#essay

Research Article

Is It Important to Prevent Early Exposure to Drugs and Alcohol Among Adolescents? Candice L. Odgers,1 Avshalom Caspi,2,3 Daniel S. Nagin,4 Alex R. Piquero,5 Wendy S. Slutske,6

Barry J. Milne,3 Nigel Dickson,7 Richie Poulton,7 and Terrie E. Moffitt2,3

1 University of California, Irvine;

2 Duke University;

3 Kings College London;

4 Carnegie Mellon University;

5 University of

Maryland; 6 University of Missouri-Columbia; and

7 University of Otago

ABSTRACT—Exposure to alcohol and illicit drugs during

early adolescence has been associated with poor outcomes

in adulthood. However, many adolescents with exposure to

these substances also have a history of conduct problems,

which raises the question of whether early exposure to

alcohol and drugs leads to poor outcomes only for those

adolescents who are already at risk. In a 30-year pro-

spective study, we tested whether there was evidence that

early substance exposure can be a causal factor for

adolescents’ future lives. After propensity-score matching,

early-exposed adolescents remained at an increased risk

for a number of poor outcomes. Approximately 50% of

adolescents exposed to alcohol and illicit drugs prior to age

15 had no conduct-problem history, yet were still at an

increased risk for adult substance dependence, herpes in-

fection, early pregnancy, and crime. Efforts to reduce or

delay early substance exposure may prevent a wide range

of adult health problems and should not be restricted to

adolescents who are already at risk.

Many adolescents experiment with drugs and alcohol, and

parents, teachers, and policymakers want to know the conse-

quences of adolescent substance use. Parents, in particular,

worry that their adolescents who use drugs or alcohol may suffer

long-term consequences, such as a dependence on drugs; be

drawn into risky sexual behaviors; contract sexually transmitted

diseases (STDs); fall behind in school; or get a criminal record.

Research has demonstrated that exposure to illicit drugs and

alcohol prior to age 15 statistically predicts substance disorders

in adulthood (Grant & Dawson, 1997; Hingson, Heeren, &

Winter, 2006). Exposure to these substances during adoles-

cence has also been linked to involvement in risky sexual

behaviors and STDs (Stueve & O’Donnell, 2005), early preg-

nancy (Ellickson, Tucker, & Klein, 2003), low educational

attainment (King, Meehan, Trim, & Chassin, 2006), and crime

(Elliott, Huizinga, & Menard, 1989). On the basis of this

evidence, the U.S. Surgeon General’s office has issued a call to

action to stop underage drinking (U.S. Department of Health and

Human Services, 2007), placing a special emphasis on risks to

youth who report drinking before the age of 15. However,

nagging doubts remain regarding whether substance use affects

adolescents’ later lives causally.

Critics of this recent policy stance allege that early exposure

to alcohol and drugs per se is not the cause of problems in ad-

olescents’ later lives (Peele, 2007). Indeed, research has re-

peatedly shown that adolescents who use substances in early

adolescence are also more likely than other adolescents to have

a childhood history of conduct problems, which itself predicts

the same adult outcomes (Moffitt, 2006). A recent review con-

cluded that adolescents at risk for developing substance-use

disorders are not ‘‘just normal adolescents who happen to be

experimenting with substances but, in many cases, are youths

with other . . . problems’’ (Armstrong & Costello, 2002, p. 1235).

Thus, parents, teachers, and policymakers are caught between

mixed messages: On the one hand, they are told that exposure to

drugs and alcohol prior to age 15 is harmful for adolescents and

should be prevented, and on the other hand, they are told that it

is normal for adolescents to try these substances and that the

majority will not become addicts or ruin their lives.

It is not practical or feasible to administer alcohol and drugs to

young adolescents in a randomized control trial in order to test

Address correspondence to Candice L. Odgers, Department of Psy- chology and Social Behavior, University of California, Irvine, 3361 Social Ecology Building II, Irvine, CA 92697-7085, e-mail: codgers@ uci.edu.

PSYCHOLOGICAL SCIENCE

Volume 19—Number 10 1037Copyright r 2008 Association for Psychological Science

the causal effect of substance use on adolescents’ future lives.

However, analytic strategies are now available to quantify

‘‘treatment’’ effects in observational studies when randomized

control trials are not a viable option (D’Agostino & D’Agostino,

2007; Stuart & Green, 2008). Such propensity-score methods

are designed to re-create the desirable features of experimental

designs by creating balance between exposed (‘‘treatment’’)

and nonexposed (‘‘control’’) groups that have formed naturally

over the course of an observational study (Haviland, Nagin, &

Rosenbaum, 2007).

Here we report the results of applying propensity-score

matching (Rosenbaum & Rubin, 1985) in a 30-year longitudinal

study to address a question of significant public-health impor-

tance that cannot be studied using randomized control trials.

