EFFEKT

Study 1

Evaluation Methodology

Design: The program seeks to be applicable in both rural and urban areas, so three types of communities (inner city, public housing, and small town) were selected for the initial test. Prior to the start of the project, the Social Medicine Unit of the County Hospital surveyed all 9th graders in Örebro County (n=3094) about alcohol use and other health-related behaviors. Data from this survey was used in the selection of schools. Because the intervention was supposed to work in different types of communities, the researchers first selected and approached intervention schools in the three types of communities mentioned above. All of the schools agreed to participate. Then, matched control schools were selected that were similar on community type, size of school, alcohol use and delinquency, but that were also distinct enough geographically to minimize the potential for cross-over effects. In each of the small towns (one intervention and one matched control), there were two schools. In each of the other community types there was one intervention or one matched control school per community. There was a total of 8 schools, including 38 classrooms (18 control, 20 experimental). A baseline assessment was conducted at the beginning of 7th grade and follow-ups during implementation were conducted near the end of the spring terms, when youths were in 8th and 9th grades.

The initial sample available consisted of 895 students (437 in the intervention group and 458 in the control group). Of those, 811 students participated in the evaluations in 7th grade (393 in the intervention group and 418 in the control group). At that initial assessment, 651 parents participated in evaluations (339 in the intervention group and 312 in the control group). At the 8th grade assessment, 768 students participated (382 in the intervention schools and 386 in the control schools). 662 parents participated in this assessment (316 in the intervention group and 346 in the control group). At the 9th grade assessment, 829 students participated in the evaluation (408 in the invention group and 421 in the control group). 708 parents participated in this assessment (367 in the intervention group and 341 in the control group).

There were 339 youth and 264 parents in the preventive condition with complete data at all three waves and 366 and 242 for the control condition.

Attrition Analysis: The researchers used logistic regression to examine whether gender, parental divorce, parents and youths' place of birth, parental education or any measures under study at baseline predicted leaving the longitudinal sample. They selected those who reported on drunkenness and delinquency at baseline, which amount to 99% of those who responded to the questionnaire at all. Then, they compared those who reported again at post-test with those who did not. Separate analyses were conducted for: adolescents in the control group, adolescents in the intervention group, parents in the control group, and parents in the intervention group. In the control group, youths who dropped out of the study reported more frequent drunkenness at baseline than those who stayed in the study. In both the intervention and control groups, parents who dropped out had more lenient attitudes towards youth drinking than those who stayed in.

Sample: For youths, participation was high, and rates did not differ between intervention and control schools. For parents, there were some unsystematic differences in participation between the intervention and control schools. The major reason for youth non-participation was absence from school due to sickness. Parents could disallow youths' participation and youths could decline participation, even if their parents did not object. In total, 60 parents or youths (6.7%) denied youth participation in grade 7, 129 (14%) did so in grade 8, and 92 (10.2%) did in grade 9. Parents who refused to participate themselves were distributed evenly across schools and conditions.

At baseline, parents in the intervention group were significantly less strict that those in the control group. Because the point of the intervention was to maintain strictness in the intervention group, this made the intervention more challenging. There were no differences between the groups on drunkenness, organized activity participation, delinquency or parent-reported education or ethnicity.

No information was provided about the gender or racial/ethnic makeup of either the student or parent participants.

Measures:

Parents' attitudes: Parents were asked which of four attitude descriptions fit them most accurately. They ranged from very lenient ("It is natural for children our son or daughter's age to be curious about trying alcohol. We trust that our son/daughter drinks in a responsible way") to very strict ("A child our son or daughter's age is way too young to drink alcohol at all. We think it is obvious that adolescents under 18 years should not concern themselves with alcohol").

Organized activity involvement: Youths were asked about their involvement in activities that took place in groups, had adult leaders, and met at least once a week. They indicated which of seven specific activities they were involved in and how many nights per week they attended each activity.

Adolescent drunkenness: At all three assessments, adolescents answered the question: "How many times during the last 4 weeks have you drunk beer, wine, or spirits to the point that you felt drunk?" They reported by writing a number on a blank line.

Delinquency: At each assessment, adolescents reported whether they had performed any of 20 delinquent acts during the last year (e.g. breaking into stores or cars, hurting someone with a weapon, painting graffiti, shoplifting, stealing a bike, vandalism). Items were rated on a three-point scale from "never" (1) to "several times" (3).

