Description of included studies
Nine publications reporting on seven prospective cohort studies were identified that met the review inclusion criteria [14–22]. The seven studies provided data on 13,255 participants aged 10 to 26 years old. Characteristics of the included studies are shown in Table 2 (see Additional file 2). Five were conducted in the USA [16–19, 21], one in Belgium  and one in New Zealand [14, 15, 22]. In one study  the cohort was part of an RCT of a school-based drug prevention programme, and in another  the cohort was a sub-set of a larger cohort study recruited in 1972 and followed through childhood to early adulthood evaluating growth and development.
The age of participants at baseline interview was 12 to 13 years (7th grade) in three studies [15, 16, 18], 14 to 15 years (9th grade) in one , one study  recruited a broader age group of youth, 15 to 26 year olds, one  used a mixed age group of first (aged 11 to 12 years) and fourth year (aged 14 to 15 years) secondary school students and one  used 10 to 14 year olds (5th to 8th grade).
In five studies participants were followed up once after baseline. Time to follow-up was one year [18, 20], 18 months , 30 months  and 13 to 26 months . One study reported outcomes at multiple time-points, six years and nine years and 14 years [14, 15, 22]. One study evaluated participants at four time points and present results for follow-up after 21 months taking the multiple time points into account in the analysis .
Each study used disparate measures of exposure; all relied on self-reported measures. One generated a composite score to reflect the amount of exposure to TV beer advertising, magazine alcohol advertising, beer concession stands and in-store advertising displays . One measured exposure to any alcohol advertising in the past month on each of four media, TV, radio, billboards and magazines . Another classified exposure as watched TV show index to quantify exposure to alcohol ads in specific TV shows in addition to self-reported exposure to alcohol ads . In the study by Connolly  recall of alcohol advertisements from different media, TV, radio, magazines, newspapers and films was evaluated. Two studies measured exposure as hours of TV and music video viewing [19, 20], and one exposure to alcohol use in popular movies .
Drinking status was measured in all studies at follow-up. Two studies reported any alcohol use in the past month [17, 18], one study reported alcohol use in the past year , one reported frequency of drinking at specific locations and average and maximum amount alcohol consumed on an occasion , one reported lifetime and past 30 days alcohol use , one alcohol use whilst going out , and one incident alcohol use without parental knowledge .
One study used a random sample of youth  three randomly selected schools and all participants at those schools were invited to participate [18, 20, 21]; in one study  all participants at six schools were eligible to participate but how schools were selected was not described; one study used the original sample of participants selected for participation in an RCT but excluded those with missing data ; and one study consisted of a sub-sample of children who had exposure and outcome data available at all follow-up periods .
Ascertainment of exposure and outcome data were by self-reported questionnaires in four studies [16, 18–20], by face-to-face interview in one  and computer-aided telephone interview in two [17, 21]. None of the studies explicitly reported that interviewers were unaware of the exposure status of participants when outcome assessments were conducted, however with participants independently reporting drinking outcomes via self-reported questionnaires there is little scope for detection bias on the part of the investigators. Not all children were non-drinkers at baseline. Two studies reported results for baseline drinkers and non-drinkers separately [16, 19].
All studies suffered, to a greater or lesser extent, from potential attrition bias. Attrition rates were 33%  and 69%  in two surveys where data were collected by telephone; 18% , 25% , 39%  and 36%  in surveys conducted in schools, and 35%  for the survey with face-to-face interviews and questionnaires.
One study used imputation to account for missing data ; all other studies excluded participants with missing data from the analyses.
Statistical adjustments for measured confounding factors were performed by each study, but the number and type of confounders varied from study to study. The results of the overall quality assessment of each study are shown in Table 3 (Additional file 3).