Specifically, we asked whether there is evidence that early

exposure to illicit substances is a causal factor in adolescents’

future lives (‘‘are drugs bad for kids?’’), or whether adolescents

with a developmental history of conduct problems are simply

more likely than other adolescents to be exposed to alcohol and

illicit substances and to experience poor adult outcomes (‘‘do

bad kids do drugs?’’). At present, the extent to which a history of

conduct problems in childhood has confounded the association

between early substance exposure and adult outcomes is un-

clear. Researchers have called for prospective longitudinal

designs that are able to adequately account for childhood con-

duct problems (Hingson et al., 2006); however, to date, only a

handful of studies have met this criterion.

In the present study, adolescents’ developmental history of

conduct problems was assessed prospectively by multiple

informants. Assessments were made when the study members

were ages 7, 9, 11, and 13. Early substance exposure was de-

fined as frequent exposure to illicit substances prior to age 15.

Adolescents were then followed prospectively into adulthood

to assess the influence of early substance exposure on their adult

lives. We estimated two types of treatment effects to test whether

early substance exposure influenced adolescent’s adult lives.

First, we estimated the effect of early substance exposure on

adolescents’ adult outcomes across the entire birth cohort.

Propensity-score matching was applied to adjust treatment-

effect estimates for nonrandom assignment to the treatment

condition (in this case, defined as exposure to illicit substances

prior to age 15, which we term early exposure) and to facilitate

causal inference by ensuring that adolescents with early exposure

were balanced relative to those with no early exposure on key

background factors that might otherwise confound the results.

Second, we estimated group-specific treatment effects to test

whether early substance exposure is similarly dangerous for

all adolescents, or whether some adolescents’ developmental

history of conduct problems signals particular vulnerability to

early substance exposure. Developmental trajectories of con-

duct problems were defined previously in this cohort using

group-based trajectory modeling (Odgers et al., 2008); this work

isolated a group of children (66% of the cohort) who followed a

no-conduct-problem trajectory between the ages of 7 and 13.

Adolescents on this trajectory had very few deficits in childhood

and were less likely than their already-at-risk peers to experience

early substance exposure and poor adult outcomes. The question

for the majority of ordinary adolescents, therefore, was whether

adolescentswhohadno history of conduct problemswere protected

from any untoward effects of early substance exposure. By contrast,

the remaining 34% of the children in this cohort entered adoles-

cence with a history of conduct problems; these adolescents were

more likely (or had a higher propensity) than their normative peers

to experience early substance exposure and poor adult outcomes.

Thequestionforthissubgroupofalready-at-riskadolescents,there-

fore, was not whether they would go on to experience poor adult

outcomes, but rather whether early substance exposure placed an

additive burden on their already-compromised future lives.

METHOD

Participants

Participants were members of the Dunedin Multidisciplinary

Health and Development Study. The cohort of 1,037 children

(52% male) was constituted at 3 years of age, when investigators

enrolled 91% of children born consecutively between April

1972 and March 1973 in Dunedin, New Zealand. Cohort fami-

lies represent the full range of socioeconomic status in New

Zealand’s South Island and were primarily White. Follow-up

assessments were conducted, with informed consent, when the

cohort members were 5, 7, 9, 11, 13, 15, 18, 21, 26, and 32 years

of age; 96% of the living study members participated in the

age-32 assessment, which took place in 2003 through 2005.

Research ethics committees at the University of Otago, Duke

University, and Maudsley Hospital approved this research.

Early Exposure and Conduct-Problem Measures

Early Substance Exposure

At ages 13 and 15, study members reported on their frequency of

exposure to illicit substances during the past year. They reported

whether they had sniffed glue, gasoline, or other inhalants;

smoked cannabis; used any illegal drugs other than cannabis;

bought or drunk alcoholic drinks; or drunk alcoholic drinks

during school. Response options were 0, never; 1, once or twice;

and 2, multiple occasions. In total, 11.2% of the study members

were classified as being exposed to substances on multiple

occasions at age 13, age 15, or both; hereafter, we refer to these

study members as early-exposed adolescents. Alcohol was the

substance most commonly used by the young adolescents in our

cohort; 10% of study members were exposed to alcohol prior

to age 15, 5% were exposed to cannabis, and fewer than 2% were

exposed to inhalants or other drugs.

Conduct Problems

Self-, parent, and teacher reports were used to assess study

members’ conduct problems at ages 7, 9, 11, and 13 years. The

1038 Volume 19—Number 10

Early Exposure to Drugs and Alcohol

following six symptoms of conduct disorder, as defined by the

fourth edition of the Diagnostic and Statistical Manual of Mental

Disorders (DSM–IV; American Psychiatric Association, 1994),

were assessed as being present or absent at each age: physical

fighting, bullying other people, destroying property, telling lies,

truancy, and stealing. A composite score, ranging from 0 to 6,

was calculated to represent the number of different types of

conduct-problem behaviors each study member had engaged

in during the past year. As noted earlier, prior group-based

trajectory modeling isolated young adolescents following a no-

conduct-problem trajectory between the ages of 7 and 13 (65.7%

of the cohort; 74.1% of females and 57.9% of males). This group

represented a ‘‘pure,’’ or super-healthy, group in that the average

number of conduct-problem symptoms at each age was below 1.