Analysis: To analyze changes in drunkenness and delinquency over time for the intervention and control groups, general linear modeling (GLM) was used. To examine gender effects, the researchers included interactions between gender and time, time and drunkenness (or delinquency). When multivariate effects were significant, univariate analyses of covariance (ANCOVAs) were used to examine post-test differences, controlling for baseline measures.

To test for group differences in sporadic and frequent drunkenness, chi-squared analyses were used. To determine whether the results might have been affected by differential attrition in the control and intervention groups, the researchers conducted the post-test ANCOVAs again after applying two separate data treatment strategies. The first limited the sample to those with data at both baseline and post-test and then deleted some cases randomly in order to match the numbers of heavy drinkers in the intervention and control groups at baseline. The second used the expectation-maximization function in SPSS to estimate post-test values for those who had participated at baseline but not post-test. The post-test values were estimated from information on all variables in the study at baseline.

A later analysis (Koutakis & Ozdemir, 2010) used latent grow curve modeling.

Outcomes

Parental attitudes became more permissive of underage drinking over 3 years. There was a significant group x time interaction. Parents in the intervention group kept their strict attitudes over time, while those in the control group became more permissive. Although there was a decrease in participation over time in organized activities, there were no differences between the intervention and control groups.

The analyses showed that youth drinking increased over time, but the increase was significantly steeper in the control group than in the intervention group. Univariate analyses showed more drunkenness in the control group than in the intervention group at post-test, controlling for baseline levels. Furthermore, the proportion of participants who had been drunk several times during the last month was twice as high in the control group as in the intervention group.

There was an increase in delinquency over time. This increase was steeper in the control group than in the intervention group. Univariate analyses showed significantly higher levels of delinquency in the control group than the intervention group at post-test, controlling for baseline levels.

Early Starters:
The researchers also tested the effects of the program on early starters, defined as those having been drunk at least at baseline (n=148) or being higher than the 80th percentile on delinquency at baseline (n=159). Drunkenness increased over time for early starters, but the increase was steeper in the control group than in the intervention group. Univariate analyses showed more drunkenness in the control group than in the intervention group at post-test, controlling for baseline levels. For early starters in delinquency, adolescents in the control group increased more over time than those in the intervention group. Univariate analyses showed more delinquency in the control group than in the intervention group at post-test, controlling for baseline levels. All outcomes were significant.

Findings Using Latent Growth Curve Modeling (Koutakis & Ozdemir 2010):
Parent attitudes remained unchanged over time in the preventive condition, while parents in the control condition gradually became more lenient towards underage drinking. Although youth in both conditions increased in drunkenness and delinquency over time, the increase was about twice as high in the control condition as in the experimental condition.

Study 2

Evaluation Methodology

Design: From a list of Dutch high schools, 80 schools randomly were selected. An independent statistician assigned 19 schools randomly to one of the four conditions: (1) parent intervention; (2) student intervention; (3) combined student-parent intervention; and (4) control condition (business as usual). Randomization was carried out centrally, using a blocked randomization scheme (block size 5) stratified by level of education, with the schools as units of randomization. Within each participating school, all first-year students participated in the intervention.

The baseline data were collected immediately after allocation at the beginning of the first year in high school (September/October 2006), before any intervention was carried out, and again 34 months later in May/June 2009. Adolescent digital questionnaires were administered in the classroom by trained research assistants. Students who were not willing to participate were free to refuse participation on the day the questionnaires were administered. Parental questionnaires were sent to parents' home address along with a letter of consent at baseline. This letter informed parents about the participation of the school in the project, and parents were given the opportunity to refuse participation of their child (0.01% refusal). Nonresponding parents were reminded after 3 weeks by a letter and after another 2 weeks by phone.

Nineteen schools, including 3490 adolescents, were selected to participate in the study. Because of initial nonresponse (n =122) and exclusion of adolescents who already drank weekly at baseline (n =306) or who responded inconsistently on the quantity-frequency items (indicated 1 or more drinks and zero on the number of days or vice versa) measuring weekly drinking (n =125), 2937 adolescents were eligible for analyses. Intention-to-treat analyses were based on 2937 students not manifesting (heavy) weekly drinking at baseline.