Connolly  investigated the relationship between alcohol consumption at 18 and alcohol-related mass media communications recalled at ages 13 and 15 years in a New Zealand cohort of young people. Among men, those who recalled more alcohol advertisements at age 15 drank significantly more beer at 18 years (average amount of beer consumption p = 0.047; maximum amount of beer consumption p = 0.008). In women a negative association of alcohol advertisement recall at age 13 years and frequency of drinking beer was found (p = 0.029). Multi-variate analyses were adjusted for potential confounders which were: media exposure, gender, current occupation, living situation, socio-economic status and peer approval of drinking. There was no significant effect on wine or spirit consumption in either women or men. Whilst significant relationships were detected, we cannot rule out the possibility they occurred due to chance. The authors reported results for more than 35 statistical tests and significant findings would be expected to occur due to chance. This coupled with the small sample sizes, 251 men and 184 women, cast some doubt on these findings being true effects. Longer follow-up from this same sample at age 21 and 26 years have been published [14, 22]. In the group that were beer drinkers at 18 years, liking of alcohol advertising and brand allegiance had a positive impact on beer consumed at age 21 years; standardised coefficients were 0.26 and 0.36, respectively. At 26 years, those showing a liking for alcohol advertising at 18 years were more likely to be in a group of heavier drinkers.
Stacy  assessed the impact of exposure to TV alcohol advertisements on alcohol use in 2,250 12 to 13 years old school children in California followed up for a year. At baseline, 16% reported drinking beer in the past month, 15% reported drinking wine in the past month, and 8% reported three-drink episodes in the past month. At follow-up, prevalence was 18% for beer, 20% for wine and 12% for three-drink episodes. At one-year follow-up, each standard deviation increase in TV viewing of programmes with alcohol advertisements at baseline was associated with a significant increase (44%) in risk of beer use ((odds ratio (OR) 1.44 95% Confidence Interval (CI): 1.27 to 1.61)), wine/liquor use (OR 1.34; 95% CI: 1.17 to 1.52) and three-drink episodes (OR 1.26; 95% CI: 1.08 to 1.48), controlling for general TV viewing frequency, participation in team sports, perception of peer alcohol use, perceived peer approval of alcohol use, intentions to use alcohol, perceptions of adults alcohol use, gender, ethnicity and school, exposure memory covariates and a propensity score to adjust for differential risk profile of those lost to attrition. A watched TV sports index was only positively associated with beer drinking, (OR 1.20; 95% CI: 1.05 to 1.37) with adjustment for confounders, and self-reported frequency of exposure was significantly associated with increased risk of beer drinking, (OR 1.21; 95% CI: 1.14 to 1.41). Other exposure measures, cued-recall memory test and draw-an-event memory test, did not show significant relationships with any of the outcomes, though most showed effects in the direction of positive associations with one exception, participants scoring one standard deviation above the mean for draw-an-event memory test were significantly less likely to drink beer one year later (OR 1.14; 95% CI: 1.01 to 1.25).
Ellickson  examined the relationship between a range of advertisement exposures and subsequent drinking among US adolescents age 12 to 13 years. Forty-eight per cent non-drinkers at baseline (n = 1,905) initiated drinking by two-year follow-up. For baseline non-drinkers, exposure to in-store beer displays predicted drinking onset at follow-up, OR 1.42 (p < 0.05) adjusted for general TV viewing, social influences, social bonds, gender, ethnicity and attitudes and behaviour. Exposure to TV beer advertisements, magazines with alcohol advertisements, and in-store advertisement displays all showed positive associations, though none were significant in adjusted analyses, OR 1.05, 1.12 and 1.06, respectively. Confidence intervals were not reported for any of the ORs. Among baseline drinkers (n = 1,206), 77% reported alcohol use in the past year at follow-up. Exposure to magazines with alcohol advertisements and to beer concession stands at sports or music events predicted frequency of drinking at follow-up, regression coefficient 0.10 and 0.09, (p-value < 0.05), respectively. Exposure to TV beer advertising or in-store advertisement displays were not significant predictors of drinking frequency in analyses adjusted for baseline drinking and multiple control variables regression coefficient -0.01 and 0.02, respectively.