Background Covariates

Psychometric properties of all background covariates in this

study have been reported in detail elsewhere (Moffitt, Caspi,

Rutter, & Silva, 2001), and all estimates of their reliability

exceeded .70. Background covariates were assessed prior to age

13, unless otherwise stated.

Family history of alcohol and drug disorders was assessed in

2003 through 2006 as part of the Dunedin Family Heath History

Study (DFHHS; Odgers et al., 2007). That study used the Family

History Screen (Weissman et al., 2000) to collect psychiatric-

history data about each study member’s biological parents,

grandparents, and siblings older than 10 years old. Ratings were

obtained from multiple informants. A family-liability score

was computed for each study member. This score indexed

the proportion of family members, across three generations, with

an alcohol or drug disorder.

Criminal conviction of a parent was also assessed through the

DFHHS; 25% of study members had at least one parent with a

criminal conviction.

Socioeconomic status (SES) was measured as the higher of the

father’s or mother’s occupation, as rated on a 6-point scale for

New Zealand (Elley & Irving, 1976); 21% of the families were

classified as low SES.

Maltreatment was measured using the following indicators:

rejecting mother-child interactions (as observed by staff mem-

bers), parental reports of harsh discipline, two or more changes

in primary caregiver, and retrospective self-reports of injurious

physical abuse or unwanted sexual contact. A study member was

considered to be maltreated if he or she had two or more indi-

cators of maltreatment (Caspi et al., 2002); on this basis, 9% of

the study members were classified as maltreated.

Mother’s IQ was tested using the SRA (Thurstone & Thurstone,

1973); scores were standardized (M 5 100, SD 5 15). Low mo-

ther’s IQ was defined as a standardized SRA score less than 85.

Child IQ was tested using the Wechsler Intelligence Scale

for Children–Revised (WISC-R; Wechsler, 1974); scores were

standardized (M 5 100, SD 5 15). Low child IQ was defined as a

standardized WISC-R score below 85.

Undercontrolled temperament was measured through staff

ratings. Ratings were made after observing the child in a 90-min

testing session with an unfamiliar examiner. Factor and cluster

analyses reduced these ratings to three temperament types,

including the undercontrolled type (Caspi & Silva, 1995).

Attention-deficit/hyperactivity disorder (ADHD) was measured

using the Diagnostic Interview Schedule for Children (Costello,

Edelbrock, Kalas, Kessler, & Klaric, 1982). Data obtained

when the study members were ages 5 to 15 were used to make

diagnoses according to the criteria of the third edition of

the Diagnostic and Statistical Manual of Mental Disorders

(DSM–III; American Psychiatric Association, 1980); diagnoses

were confirmed through parent or teacher report. Six percent of

study members met diagnostic criteria for ADHD.

Adult Outcomes

Substance-use disorders at age 32 were assessed from data

obtained in private structured interviews using the Diagnostic

Interview Schedule (Robins, Cottler, Bucholz, & Compton,

1995). Diagnoses were made according to DSM–IV criteria

(American Psychiatric Association, 1994). For this report,

substance dependence was defined as cannabis dependence,

dependence on other drugs, or alcohol dependence within the

past year. Dependence at age 32 signals a substance-use

problem serious enough to outlast early adulthood, a develop-

mental period when large numbers of young people can meet

criteria for substance disorders on a short-term basis.

Type 2 herpes infection at age 32 was assessed using blood

samples. Diagnoses were made on the basis of an indirect en-

zyme-linked immunosorbent assay (HerpeSelect s

2 ELISA

IgG, Focus Technologies, Cyprus, CA; Eberhart-Phillips et al.,

2001). Herpes infection was diagnosed using a cutoff value of

3.5, and any equivocal result (between 0.9 and 3.5) was resolved

using the Western Blot test (Ho, Field, Irving, Packham, &

Cunningham, 1993).

Early pregnancy was defined as having at least one pregnancy

prior to age 21 and was assessed by females’ self-report. The

failure to delay pregnancy until age 20 or 21 is associated with

economic costs and poor social consequences for both mothers

and children in contemporary cohorts (Maynard, 1996).

No educational qualifications was defined as ending second-

ary education prior to receiving qualifications and not returning

to earn qualifications by age 32. In New Zealand, students re-

ceive qualifications on the basis of national exams that almost

all students take by age 16; the results determine promotion in

secondary and technical schools and help people secure better

employment in the labor market (Miech, Caspi, Moffitt, Wright,

& Silva, 1999).

Number of criminal convictions between ages 17 and 32

was determined for each study member by searching the

computerized New Zealand Police database. Convictions for

nonviolent and violent crimes were included, but traffic con-

victions were not.

Volume 19—Number 10 1039

C.L. Odgers et al.

STATISTICAL ANALYSES

Analyses proceeded in four steps. First, because study members

were not randomly assigned to a substance-use condition,

potential selection biases were addressed by developing a

propensity score for early substance exposure. This score, e(x),

represented the conditional probability of early substance ex-

posure (z 5 1), versus no early substance exposure (z 5 0), given

a combination of key familial, social, and child covariates (x);

that is, e(x) represented P(z 5 1|x). Propensity scores were

calculated for all adolescents using a multivariate logistic re-

gression that included the 13 covariates listed in Table 1. The

c statistic for this model was .72 (95% confidence interval:

.67–.77), which indicates a fair-to-good ability to discriminate

between early-exposed and non-early-exposed adolescents.