Parent intervention: This intervention targets parental rules for their children's alcohol use. The intervention was modeled after a Swedish intervention, the Örebro Prevention program. The intervention was carried out at the first parents meeting at the beginning of each school year (September/October 2006 and 2007), in which also other school-related topics were discussed. The intervention consisted of three elements: 1) a brief presentation (20 minutes); 2) consensus building among a shared set of rules among parents of children of the same class; and 3) an information leaflet with a summary of the presentation and the outcome of the class meeting was sent to the parents' home addresses.

Student intervention: The student intervention is the renewed digital alcohol module of the Dutch prevention program, The Healthy School and Drugs (HSD). The alcohol module targets the students' abilities to develop a healthy attitude toward alcohol use and to train their refusal skills. After receiving training, the teachers conducted the intervention (four lessons) in all first-year classes in March/April 2007. A booster session was provided 1 year later in March/April 2008.

Combined intervention: Schools in this condition carried out both the parent and student intervention.

Control condition: Schools in the control condition were contracted not to start any alcohol-related interventions throughout the study period. However, because basic information about alcohol use is part of the standard curriculum in the Netherlands, schools were allowed to continue this practice (business-as-usual).

Sample: The final sample (n =2937) is characterized by an average age of 12.6 years (SD=0.49) at baseline, consisting of 51% boys and 40% in lower secondary vocational education.

Measures: Dichotomous variables were computed indicative of the onset of (heavy) weekly drinking in those who did not report drinking weekly at baseline. The primary and secondary outcomes of interest were onset of heavy weekly and weekly alcohol use, respectively.

Heavy weekly drinking was measured by asking how many glasses of alcohol the student usually drank on a weekend day. In accordance with the definition of heavy drinking in adults, separate outcome variables for boys and girls were used. Boys drinking at least 5 glasses and girls drinking at least 4 glasses every week were considered to be heavy drinkers. The scale was recoded into a dichotomous variable, with 0=no heavy weekly drinking and 1=heavy weekly drinking.

Weekly alcohol use was defined by the Quantity-Frequency measure. The scale was recoded into 0=no weekly user and 1=weekly user, if at least one glass of alcohol was consumed on a weekly basis. In addition, the quantity-frequency was set to zero if adolescents reported not to have drunk alcohol in the previous month (analysis where the quantity-frequency measure was left unchanged revealed the same results). Onset of (heavy) weekly alcohol use was defined if students who were not weekly drinkers at baseline became (heavy) weekly drinkers at follow-up. Self-report measures of adolescents on alcohol use have proven to be reliable and valid methods to measure alcohol use.

Analysis: Data were analyzed (Mplus, version 6.0) in accordance with the intent-to-treat principle. No missing data appeared on confounders because of zero nonresponse on the item level for the adolescents. Missing data on the dependent variables were handled by using full information maximum likelihood (FIML). An earlier report showed that the randomization resulted in a slightly uneven distribution across the active conditions compared to the control condition in terms of age, gender, and level of education. Therefore, all subsequent analyses were conducted with these variables as covariates to control for any possible bias stemming from the imbalance.

Non-independence of observations due to cluster sampling-students were "nested" in classes-was taken into account by obtaining standard errors as implemented in Mplus. The cluster effect was corrected for at the class level, as the interventions were carried out in classes and higher intraclass correlations were found at the class level compared to the school level. The intraclass correlations were calculated without adjustment for confounders in an intercept-only model.

To examine the effect of the interventions on the incidence of heavy weekly and weekly drinking, two models were tested. In the first model, each of the experimental conditions was compared with the control condition. Odds Ratios of weekly (heavy) drinking were obtained using logistic regression of the binary outcome on the treatment dummies (experimental versus control), while adjusting for the confounders and the nested data. This model informs about the incidence of alcohol use at wave 3 compared to baseline. In the second model, alcohol use at previous time points was added to the first model so that an autoregressive logistic model was tested. This model shows the stability or additional effect at Wave 3 compared to the previous waves. Number needed to treat (NNT) represents the number of students who need to receive the intervention rather than its alternative in order to avoid one adverse outcome. NNT was obtained as the inverse of the risk difference.

