Steps Towards Alcohol Misuse Prevention Programme (STAMPP): a school-based and community-based cluster randomised controlled trial

Objectives To assess the effectiveness of a combined classroom curriculum and parental intervention (the Steps Towards Alcohol Misuse Prevention Programme (STAMPP)), compared with alcohol education as normal (EAN), in reducing self-reported heavy episodic drinking (HED) and alcohol-related harms (ARHs) in adolescents. Setting 105 high schools in Northern Ireland (NI) and in Scotland. Participants Schools were stratified by free school meal provision. Schools in NI were also stratified by school type (male/female/coeducational). Eligible students were in school year 8/S1 (aged 11–12 years) at baseline (June 2012). Intervention A classroom-based alcohol education intervention, coupled with a brief alcohol intervention for parents/carers. Primary outcomes (1) The prevalence of self-reported HED in the previous 30 days and (2) the number of self-reported ARHs in the previous 6 months. Outcomes were assessed using two-level random intercepts models (logistic regression for HED and negative binomial for number of ARHs). Results At 33 months, data were available for 5160 intervention and 5073 control students (HED outcome), and 5234 and 5146 students (ARH outcome), respectively. Of those who completed a questionnaire at either baseline or 12 months (n=12 738), 10 405 also completed the questionnaire at 33 months (81.7%). Fewer students in the intervention group reported HED compared with EAN (17%vs26%; OR=0.60, 95% CI 0.49 to 0.73), with no significant difference in the number of self-reported ARHs (incident rate ratio=0.92, 95% CI 0.78 to 1.05). Although the classroom component was largely delivered as intended, there was low uptake of the parental component. There were no reported adverse effects. Conclusions Results suggest that STAMPP could be an effective programme to reduce HED prevalence. While there was no significant reduction in ARH, it is plausible that effects on harms would manifest later. Trial registration number ISRCTN47028486; Post-results.


FULL PRIMARY OUTCOME MODELS
For reasons of space, the full primary outcome models were not presented in the main text.
Table S1 presents the parameter estimates from a two level random intercepts logistic regression model for the heavy episodic drinking (HED) primary outcome at T3.  Table S2 gives the parameter estimates from a two level random intercepts negative binomial model for the drinking harms primary outcome at T3.

SECONDARY OUTCOMES
A range of secondary outcomes were also examined within the study. These included the primary outcomes assessed at T2: Heavy episodic drinking (HED) (T2): Self-reported alcohol use defined as self-reported consumption of >5 drinks, assessed at +24 months (T2) from baseline. This was dichotomised at none/one or more occasions. This outcome was assessed via a two level random intercepts logistic regression model. Around 12.4% of respondents reported HED at T2 using this measure. In the intervention arm HED was reported by 10.9% (N=573) and in the control arm by 13.9% (N=722).

Alcohol related harms (T2):
The number of self-reported harms (harms caused by own drinking) assessed at +24 months (T2) from baseline. Items included harms such as getting into a physical fight or being sick after drinking. The outcome was a count of the number of discrete harms reported (0-16) and was assessed by a two level random intercepts negative binomial model. In the intervention arm 74.3% reported no drinking harms, while in the control arm 71.5% reported no harms.
In addition, a number of secondary outcomes at T3 and T2 were also examined, including: Lifetime drinking (T3): Whether the pupils had ever consumed a full drink of alcohol at +33 months (T3) (two level random intercepts logistic regression model).
Last year drinking (T3): Whether the pupils had consumed a full drink of alcohol in the last year, assessed at +33 months (T3) (two level random intercepts logistic regression model).
Last month Drinking (T3): Whether the pupils had consumed a full drink of alcohol in the last month, assessed at +33 months (T3) (two level random intercepts logistic regression model).
Harm from others (T3 and T2): The number of self-reported harms experienced that were the result of other people's drinking, assessed at both +33 months (T3) and +24 months (T2) from baseline (two level random intercepts negative binomial models). Harms included being hit or having property damaged by someone who had been drinking.
Age of onset (T3 and T2): Self-reported age at which respondent first consumed a full drink, assessed at both +33 months (T3) and +24 months (T2) from baseline (two level random intercepts Cox regression model).

Unsupervised drinking (T3 and T2):
Whether the pupils were permitted, by their parents(s), to consume alcohol (with small group of friends or at parties) with no adult present, assessed at both +33 months (T3) and +24 months (T2) from baseline (two level random intercepts logistic regression model).

Number of drinks consumed (T3 and T2):
Pupils were asked whether they usually drank from a range of different alcohol drinks (beer, alcopops, spirits cider, wine, Buckfast [a popular brand of fortified wine, with caffeine], others) and if so, how much did they usually drink. The values for each drink were summed together to give a total. As the underlying items continued decimals the total value was multiplied by 10 to create whole numbers.
The secondary outcome analysis also included covariates at level 1 (individual) and level 2 (school) where appropriate: The models use for the secondary outcome were similar to those employed in the primary outcome analysis with a single level one covariate, and the treatment indicator and stratification variables used in the randomisation as level two covariates. School type (Randomisation stratification factor): Given the larger number of schools in Northern Ireland, an additional stratification factor was used in the randomisation. This was school type (all boys' school/ all girls' school/coeducation school). Schools in Glasgow/Inverclyde were all assigned to the co-education type. This indicator was used represented by two dummy variables (co-education was the comparison category).

Level 1 covariate
Location: A dummy variable was generated to indicate the location of the schools (Northern Ireland/Scotland). Table S3 presents the random intercept models for the primary outcomes at +24 months. The baseline measures were significant, as was location. For the HED outcomes both free school meals (tertile split) and school type were significant. The intervention arm was significant at a 0.05 level (β=-0.241; p=0.041). However, it failed to reach the much stricter threshold used in the primary analysis (0.025). It should be noted that the HED indicator used at +33 months, and as specified in the DAP, was different that that used at +24 months. In particular, this measure did not use gender specific splits, referred to drinks rather than units, and did not provide any visual guides to help with the estimation of amount consumed. This suggests that the significant intervention effect may have been partly dependent on the precision of the measurement instrument used to collect the primary outcome data. The age at which differences in HED were assessed may have been important when assessing intervention outcomes.  Table S4 presents the outcome models for the additional secondary outcomes assessed at T3.

Results from the analysis of secondary outcomes
The treatment indicator was not significant in any of these models.  Table S5 presents the models for the secondary outcomes assessed at T2. Again, the treatment indicator was not significant in any of these models.

Subgroup analyses
To explore differential treatment effects on the primary and secondary outcome measures, prespecified interaction terms were fitted between trial arm and baseline measures thought to predict the effect of treatment. These were:  Age, in months, of pupil at baseline;  Gender;  Socioeconomic status (using the proportion of free school meals indicator);  Alcohol use behaviour at baselineever use, last year use, age of onset, and context of use (abstainer/supervised/unsupervised);  and in NI, a Grammar/Secondary school analysis.