The propensity scores ranged from .02 to .73.

Second, we used the STATA module PSMATCH2 (Leuven &

Sianesi, 2003–2006) to perform 3-to-1 nearest-neighbor pro-

pensity-score matching. Specifically, propensity scores were

used to match each early-exposed adolescent to 3 non-early-

exposed adolescents who had a similar probability of early

exposure.

Third, to determine whether early substance exposure influ-

enced adolescents’ later lives, we compared the adult outcomes

of early-exposed adolescents with the adult outcomes of

propensity-matched non-early-exposed adolescents.

Fourth, adolescents were stratified by their conduct-problem

history to estimate group-specific treatment effects. More spe-

cifically, we tested (a) whether early substance exposure pre-

sents a risk for the majority of adolescents, who have no conduct-

problem history (no-conduct-problem group), and (b) whether

adolescents with a prior history of conduct problems (conduct-

problem group) have a particular vulnerability to early substance

exposure.

RESULTS

Does Propensity-Score Matching Create Balance Between

Early- and Non-Early-Exposed Adolescents?

As Table 1 shows, early-exposed adolescents were well matched

to non-early-exposed adolescents following propensity-score

matching. Prior to propensity-score matching, the standardized

bias (SB) between groups ranged from �8% to 85% across the background covariates; the average SB was 17%. After propen-

sity-score matching, the SB between groups ranged from �6% to 8% across the background covariates, and the average SB

was 0%. We obtained a similar reduction in SB within the

conduct-problem subgroups after propensity-score matching:

TABLE 1

Risk Factors in Adolescents With Versus Without Early Substance Exposure, Before and After Propensity-Score

Matching

Risk factor

Before propensity-score matching After propensity-score matching

Mean score or prevalence Mean score or prevalence

No early exposure

Early exposure

SB (%)

No early exposure

Early exposure

SB (%)(n 5 813) (n 5 114) (n 5 342) (n 5 114)

Family history of alcohol or drug

disorder 0.14 0.21 43n 0.20 0.21 5

Conduct problems, age 7 1.65 1.90 16 1.97 1.90 �5 Conduct problems, age 9 1.47 1.75 18n 1.83 1.75 �6 Conduct problems, age 11 1.36 1.72 23n 1.79 1.72 �6 Conduct problems, age 13 1.31 2.49 85n 2.45 2.49 3

Parent criminal conviction (%) 25.0 31.8 16 29.3 31.8 6

Low socioeconomic status (%) 20.0 20.0 0 17.8 20.0 7

Maltreatment (%) 8.4 10.0 6 10.0 10.0 1

Low mother’s IQ (%) 14.1 18.1 11 15.7 18.0 8

Low child IQ (%) 13.1 15.2 6 16.3 15.2 �4 Undercontrolled temperament (%) 10.7 8.2 �8 9.1 8.2 �4 ADHD (%) 6.1 7.2 5 7.9 7.2 �3 Male (%) 50.6 51.8 3 53.3 51.8 �4 Average — — 17 — — 0

Note. The sample included 927 adolescents, categorized as early-exposed (i.e., exposed to alcohol or drugs on multiple occasions before age 15) and non-early-exposed (i.e., not exposed to alcohol or drugs on multiple occasions before age 15). Standardized bias (SB) between the treatment (exposed) group and the control (unexposed) group (subscripts T and C, respectively) was computed as follows: SB 5 MT � MC/

p [sT

2 1 sC 2)/2]. Negative values of SB indicate greater risk in the non-early-exposed group than in the early-exposed group. ADHD 5

attention-deficit/hyperactivity disorder. np < .10.

1040 Volume 19—Number 10

Early Exposure to Drugs and Alcohol

Within the no-conduct-problem subgroup, the average SB for

early- versus non-early-exposed adolescents was reduced from

2% to 1%, and within the conduct-problem subgroup, the aver-

age SB was reduced from 11% to 0%. For all analyses, the SB was

reduced to below 20% for each of the 13 covariates following

propensity-score matching, an indication of a high degree of

similarity in the distributions of the background covariates.

Does Early Substance Exposure Predict Poor Adult

Outcomes for Adolescents After Propensity-Score

Matching?

As Table 2 shows, early-exposed adolescents were at an increased

risk for the adult outcomes of substance dependence, herpes in-

fection, early pregnancy, failure to obtain educational qualifica-

tions, and criminal convictions. The adjusted effects after

propensity-score matching remained statistically significant for

all of these outcomes. Early-exposed adolescents were approxi-

mately 2 to 3 times more likely than non-early-exposed adoles-

cents to be substance dependent, to have herpes infection, to have

had an early pregnancy, and to have failed to obtain educational

qualifications; early-exposed adolescents also had significantly

more criminal convictions than non-early-exposed adolescents.