Outcomes:
Baseline equivalence and attrition: At baseline, the intervention conditions differed significantly from the control condition with respect to the number of boys/men and of adolescents with a low level of education, where significantly more men were in the combined intervention group than all other groups and the control group members had significantly lower levels of education.

A total of 2533 adolescents (86.2%) stayed in the program and completed the follow-up assessment after 34 months. Attrition analyses on demographic variables and alcohol use indicated that participating adolescents were more likely to be younger, were more often in lower education, and drank a lower average number of alcohol beverages per week at baseline. Attrition was unrelated to conditions.

Intervention effects: At follow-up, significantly fewer students in the combined intervention had started to drink heavily on a weekly basis compared to the control condition. No significant effects of either the parent intervention or student intervention were found on the incidences of heavy weekly drinking. This finding is replicated when previous alcohol use was included in the analysis. So, when parents and adolescents are targeted simultaneously, the proportion of heavy weekly drinking adolescents is reduced.

Significantly fewer students in the combined intervention had started to drink on a weekly basis relative to the control condition at follow-up. No significant effects of either the parent intervention or student intervention were found. Again, these findings are in line with the autoregressive analysis. So, the combined intervention can effectively reduce the proportion of weekly drinking adolescents.

All schools implemented the interventions as stipulated in the protocol; only some small differences were found in the way consensus among parents was strived for; five schools (of 10) did not try to reach consensus among the parents about the rules for adolescent drinking while being in the meeting with the mentor, but instead distributed copies of a preprinted list of plausible rules, for the parents to take home. Parents returned the lists to the mentor after having chosen the rules they complied with. A summary of these lists was used to compose the leaflet sent to parents afterwards, in order to notify all parents about the chosen set of rules. No other differences with respect to the parent intervention were found. No differences in the implementation of the student intervention were found.

Mediating Effects: None examined.

Effect Size: Odds ratios for heavy weekly drinking and weekly drinking ranged from .59 to .88 - small to small medium effect sizes.

Limitations:

• Generalizability is limited because the current findings may not reflect the situation in other drinking cultures.
• Because of the exclusion of weekly drinking adolescents at baseline, necessary for analyzing incidence, the current findings apply only to adolescents who were not involved in weekly drinking at age 12 years. As this percentage was rather large (87.3%), the current results can be considered applicable to the large majority of students at this age.
• Only 25% of all schools that were approached actually participated.
• Some drop-out occurred, specifically among older students and those in lower-level types of education.

Study 3

This study was designed to provide an independent trial of the Orebro program (ÖPP), testing the hypothesis that children of parents exposed to ÖPP would report lower frequencies and later initiation of drunkenness than children of non-exposed parents. However, this trial does not appear to actually replicate either of the previous studies included here. First, Study 1 involved only parents and not students. Additionally, while Study 2 includes both parent and student elements as seen in this study, parents were only given 4 presentations while in the current independent trial up to six presentations were given.

Evaluation Methodology

Design : The study was implemented between 2007 and 2010 as a cluster-randomized trial, with measurements prior to randomization and 12 and 30 months post-baseline. It appears that the 12 month follow-up acted as a post-test, and the 30 month follow-up was completed 18 months after the intervention was completed. The program format required the allocation of schools rather than individuals, thus schools were considered the randomization unit. Schools were assigned to the intervention or to a control condition which allowed prevention activities other than ÖPP (i.e., business as usual).

The study invitation, which was e-mailed to middle school principals, described ÖPP, the study objectives and conditions for participation, and the randomization procedure. Eligible schools were municipal schools with students ages 13-16 years and no previous exposure to the ÖPP. Study logistics required that only schools with experienced local ÖPP presenters within a travelling distance could be invited. After a redesigning of the sampling criteria due to lack of responses from schools, study information was mailed to 716 schools in 13 counties, of which 40 schools (6%) volunteered to participate. In each school, two to three classes (∼50 students) of the school's own choice were measured. Schools were stratified by size and the proportion of graduating students, with the latter intended as a socio-economic proxy. Randomization generated 20 intervention schools (46 classes, 893 youth) and 20 control schools (41 classes, 859 youth).

