Early Drinking Onset and Subsequent Alcohol Use in Late Adolescence: a Longitudinal Study of Drinking Patterns

Purpose: The age of drinking onset is a central concept for both policy and prevention of alcohol-related harm, yet evidence on the predictive value of the age of onset is lacking. This study compares alcohol outcomes of adolescents who started to drink early with those who started later, and tests if associations are moderated by other explanatory factors. Methods: Data from a two-wave longitudinal prospective cohort survey with a Swedish nationwide sample of 4,018 adolescents aged 15/16 years at baseline (T1) and 17/18 years at follow-up (T2) were used. Outcome variables at T2 were Alcohol Use Disorders Identi ﬁ cation Test (AUDIT) e C, risky drinking, and binge drinking monthly or more often. A vast number of explanatory factors at T1 were controlled for. Results: Early drinking onset predicted later higher AUDIT-C scores ( b ¼ 0.57, p value < .001), and higher probability of risky drinking (odds ratio ¼ 1.95, 95% con ﬁ dence interval ¼ 1.56 e 2.44), and binge drinking (odds ratio ¼ 1.38, con ﬁ dence interval ¼ 1.06 e 1.81), controlled for other explanatory factors. If binge drinking frequency at T1 was included, the associations remained for AUDIT-C and risky drinking, but not for binge drinking at T2. No signi ﬁ cant interactions between early drinking onset and the explanatory factors were found. Discussion: Early drinking onset predicts

The current evidence of early drinking onset on future drinking habits is insufficient.This study shows that an early age of onset was related to higher alcohol use in late adolescence, after controlling for other explanatory risk factors.The age of onset predicted drinking outcomes 2 years later.
Delaying the onset of alcohol use has been considered an important policy that underpins many actions on alcohol prevention [1].An early drinking onset has been suggested to increase the risk of higher subsequent alcohol use, but more research is needed as the current state of evidence is contradictory [2,3].
An early drinking onset may contribute to higher alcohol use later in life [4], primarily through higher vulnerability at a younger age [5e8], by affecting relationships, social status, and self-conception, e.g., by engaging with risk-taking peers [9e11], or through longer exposure to a substance with addictive and habit-forming properties [10,12].
In recent decades, young people's drinking has decreased in most of the Western world, with the age of drinking onset shifting upward [13,14].For example, a study from Sweden shows that 17year-olds today drink as 15-year-olds did 13 years ago [15].The older age of onset may have changed the meaning of early drinking onset from early to mid-adolescence.Thus, the developmental period of mid to late adolescence has become important to understand the initiation and progress of alcohol use.
Further, the significance of age of drinking onset has been questioned [1,16e22].The debate concerns the causal chain and whether the age of onset directly or indirectly affects later drinking habits.This has given rise to two diverging ideas.The "causality hypothesis" [22] acknowledges that early determinants, distal predictors, and pre-existing differences among adolescents manifest in early drinking onset, but also that early drinking on its own right affects subsequent alcohol consumption.The other idea, the "marker hypothesis" [1,21], stipulates that early drinking onset is confused as a predictor.Instead, early onset is simply a proxy and a marker of other risk factors, without causal importance per se.The underlying actual cause is instead believed to be a third factor that induces vulnerability and susceptibility to both early drinking onset and adult alcohol problems, and the association should disappear if other risk factors are considered.
However, a systematic review has concluded that the empirical evidence is weak, as few longitudinal studies have examined the influence of adolescent alcohol use on later drinking habits [1].Most studies are based on cross-sectional data impaired by recall bias and therefore not suited for claims spanning over time and causal inference [17,19].Furthermore, many studies do not control for potential confounding factors and thus cannot rule out alternative explanations.The few prospective cohort studies that have examined the link between early drinking onset and later alcohol problems have yielded inconsistent findings.For example, when potential confoundersdsuch as conduct problems and alcohol problems among parentsdwere controlled for, the effect of early drinking onset diminished in one study [23], whereas it remained in another [24].Consequently, an inconsistency remains of whether the effect of early onset remains when other explanatory factors are controlled for.Further, such other explanatory factors can also moderate the effect that early onset has on subsequent drinking outcomes.
Other explanatory factors of adolescent drinking are personality traits and alcohol-related habits within the social environment.Sensation-seeking and impulsivity are often elevated during adolescence [25], and these traits are known risk factors for drinking.Aggressivity is related to intoxication [26].Externalizing behaviors can increase alcohol use, while internalizing behaviors can decrease it [27].Children exposed to parents' drinking have a higher probability of drinking themselves [28].Adolescents whose parents approve [29] or permit drinking have an elevated risk of drinking [30].Alcohol-using friends are also predictive of adolescent drinking [29].
The present study aims to determine the importance of the age of drinking onset in a large, nationally representative Swedish birth cohort by examining whether an association between age of onset and subsequent alcohol use in adolescence holds when other explanatory factors are controlled for.Further, we test if any of these explanatory factors moderate associations.
Suppose the "marker hypothesis" holds true.In that case, we anticipate that any significant association between the age of onset and subsequent drinking will disappear when the other explanatory factors are incorporated into our model.Conversely, if the "causality hypothesis" is correct, we expect that such association persists by still being significant after these factors are considered.