Does Early Substance Exposure Influence the Majority of

Ordinary Adolescents With No Prior History of Conduct

Problems?

As expected, early substance exposure was not a random event;

adolescents with a conduct-problem history were 2 times (odds

ratio 5 2.1, confidence interval 5 1.4–3.1) more likely to be

exposed to illicit substances prior to age 15, compared with ad-

olescents without a conduct-problem history. More specifically,

17.0% of adolescents with a conduct problem history, versus

9.1% of adolescents with no conduct-problem history, experi-

enced early substance exposure. Fifty-six of the 114 early-

exposed adolescents were assigned to the no-conduct-problem

subgroup; this means that approximately 50% of adolescents

exposed to substances prior to age 15 had no prior history of

conduct problems.

Table 3 presents the group-specific effects of early substance

exposure on adolescents’ adult outcomes. Adjusted effects after

propensity-score matching are presented separately for

adolescents with versus without a conduct-problem history. The

results in the table illustrate two main findings. First, after

propensity-score matching, ordinary adolescents without a

conduct-problem history were at an increased risk for adult

substance dependence, herpes infection, early pregnancy, and

criminal convictions if they had experienced early substance

exposure, although early substance exposure did not increase

risk for not finishing school in this subgroup. Second, early

substance exposure further elevated the risk for adult substance

dependence, early pregnancy, failure to obtain educational

qualifications, and criminal convictions—but not herpes in-

fection—among adolescents who were already at risk because of

their conduct-problem history.

Sensitivity Analyses

Are the effects of early substance exposure different for alcohol

versus illicit drugs? The substance most frequently used by

TABLE 2

Effects of Early Substance Exposure on Adolescents’ Adult Outcomes, Before and After Propensity-Score Matching

Adult outcome

Before propensity-score matching After propensity-score matching

Mean risk Mean risk

No early exposure

Early exposure Effect

size

No early exposure

Early exposure Effect

size(n 5 813) (n 5 114) (n 5 342) (n 5 114)

Substance dependence at age 32 (%) 11.1 28.8 3.25nn 14.7 28.8 2.25nn

(2.04–5.18) (1.35–3.73)

Herpes infection at age 32 (%) 17.0 28.0 1.90nn 16.1 28.0 2.02nn

(1.18–3.07) (1.18–3.44)

Early pregnancy (prior to age 21) a (%) 18.1 44.7 3.65nn 22.5 44.7 2.78nn

(1.93–6.86) (1.38–5.57)

No educational qualifications by age 32 (%) 14.7 32.1 2.74nn 19.3 32.1 1.97nn

(1.76–4.27) (1.22–3.19)

Number of criminal convictions between

ages 17 and 32 b

1.31 7.12 5.35nn 1.74 7.12 3.95nn

(2.85–10.05) (2.23–6.97)

Note. Study members were categorized as early-exposed (i.e., exposed to alcohol or drugs on multiple occasions before age 15) versus non- early-exposed (i.e., not exposed to alcohol or drugs on multiple occasions before age 15). The reported effect sizes are odds ratios for all outcomes except number of criminal convictions, for which incidence-rate ratios are reported. The numbers in parentheses are 95% confidence intervals. All findings remained statistically significant after adjustment of standard errors via bootstrapping in PSMATCH2. a Early pregnancy was estimated for females only and was defined as having at least one pregnancy prior to age 21; sex-specific propensity scores were used for this female-only analysis.

b Negative binomial regressions were applied to model incidence-rate ratios for the number

of criminal convictions. nnp < .05.

Volume 19—Number 10 1041

C.L. Odgers et al.

adolescents in our cohort was alcohol, and the vast majority of

early-exposed adolescents were exposed to alcohol. However,

65 of these adolescents were exposed to alcohol only and did not

use drugs. We tested whether adolescents exposed only to

alcohol differed from adolescents with no early substance ex-

posure. After propensity-score matching, early alcohol exposure

alone predicted a cumulative index of poor adult outcomes,

including substance dependence, herpes infection, failure to

obtain educational qualifications, and criminal convictions

(Cohen’s d 5 0.51, p < .01).

The remaining early-exposed adolescents (n 5 49) were ex-

posed to multiple substances. This reflects the reality of ado-

lescent substance use: It is extremely rare for a young adolescent

to specialize in the use of cannabis, inhalants, or any other illicit

drug. Rather, adolescents who use illicit drugs typically use

alcohol as well. After propensity-score matching, early poly-

substance exposure predicted a cumulative index of poor adult

outcomes (Cohen’s d 5 1.15, p <.01). Thus, although early

alcohol exposure alone was a significant predictor of adult

risk, early poly-substance exposure was associated with more

pronounced risk.

