention (after 1 session: OR 0.41, 95% CI 0.21-0.81; after 2 sessions: OR 0.24, 95% CI 0.11-0.57); after 3 sessions: OR 0.22, 95% CI 0.09-0.54). A similar pattern was found for 16-year-old adolescents. There was a significant effect of the intervention after 2 sessions (OR 0.43, 95% CI 0.20-0.95, P=.04) which became stronger after 3 sessions (OR 0.37, 95% CI 0.16-0.87, P=.02). There was no significant effect for older adolescents. The analyses for excessive drinking revealed a significant interaction effect between condition and educational level (OR 2.37, 95% CI 0.98-5.73, P=.05) for adolescents that adhered to at least 1 session. However, the subgroup effects for higher (OR 0.87, 95% CI 0.22-3.52, P=.85) and lower (OR 0.46, 95% CI 0.12-1.83, P=.27) educated adolescents were both not significant. Weekly consumption revealed a similar result with a significant interaction effect with educational level (β=–0.22, SE 1.27, P=.09) for adolescents that completed at least one session, but only small and nonsignificant subgroup effects for higher- (β=–0.19, SE 0.94, P=.84) and lower- (β=0.17, SE 2.68, P=.95) educated adolescents. Furthermore, there was a significant interaction effect between condition and age on weekly consumption for adolescents that completed at least 2 sessions (β=–0.99, SE 0.52, P=.05) and for those who completed at least 3 sessions (β=–1.03, SE 0.59, P=.08); however, although the effects were more positive for the younger age groups, no effect reached a significant level. Finally, significant predictors of adherence were being Protestant, being female, being younger, having a higher educational background, and being a nonbinge drinker (Table 7).