Data
The data for the current study are derived from a prospective longitudinal research project following a cohort of adolescents born in 2001.The project started in 2017 and was approved by the ethical review board in Stockholm (Dnr 2017/ 103-31/5).The participants were recruited as ninth grade students (age 15/16 years) in the Swedish school system.In total, 343 schools (68.6%) and 5,537 (81.7%) students chose to participate.All respondents answered a paper-and-pencil questionnaire while at school.
The first follow-up (T2) occurred 2 years later.In 2019, the respondents were 11th grade students (age 17/18 years).The second questionnaire was mailed to each student's registered home address.Answering the questionnaire on the web was an option.Most preferred the digital option (82.2%) over the traditional paper format (17.8%).In total, by 4,018 respondents (72.4%) answered the second questionnaire.Drop-out analyses showed that those who did participate in T2 differed slightly at baseline from those who only participated in T1: there was a higher proportion of girls, they had slightly lower sensation-seeking, impulsivity, and aggressivity, furthermore they drank slightly less on an average.
The analytical sample only included observations on those who reported past-year drinking status either at both time points or only at the follow-up.Thus, 1,257 of the original 4,018 responders (31.3%) were excluded, including non-drinkers (28.1%) or those only drinking at T1 but not at T2 (3.2%).Next, listwise deletion was applied to the remaining 2,761 responders, where 235 were excluded due to missing reports on drinking status or other variables (8.5%).
In the final analytic sample of 2,526, 56.7% were girls, whereas in the full sample was 54.9% girls.Participating schools did not differ in parents' education level, immigration status, or grade point average, from non-participating schools.Thus, survey weights were not used.However, responses from the same school can be correlated, narrowing the spread and variance in estimates.To adjust for this, cluster-robust standard errors were used.