DISCUSSION

This application of propensity-score matching within a 30-year

prospective study helps to advance what is known about the

effects of early substance exposure in two ways. First, the prior

consensus in child psychology and psychiatry has been that

adolescents who go on to develop substance dependence are not

normal adolescents who are experimenting with substances, but

rather are highly likely to be adolescents with a prior history of

conduct problems (Armstrong & Costello, 2002). If this is the

case, the documented association between early substance

exposure and adult outcomes would not be due to exposure per

se, but instead would be the result of who is exposed (Wells,

Horwood, & Fergusson, 2004). Prior research has not resulted

in a consensus regarding the causal status of substance exposure

(Agrawal, Neale, Prescott, & Kendler, 2004; Kandel, 2003;

Lynskey et al., 2003; Prescott & Kendler, 1999). However,

results from this study are consistent with a causal effect of early

substance exposure among adolescents with no prior history

of conduct problems. That is, early-exposed adolescents with

no conduct-problem history, although they did not have an in-

creased risk of failing to complete school, were more likely than

their matched non-early-exposed counterparts to develop sub-

stance dependence, test positive for herpes, have an early

pregnancy, and be convicted of criminal offenses.

Second, findings from this prospective study support a causal

link between early substance exposure and a wide range of adult

outcomes. Propensity-score-adjusted effects indicate that early

substance exposure more than doubles the odds of adult

substance dependence, herpes infection, early pregnancy, and

criminal convictions. With a few notable exceptions (Brook,

Brook, Zhang, Cohen, & Whiteman, 2002; Dooley, Prause,

Ham-Rowbottom, & Emptage, 2005; Fergusson & Horwood,

1997; Wells et al., 2004), most prior tests of the association

between early substance exposure and adult outcomes have

been based on cross-sectional surveys of adults, which are

TABLE 3

Propensity-Adjusted Effects of Early Substance Exposure on Adolescents’ Adult Outcomes, for Adolescents With

Versus Without a Conduct-Problem History

Adult outcome

No conduct-problem history Some conduct-problem history

Mean risk Mean risk

No early exposure

Early exposure Effect

size

No early exposure

Early exposure Effect

size(n 5 168) (n 5 56) (n 5 165) (n 5 55)

Substance dependence at age 32 (%) 7.6 23.2 3.64nn 15.3 32.0 2.60nn

(1.57–8.45) (1.24–5.46)

Herpes infection at age 32 (%) 15.4 32.7 2.66nn 20.5 23.4 1.18

(1.26–5.60) (0.54–2.60)

Early pregnancy (prior to age 21) a (%) 12.3 34.4 3.75nn 35.9 69.2 4.02nn

(1.45–9.69) (1.04–15.46)

No educational qualifications by age 32 (%) 10.9 14.2 1.37 28.7 49.0 2.38nn

(0.56–3.34) (1.24–4.57)

Number of criminal convictions between

ages 17 and 32

0.65 1.59 2.86nn 2.52 13.12 4.96nn

(1.2–6.6) (2.5–10.0)

Note. Study members were categorized as early-exposed (i.e., exposed to alcohol or drugs on multiple occasions before age 15) and non- early-exposed (i.e., not exposed to alcohol or drugs on multiple occasions before age 15). The reported effect sizes are odds ratios for all outcomes except number of criminal convictions, for which incidence-rate ratios are reported. The numbers in parentheses are 95% confidence intervals. All findings remained statistically significant after adjustment of standard errors via bootstrapping in PSMATCH2. a The within-group treatment effects for pregnancy prior to age 21 are exploratory because each subgroup contained fewer than 50 early- exposed females. nnp < .05.

1042 Volume 19—Number 10

Early Exposure to Drugs and Alcohol

limited to retrospective recall of both childhood behaviors and

age of onset for substance use.

This study has limitations. First, although propensity-score

matching is designed to mimic the desirable features of ran-

domized control trials within observational studies, the lack

of random assignment to a treatment condition (which in this

case would have involved administering illicit drugs to children)

restricts an ideal test of causal association. Second, children

with conduct problems are a heterogeneous group; future studies

are required to estimate the impact of early substance exposure

separately for distinct subtypes of children with conduct prob-

lems. Third, this study was based on a single New Zealand

cohort and requires replication across ethnicities and cultures.

Fourth, the culture of drug and alcohol use among adolescents

has shifted over time. The cohort we studied was exposed to

substances for the first time in 1984 through 1987, which meant

that exposure was restricted mainly to alcohol and cannabis.

Ideally, a study would isolate the effects of specific types of

drugs on adolescents’ future lives. However, the reality is that,

aside from adolescents who are exposed to alcohol only, ado-

lescents exposed to substances prior to age 15 do not specialize

in their substance use. Only 1.2% of the cohort restricted their

substance use to cannabis prior to age 15, and less than 1% used

inhalants or other drugs exclusively. Therefore, we could not

examine the specific effects of cannabis, inhalants, or other

drugs separately, as the necessary groups of adolescents did not

exist. Research needs to examine the effects of other substances

that are now used more frequently by adolescents (e.g., cocaine,

ecstasy). Future studies should also be designed to include

a more rigorous assessment of early substance exposure based

on multiple informants and finer-grained measurement of the

precise timing and dosage of exposure.

With these limitations in mind, the implications of these re-

sults for public policy and prevention can be noted. With respect

to public policy, out results are consistent with the U.S. Surgeon

General’s call to action to prevent early substance exposure.