Snyder  evaluated the relationship between self-reported advertising exposure to four media (TV, radio, billboards and magazines) and the prevalence of advertising in the same media sources and alcohol consumption in 15 to 26 year olds in 24 media markets in USA. Participants were followed up at four time-points over a 21 month period. Sixty-one per cent had at least one drink in the past month at baseline and consumed an average of 38.5 drinks a month. Participants reported seeing an average of 22.7 alcohol advertisements per month. For each additional advertisement seen, the number of drinks consumed increased by 1% Event Rate Ratio (ERR) 1.01 (95% CI: 1.01 to 1.02). Also for each additional dollar per capita spent on advertising the number of alcoholic drinks consumed per month increased by 3% ERR 1.03 (95% CI: 1.01 to 1.05). In the sub-group of participants aged less than 21 years (60% of sample), who were below the legal drinking age, similar patterns were seen, ERR 1.01 (95% CI: 1.0 to 1.02) and 1.03 (95% CI: 1.0 to 1.06) increase in number of drinks consumed per month for self-reported advertising exposure and advertising expenditure, respectively. All analyses were adjusted for gender, age, ethnicity, school status and alcohol sales per capita, however the high degree of attrition in this study (more than 50% for two of the four follow-up assessments) precludes firm conclusions on the basis of these findings.
Two studies evaluated exposure to TV and music videos and alcohol use in adolescents [19, 20]. In the study by Robinson et al the association between hours of TV, music video and videotape viewing, computer and video game use and subsequent alcohol use at 18 months follow-up was investigated in 1,533 14 to 15 year olds from six public high schools in California. During follow-up, 325 (36.2%) baseline non-drinkers began drinking and 322 (50.7%) drinkers continued to drink. In baseline non-drinkers (n = 898), onset of drinking was significantly associated with hours of TV viewing at baseline. For each additional hour of TV viewing per day the average increased risk of starting to drink during the next 18 months was 9% OR 1.09 (95% CI: 1.01 to 1.18), for each additional hour of music video viewing OR 1.31 (95% CI: 1.17 to 1.47). For each additional hour of videotape viewing the average risk decreased, 11% OR 0.89 (95% CI: 0.79 to 0.99) in analyses controlling for age, sex, ethnicity and other media use. Computer and video game use was not significantly associated with subsequent onset of drinking, OR 0.94 (95% CI: 0.84 to 1.05). In baseline drinkers (n = 635), there were no significant associations between baseline media use and maintenance of drinking. For each additional viewing hour per day the risk, OR (95% CI), of maintenance of drinking was: 1.01 (0.93, 1.11) for television, 1.05 (0.95, 1.17) for music videos, 0.97 (0.86, 1.10) for videos and 1.00 (0.89, 1.12) for computer or video games.
Van Den Bulck  examined the relationship between television viewing and music video exposure and subsequent alcohol consumption while going out one year later in 2,546 first and fourth year secondary school students in Flanders, Belgium. Only 65% of the original sample with complete data at both time-points was analysed. The majority of students (63.6%) watched music videos at least several times a week, about a third watched daily. Overall television viewing and music video viewing at baseline significantly predicted the amount of alcoholic beverages adolescents consumed while going out at follow-up. Results of a regression model controlling for gender, school year, smoking and pubertal status were reported: R2 = 0.568 (F = 230.374; df = 7; p < 0.0001).
Sargent evaluated the exposure to alcohol use in popular contemporary movies in a cross-sectional survey with prospective follow-up of never drinkers and recorded incident alcohol drinking 13 to 26 months later. Adolescents, 10 to 14 years old, were recruited from 15 randomly selected schools in New Hampshire and Vermont, USA. Never-drinkers at baseline were followed up (n = 2,406). Baseline median exposure to alcohol use in 601 movies was 8.6 hours, (inter-quartile range (IQR): 4.6 to 13.5). At follow-up, 14.8% reported having tried alcohol, which was significantly associated with alcohol exposure (viewing hours). For each additional hour of movie alcohol exposure the risk of initiating alcohol use was increased by 15%, OR 1.15 (95% CI: 1.06, 1.25) adjusted for school grade, school, gender, parent education, sensation seeking, rebelliousness, self-esteem, school performance, parenting style and smoking experimentation.