Measures
The age of drinking onset was measured with the same question at both time points: "Have you ever had a drink of alcohol (disregard drinks below 2.8%, such as light beer or weak cider)?"The predefined response options were as follows: "No", "Yes, during the past 30 days", "Yes, during the past 12 months", and "Yes, more than 12 months ago".This item was operationalized into (1) "Early drinking onset" based on reported past-year drinking at 15/16 and 17/18 years of age, and (2) "Late drinking onset" based on reports of nondrinking at age 15/16 years and past-year drinking at age 17/18 years.
AUDIT-C at age 17/18 years was measured using the first three items of the Alcohol Use Disorders Identification Test (AUDIT) [31] which capture drinking frequency, quantity, and binge frequency.A total score scale ranging from 0 to 12 was created based on these items.Risky drinking had a cut-off of four points for boys and three points for girls on the AUDIT-C indicating (0) "Nonrisky drinking" and (1) "Risky drinking".Previous studies suggested a cut-off of three for adolescent girls and four for adolescent boys [32].
Binge drinking monthly or more often was measured at both time points and based on a frequency measure from AUDIT-C (26), "How often do you drink six such "drinks" or more during the same occasion?",with an illustration of what typical alcohol beverages that corresponds to a "drink".Response options were categorized into (0) "Never"/"More seldom than once a month" or (1) "Every month"/"Every week"/"Daily or almost every day".
Sex was derived from the national identification numbers of the study participants.This number is assigned when listed in the Swedish population register.The next-to-last digit is even for females and odd for males.A binary variable was used with the following values: (0) "Female" and ( 1) "Male".
Sensation-seeking, Impulsivity, and Aggressivity at age 15/16 years were measured with the question "How well do the following statements describe how you are as a person?", with the following predefined response options: "Not at all," "Quite badly," "Neither well nor badly," "Quite well" and "Completely accurately."Sensation-seeking consisted of four items from the Brief Sensation-Seeking Scale [33] and two other tested items [34].Impulsivity was measured with three items: "I often do things impulsively without thinking in advance," "I try to avoid difficult tasks," and "I do what I feel like, without considering whether it is good or bad in the long term."Aggressivity was measured with two items: "When I get angry, I have a hard time not screaming, slamming doors and such" and "It's quite easy for me to get angry."Composite mean score scales were used with moderate-to-very good internal consistency: sensation-seeking (0.88), impulsivity (0.62), and aggressivity (0.70).
Emotional symptoms, peer relationship problems, conduct problems, and hyperactivity at age 15/16 years were measured with five items each from the Strengths and Difficulties Questionnaire [35].The factor structure of these measures using this study sample has been reported elsewhere [36].A total score scale (0e10) was calculated for each of these four variables.
Parents' permissiveness of drunkenness at age 15/16 years was measured for each parent with the question: "What do you think your mother's/father's reaction would be if you did any of the following?",followed by the statement "Drinking until you are drunk".The predefined response options were as follows: (1) "She/He would absolutely not allow it", (2) "She/He would try to prevent it", (3) "She/He would not care" and (4) "She/He would accept it".The alternative "Do not know" was coded as a missing value.A mean score scale was created based on both variables, with good internal consistency (0.80).
Alcohol problems among parents at participant age 15/16 years was measured with the six-item scale Children of Alcoholics Screening Test [37].The response options were as follows: (1) 'Yes' or (0) 'No'.A total score scale (0e6) was created, with very good internal consistency (0.84).

Analysis
Cronbach's alphas were used to test the internal consistency of composite measures.Early and late drinking onset groups were compared using mean values, prevalence rates, and confidence intervals on explanatory risk factors at baseline and alcohol habits at follow-up.T-tests were applied to estimate significant differences.Multivariable linear and logistic regression models were used to examine associations.The age of onset (15/16 years vs. 17/18 years) was used as the predictor of drinking habits at age 17/18 years.Those with early drinking onset were used as the reference group.The analyses were not preregistered, and all results are to be considered exploratory.The risk of multicollinearity was assessed with the Spearman rank test and no problems were found.The highest correlation among the independent variables was between impulsivity and hyperactivity (0.56).Interaction analyses were conducted to examine modifications of associations.A 95% confidence level was used