Results support the policy position that early substance expo-

sure poses independent risks for adolescents’ future lives; these

risks extend beyond the development of later substance de-

pendence, to include risks for herpes infection, early pregnancy,

and crime. With respect to prevention, it is important to note that

in our study, 50% of the adolescents exposed to substances

before age 15 did not have a prior history of conduct problems,

yet they experienced many of the same ill consequences as their

peers who were already at risk. This means that prevention

efforts should not focus solely on at-risk or conduct-problem

adolescents. Moreover, this research provides a rationale for

evaluating the cost-benefit ratios of substance-use prevention

programs separately for adolescents with versus without conduct

problems. In short, universal interventions are required to

ensure that all children—not only those entering early adoles-

cence on at at-risk trajectory—receive an adequate dose of

prevention.

Acknowledgments—This work was supported by the U.S.

National Institute of Mental Health (Grants MH45070 and

MH49414), the U.K. Medical Research Council (G0100527),

the William T. Grant Foundation, the Health Research Council

of New Zealand, and the National Institute on Drug Abuse

(Grant 1 P20 DA017589, awarded to the Duke University

Transdisciplinary Prevention Research Center). We thank the

Dunedin Multidisciplinary Health and Development Study

members and their families, the study unit’s research staff, and

study founder Phil Silva. We also thank the National Consortium

on Violence Research and Director Alfred Blumstein, as well

as the Center for Child and Family Policy and Director Kenneth

Dodge, for facilitating this research.

REFERENCES

Agrawal, A., Neale, M.C., Prescott, C.A., & Kendler, K.S. (2004). A

twin study of early cannabis use and subsequent use and abuse/

dependence of other illicit drugs. Psychological Medicine, 34, 1227–1237.

American Psychiatric Association. (1980). Diagnostic and statistic manual of mental disorders (3rd ed.). Washington, DC: Author.

American Psychiatric Association. (1994). Diagnostic and statistical manual of mental disorders (4th ed.). Washington, DC: Author.

Armstrong, T.D., & Costello, E.J. (2002). Community studies on

adolescent substance use, abuse, or dependence and psychiatric

comorbidity. Journal of Consulting and Clinical Psychology, 70, 1224–1239.

Brook, D.W., Brook, J.S., Zhang, C.S., Cohen, P., & Whiteman, M.

(2002). Drug use and the risk of major depressive disorder,

alcohol dependence, and substance use disorders. Archives of General Psychiatry, 59, 1039–1044.

Caspi, A., McClay, J., Moffitt, T.E., Mill, J., Martin, J., Craig, I.W., et al.

(2002). Role of genotype in the cycle of violence in maltreated

children. Science, 297, 851–854. Caspi, A., & Silva, P.A. (1995). Temperamental qualities at age three

predict personality traits in young adulthood: Longitudinal

evidence from a birth cohort. Child Development, 66, 486–498. Costello, A., Edelbrock, C., Kalas, R., Kessler, M., & Klaric, S.A.

(1982). Diagnostic Interview Schedule for Children (DISC). Rockville, MD: National Institute of Mental Health.

D’Agostino, R.B., & D’Agostino, R.B. (2007). Estimating treatment

effects using observational data. Journal of the American Medical Association, 297, 314–316.

Dooley, D., Prause, J., Ham-Rowbottom, K.A., & Emptage, N. (2005).

Age of alcohol drinking onset: Precursors and the mediation of

alcohol disorder. Journal of Child & Adolescent Substance Abuse, 15, 19–37.

Eberhart-Phillips, J.E., Dickson, N.P., Paul, C., Herbison, G.P., Taylor,

J., & Cunningham, A.L. (2001). Rising incidence and prevalence

of herpes simplex type 2 infection in a cohort of 26 year old New

Zealanders. Sexually Transmitted Infections, 77, 353–357. Elley, W.B., & Irving, J.C. (1976). Revised socio-economic index for

New Zealand. New Zealand Journal of Educational Studies, 11, 25–36.

Ellickson, P.L., Tucker, J.S., & Klein, D.J. (2003). Ten-year pro-

spective study of public health problems associated with early

drinking. Pediatrics, 111, 949–955.

Volume 19—Number 10 1043

C.L. Odgers et al.

Elliott, D.S., Huizinga, D., & Menard, S. (1989). Multiple problem youth: Delinquency, substance use, and mental health problems. New York: Springer-Verlag.

Fergusson, D.M., & Horwood, L.J. (1997). Early onset cannabis use

and psychosocial adjustment in young adults. Addiction, 92, 279– 296.

Grant, B.F., & Dawson, D.A. (1997). Age at onset of alcohol use and its

association with DSM-IValcohol abuse and dependence: Results

from the National Longitudinal Alcohol Epidemiologic Survey.

Journal of Substance Abuse, 9, 103–110. Haviland, A.M., Nagin, D.S., & Rosenbaum, P.R. (2007). Combining

propensity score matching and group-based trajectory analysis in

an observational study. Psychological Methods, 12, 247–267. Hingson, R.W., Heeren, T., & Winter, M.R. (2006). Age at drinking onset

and alcohol dependence: Age at onset, duration, and severity.