Youth questionnaires were classroom-administered while parents' questionnaires were posted. Teachers were instructed on how to administer the baseline questionnaires (pre-randomization, i.e. blinded assessors), while the research team administered the follow-ups (non-blinded assessors). The current version of ÖPP encompasses six short (∼20 minutes) standardized Power-Point presentations, administered by trained program presenters to parents of 13-16-year-olds at regular teacher-parent meetings during each term in grades 7-9. The presentations, which show a minimal variation in contents, advise parents to adopt and maintain strict attitudes towards youth drinking, and to not allow their adolescents to have a sip or drink of alcohol at home. Parents were also encouraged to make written agreements on how to prevent their youth from drinking. A summary and the class agreement were subsequently mailed to all parents in the class. The program was delivered by 34 experienced ÖPP presenters, of whom 23 (68%) were also authorized ÖPP trainers. Presenters were instructed to adhere carefully to the present program version during administrations.

Sample : Schools were located in diverse municipality types, with intervention and controls distributed fairly evenly within each type. The grade point average (range 0-320) was slightly higher for participating schools compared to the national rates, 208.5 versus 205.3. The baseline measurement in the 7th grade encompassed 1752 students, and follow-up rates in the 8th (T2) and 9th (T3) grades were 1613 (92.1%) and 1548 (88.4%), respectively. Parent's response rates were 1314 (75.0%), 1227 (70.0%) and 1184 (67.6%) at the three assessments. Dropouts had changed schools or were absent on days measures were taken.

Measures : The primary outcome was drunkenness frequency. Secondary outcomes were onset of (i.e. life-time) drunkenness and alcohol consumption. No data on reliability or validity of any of the measures used was given.

Primary Outcomes: The drunkenness outcomes were measured by the items: 'How many times /during the last four weeks/ have you drank alcohol to the point that you felt drunk?'. Response categories were 0, 1, 2, 3-4, 5-7 and >8 times for drunkenness frequency and 0, 1, 2-4, 5-10, 11-20 and >20 for life-time drunkenness. Responses to the drunkenness frequency item were dichotomized into frequent versus no frequent drunkenness during the last 4 weeks (≥2 times ≤1 time). Responses to the life-time item were dichotomized to enable the study of program effects on drunkenness onset.

Alcohol consumption was measured with a beverage-specific frequency by quantity measure regularly used in Swedish nation-wide school surveys. Respondents reporting at least one drink of alcohol (can of beer, glass of wine, etc.), at least weekly, were coded as weekly drinkers (versus non-weekly drinkers).

Secondary Outcomes: To measure parent's attitudes towards youth drinking, parents chose among four statements, from the lenient: 'It is natural for children our son or daughter's age to be curious about trying alcohol. We trust that our son/daughter drinks in a responsible way' (1) to the strict: 'A child our son or daughter's age is way too young to drink alcohol at all. We think it is obvious that adolescents under 18 years should not concern themselves with alcohol' (4). Youth were asked whether they were served alcohol at home with the response alternatives: 0 = 'No, my parents do not drink alcohol', 1 = 'No, never', 2 = 'Yes, sometimes I may have a sip from my parents glasses', 3 = 'Yes, sometimes I can have a glass of alcohol' and 4 = 'Yes, sometimes I can have a bottle of wine or a number of beers'. Responses were dichotomized (0-1 = 0, 2-4 = 1).

At follow-up measurements, teachers were asked to indicate which among 40 listed prevention programs had been used in the participating classes. Parents were asked whether an adult person in the household had participated in any of 10 listed prevention programs on behalf of any of their children.

Analysis : Four sets of analyses, or scenarios, were performed for each outcome, treating missing values as missing (i.e. completers only), as negatives (0), positives (1) and replaced by the Multiple Imputation procedure in SPSS. Parent's reports on restrictive attitudes were not imputed due to the high dropout rates (>30%). Inconsistent reports on life-time drunkenness (0 = no, 1 = yes) were re-coded if similar responses were provided on two consecutive measurements; that is, 1/0/0 were coded into 0/0/0 (n = 9) and 1/1/0 into 1/1/1 (n = 5). Remaining inconsistencies (0/1/0) were treated as missing at follow-up (n = 27). One-way analysis of variance (ANOVA), Pearson's χ² test and Fisher's exact test were used to compare groups on baseline characteristics. The binary outcomes were modeled as two-level logistic regressions of condition (ÖPP/control), incorporating school as a random effect and adjusting for the life-time drunkenness status at baseline.