Results
Early drinking onset was reported by 49.9%, whereas late drinking onset was reported by 50.1%.Descriptive statistics on all included variables by age of drinking onset are shown in Table 1.Every indicator of drinking at T2 differed significantly between early and late drinking onset, with the early-onset group showing higher drinking.Further, risk profiles at baseline also differed, with higher levels of all explanatory risk factors in the group with early drinking onset except peer problems, which were reported to be higher in the late-onset group.
Linear and logistic regression analyses predicting drinking at T2 are presented in Table 2. First, crude models were used to assess how early drinking onset as a single factor predicted each alcohol use outcome.In these bivariate unadjusted models, early onset of alcohol use was significantly predictive of higher levels of all three drinking outcomes.The age of onset accounted for 8.0% of the variability in AUDIT-C scores, 5.5% in risky drinking and 2.9% in monthly binge drinking.In the next step, additional explanatory factors were included in an adjusted model, to examine if the effects remained if these risk factors were controlled for.The age of onset remained significantly associated with all three outcomes in the adjusted models.The coefficient for AUDIT-C score dropped from 1.17 to 0.57 in the adjusted model.However, early drinking onset had the largest standardized coefficient (0.14) of all explanatory factors for predicting AUDIT-C score (not shown in Table 2).In the adjusted models, the unique contribution from age of onset for variation in AUDIT-C scores was 3.9% of the total contribution of 15.1% from all explanatory risk factors.In order to explore the contribution of the other explanatory variables we divided them into three groups; these can be labeled as rigid traits, personal problems, and alcohol factors within the social environment.Rigid traits consisted of sex, sensations-seeking, impulsivity, and aggressiveness, and accounted for 4.3%.Personal problems consisted of emotional symptoms, peer problems, conduct problems, and hyperactivity, and accounted for 2.7%.Alcohol factors within the social environment consisted of parents' permissiveness of drunkenness, alcohol problems among parents, and drinking and drunkenness among friends, and accounted for 4.2% of the total contribution of our model in explaining the variation in AUDIT-C.Coefficients for the additional explanatory factors are shown in a supplementary table online (Table S1).
The age of onset and T1 binge drinking frequency were included in sensitivity analyses and all associations remained significant.However, the association between age of onset and binge drinking monthly or more often at T2 was no longer significant when additional explanatory factors and T1 binge drinking were included.
To examine if the different explanatory risk factors moderated the effect of early drinking onset on later drinking, interaction effects were added to the adjusted models (see Table 3).No interactions were found to be significant.

Discussion
In this study, the importance of the age of drinking onset on adolescents' subsequent alcohol use was examined.Among alcohol users aged 17/18 years in Sweden, about half had begun drinking at age 15/16 years.
First, the major finding of this study was that early drinking onset predicted higher alcohol consumption as indicated by AUDIT-C scores, as well as risky drinking and binge drinking monthly or more often.These associations remained after rigorous control of other explanatory factors such as conduct problems and friends' use of alcohol.For AUDIT-C scores, early drinking onset had the strongest standardized estimates of all included factors.These results support longitudinal research from other countries, demonstrating that early drinking onset is associated with increased alcohol consumption among adolescents [24].However, a systematic review of prospective cohort studies [1] showed that the effect of early drinking onset disappeared or was markedly attenuated when confounders were added.In our study, the effect of early onset remained for all three alcohol outcomes when other explanatory factors were adjusted for.A difference between our study and those included in the review is that those studies had a longer period between exposure and outcome.This may indicate that the effect of early drinking onset weakens with age.Another explanation could be that these studies were published between 1998 and 2013 and that the decline in youth drinking since then has affected the impact of the age of drinking onset.
Second, the results in this study were more in support of the "causality hypothesis" [22], as associations between age of onset and later alcohol use remained when other explanatory factors were taken into account.However, the support is indicative since the data did not allow mediation analysis.This result contrasts with those of a study [23] supporting the "marker hypothesis", where the effect of early drinking onset vanished if conduct problems were controlled for.Adolescents in that study were younger at baseline and followed for much longer.The inconsistency of the findings can also be attributed to shifts in drinking prevalence.Researchers have put forward that the decline in youth drinking may have been so profound that it can have shifted associations between risk factors and alcohol use [38].Thus, comparisons between the studies should be cautiously interpreted.Another caveat is that our study cannot rule out that latent unmeasured factors could influence the relationship and alter the associations we found.Examples of influential factors we could not consider were genetic variation and structural forces such as physical availability of alcohol [39].Third, some studies have found early drunkenness onset to be more important than early drinking onset [17,40].We addressed this issue in further analyses by considering binge drinking frequency at age 15/16 years.By taking early binge drinking into account, it was possible to distinguish the effect of early drinking onset from that of early binge drinking onset.Our results showed that the relationship between age of onset and alcohol use remained for AUDIT-C and risky drinking, but not for binge drinking monthly or more often, when all other explanatory factors, including previous binge drinking, were controlled for.We conclude that the effect of early drinking onset on later alcohol use cannot be attributed solely to the onset of high alcohol intake.However, we found an effect of early binge drinking escalating into more binge drinking at T2, suggesting that early binge drinking onset is more predictive of this outcome.Early binge drinking onset appeared to overshadow any effect of early drinking onset regarding subsequent binge drinking only.
All three alcohol outcomes in this study are based on AUDIT-C items.That the age of onset predicted all three outcomes should therefore not be interpreted as a guarantee for robustness.
The cut-off used for risky drinking among adolescents varies somewhat between studies [41].Sex-specific cut-offs of three and four points for girls respectively boys were used for risky drinking in this study.This might be considered a high bar for risky drinking, and the implication of this is that our estimates are underestimated.This was specifically tested in our study by using a one-point lower cut-off and the associations were then shown to be even stronger.A potential weakness of this study is that early drinking onset was defined as being before age 15/ 16 years, which some may not consider early.On the other hand, the age of drinking onset has shifted [13,14], moving the boundaries of early drinking onset upward.However, half of our participants had started drinking at age 15/16 years, which might indicate that this point is late as an "early" drinking onset.One likely consequence is that we found weaker associations compared with if early drinking onset had been at a lower age.Another shortcoming was the limited information on when those with later onset started to use alcohol.The only information available is that it occurred between the ages of 15/16 years and 17/18 years.A limitation of longitudinal studies is the risk of attrition bias because of study participant dropouts.This was seen in our study; participants in T2 differed slightly from those who participated in T1, as the group at follow-up drank less at baseline, and had higher scores on sensation-seeking, impulsivity, and aggressivity, in addition to higher percentage of girls than those dropping out.The implication is that our results generalize better to lowerrisk groups and that the associations we found might be underestimated, as the participants who dropped out were early drinkers to a greater extent.
One main strength of this study is the use of an extensive national sample with a good response rate, including both boys and girls.This allowed us to examine subgroups based on early and late drinking onset.Although caution needs to be taken, the considerable power of the sample allowed us to detect minor differences, whether or not they have practical relevance.The many and varied measures of explanatory factors studied with well-established instruments are also a major advantage with relevance for this study.
This topic is a central concept in our scholarly understanding of how alcohol consumption initiates and develops in adolescence, and the findings have important policy and prevention implications.Previous findings among Swedish adolescents indicate that those with early drinking onset are worse off on many psychosocial factors compared with their peers with later onset [42], and in our study, we could also observe that those with early onset scored higher on nearly all explanatory risk factors (shown in Table 1).Thus, prevention should target adolescents with this risk profile to postpone their drinking onset.If we push the age of onset upward, we will reduce drinking among adolescents here and now, as well as in the futureeas this will render them likely to drink less when they start to drink.This will also likely reduce alcohol-related harm in both the short and long term.Future studies should include three data points and conduct mediation analyses on whether adolescents with a midonset drink more than those with a later onset.Additional recommendations for this line of research are to provide a theoretical framework for the examination of confounding factors.