Archives of Pediatrics & Adolescent Medicine, 160, 739–746. Ho, D.W.T., Field, P.R., Irving, W.L., Packham, D.R., & Cunningham,

A.L. (1993). Detection of immunoglobulin-M antibodies to

glycoprotein G-2 by western-blot (Immunoblot) for diagnosis of

initial herpes-simplex virus Type-2 genital infections. Journal of Clinical Microbiology, 31, 3157–3164.

Kandel, D.B. (2003). Does marijuana use cause the use of other drugs?

Journal of the American Medical Association, 289, 482–483. King, K.M., Meehan, B.T., Trim, R.S., & Chassin, L. (2006). Marker or

mediator? The effects of adolescent substance use on young adult

educational attainment. Addiction, 101, 1730–1740. Leuven, E., & Sianesi, B. (2003–2006). PSMATCH2: Stata module to

perform full Mahalanobis and propensity score matching, com-

mon support graphing, and covariate imbalance testing [Com-

puter software]. Retrieved June 2006 from http://ideas.repec.org/

c/boc/bocode/s432001.html

Lynskey, M.T., Heath, A.C., Bucholz, K.K., Slutske, W.S., Madden,

P.A.F., Nelson, E.C., et al. (2003). Escalation of drug use in early-

onset cannabis users vs co-twin controls. Journal of the American Medical Association, 289, 427–433.

Maynard, R.A. (Ed.). (1996). Kids having kids: The economic costs and social consequences of teen pregnancy. Washington, DC: Urban Institute Press.

Miech, R.A., Caspi, A., Moffitt, T.E., Wright, B.R.E., & Silva, P.A.

(1999). Low socioeconomic status and mental disorders: A lon-

gitudinal study of selection and causation during young adult-

hood. American Journal of Sociology, 104, 1096–1131. Moffitt, T.E. (2006). Life-course-persistent and adolescent-limited

antisocial behavior. In D. Cicchetti & D.J. Cohen (Eds.), Devel- opmental psychopathology: Vol. 3. Risk, disorder, and adaptation (pp. 570–598). New York: John Wiley & Sons.

Moffitt, T.E., Caspi, A., Rutter, M., & Silva, P.A. (2001). Sex differences in antisocial behaviour: Conduct disorder, delinquency, and vio-

lence in the Dunedin Longitudinal Study. New York: Cambridge University Press.

Odgers, C.L., Milne, B., Caspi, A., Crump, R., Poulton, R., & Moffitt,

T.E. (2007). Predicting prognosis for the conduct-problem boy:

Can family history help? Journal of the American Academy of Child and Adolescent Psychiatry, 46, 1240–1249.

Odgers, C.L., Moffitt, T.E., Broadbent, J.M., Dickson, N.P., Hancox, R.,

Harrington, H., et al. (2008). Female and male antisocial tra-

jectories: From childhood origins to adult outcomes. Development and Psychopathology, 20, 673–716.

Peele, S. (2007). Addiction-proof your child. New York: Random House/Three Rivers Press.

Prescott, C.A., & Kendler, K.S. (1999). Age at first drink and risk

for alcoholism: A noncausal association. Alcoholism: Clinical and Experimental Research, 23, 101–107.

Robins, L.N., Cottler, L., Bucholz, K.K., & Compton, W. (1995). Di- agnostic Interview Schedule for DSM-IV. St. Louis, MO: Wash- ington University School of Medicine.

Rosenbaum, P.R., & Rubin, D.B. (1985). Constructing a control-group

using multivariate matched sampling methods that incorporate

the propensity score. American Statistician, 39, 33–38. Stuart, E.A., & Green, K.M. (2008). Using full matching to estimate

causal effects in nonexperimental studies: Examining the rela-

tionship between adolescent marijuana use and adult outcomes.

Developmental Psychology, 44, 395–406. Stueve, A., & O’Donnell, L.N. (2005). Early alcohol initiation and

subsequent sexual and alcohol risk behaviors among urban

youths. American Journal of Public Health, 95, 887–893. Thurstone, T.G., & Thurstone, L.L. (1973). The SRA verbal form.

Chicago: Science Research Associates.

U.S. Department of Health and Human Services. (2007). The Surgeon General’s call to action to prevent and reduce underage drinking. Washington, DC: U.S. Department of Health and Human Ser-

vices, Office of the Surgeon General.

Wechsler, D. (1974). Wechsler Intelligence Scale for Children—Revised. New York: Psychological Corp.

Weissman, M.M., Wickramaratne, P., Adams, P., Wolk, S., Verdeli, H.,

& Olfson, M. (2000). Brief screening for family psychiatric

history: The family history screen. Archives of General Psychiatry, 57, 675–682.

Wells, J.E., Horwood, L.J., & Fergusson, D.M. (2004). Drinking

patterns in mid-adolescence and psychosocial outcomes in late

adolescence and early adulthood. Addiction, 99, 1529–1541.

(RECEIVED 2/6/08; REVISION ACCEPTED 4/15/08)

1044 Volume 19—Number 10

Early Exposure to Drugs and Alcohol