Post-hoc examinations of dose-response relationships in the intervention group were performed, controlling for life-time drunkenness at baseline.

Outcomes

Implementation fidelity : Attempts to monitor fidelity were unsuccessful, because many presenters had technical problems or felt uncomfortable with recording devices during administrations. However, the authors claimed that use of experienced presenters of whom the majority were authorized trainers, and an adequate dosage, suggest that program quality did not fall below the one in real-world settings. The mean number of ÖPP presentations was 4.7 (out of 6) and numbers varied between three (n = 3), four (n = 5), five (n = 8) and six (n = 4) in the 20 program schools.

Baseline equivalence: There were no significant differences between conditions on background measures, while a near-significant difference for life-time drunkenness was observed (P = 0.06). The proportions reporting abstaining parents were similar in the ÖPP and control groups at T1 (7.1 versus 6.2%; P = 0.43) and T2 (7.1 versus 6.0%; P = 0.44), but they were more different at T3 (8.2 versus 6.2, P = 0.09).

Attrition : Life-time drunkenness at T1 was reported more frequently among dropouts than among responders at T2 (19.6% versus 11.4%) and at T3 (24.6% versus 10.4%). Assessment of differential attrition indicated significantly higher dropout rates in the control group at T2 (9.4 versus 6.5%) and a non-significant difference at T3 (12.7 versus 10.6%).

Primary Outcomes: The results were generally weak, with a statistically significant program effect found under only one imputation scenario for one of three drinking outcomes, which was present at T2 (12 months post baseline) but not at T3 (30 months post baseline). At T2, there was a statistically significant program effect on frequent drunkenness under the worst-case missing data scenario (P < 0.02); under remaining missing data scenarios P ranged between 0.07 and 0.10. At T3, no program effect was observed. There were no significant program effects on life-time drunkenness or weekly drinking at T2 or T3.

S econdary Outcomes: There was a statistically significant program effect at T2 on youth-reported alcohol servings at home under the worst-case missing data scenario (P < 0.03), and under all four missing data scenarios at T3 (P range 0.01-0.04). Among follow-up responders, ÖPP parents reported more restrictive attitudes than control parents at T2 and T3 (P < 0.01).

Dose-Response Analysis : Analyses in the ÖPP group revealed no relationship between the number of program presentations and frequent drunkenness, life-time drunkenness, or weekly drinking.

Effect Size: Odds ratios are reported but all are small, with significant odds ratios around .7.

Mediation Effects: The results showed that parents in the program had more restrictive attitudes toward alcohol use than parents in the control group. Changes in parental attitudes can be seen as be seen as mediating between the intervention and the alcohol use.

Limitations:

• Most results are highly sensitive to the method of assigning missing data.
• There is a susceptibility of self-reported alcohol consumption to both over and under-reporting.
• Parents had high dropout rates (>30%) and students showed differential attrition between groups where drop-outs had higher rates of life-time drunkenness.
• The generalizability of findings may be questioned on the basis of recruitment difficulties where an initial low response rate forced researchers to alter their methodology since many principals had simply ignored the invitation to participate. Even then, only 6% of principals who were approached by the research team agreed to participate.
• Schools were randomized, but classrooms that participated in the program were selected by individual schools and not the research team.
• Despite attempts, fidelity was not appropriately measured and efforts to do a dose-response analysis show no relationship between the number of presentations and outcomes.

Reanalysis of Data from Bodin & Strandberg (Ozdemir & Stattin, 2012)

The data was reanalyzed using the original metric of the variables (frequency measures), rather than the dichotomized measures used by Bodin & Strandberg. These analyses were complemented by tests of the mediating mechanisms. Latent growth modelling was used, taking into account the clustering of the data. Tests on four undichotomized measures: lifetime drunkenness, drunkenness in past month, frequency of drinking, and amount of drinking on each occasion were conducted. There were significant positive intervention effects found for lifetime drunkenness and a marginally significant effect on past-month drunkenness. Using dichomotized measures, there were no significant effects.

The test of mediation first showed a significant effect of the program on parental attitudes towards youth drinking. Next, the program had significant indirect effects on all four drinking measures through its effect on parents' attitudes toward youth drinking.