Conclusions
Adolescents in Sweden who had started drinking at age 15/16 years drank more 2 years later than those who had started drinking at 17/18 years.The age of drinking onset predicted subsequent alcohol use in late adolescence.Early drinking onset had an effect beyond well-known risk factors for drinking.However, early binge drinking onset may have a stronger effect on subsequent binge drinking.

Table 1
Mean scores (confidence intervals) and prevalence rates of included variables by early and late drinking onset .Sjödin et al. / Journal of Adolescent Health 74 (2024) 1225e1230and p values below .05were considered significant.Data were analyzed using Stata/SE 15.1.To assess what impact age of onset and our full model have on the variation in drinking, we used the 'Shapely2' command in Stata, which provides us with estimates of our model's capability of explaining the AUDIT-C outcome and each factor's relative influence. L

Table 2
Linear and logistic regression models of early drinking onset as a predictor of subsequent drinking outcomes (n ¼ 2,526) CI ¼ confidence interval; OR ¼ odds ratio; AUDIT ¼ Alcohol Use Disorders Identification Test. a Controlled for sex at birth and for sensation-seeking, impulsivity, aggressiveness, emotional symptoms, peer problems, conduct problems, hyperactivity, parents' permissiveness of drunkenness, alcohol problems among parents, drinking and drunkenness among friends at T1. L. Sjödin et al. / Journal of Adolescent Health 74 (2024) 1225e1230

Table 3
Interaction analysis using adjusted multiple linear and logistic regression models with early drinking onset as a predictor of subsequent drinking outcomes (n ¼ 2,526)