Addiction (2000) 95(11), 1679– 1690
RESEARCH REPORT
Initiation and progression of cannabis use in a
population-based Australian adolescent
longitudinal study
C. COFFEY,1 M. LYNSKEY,2 R. WOLFE3 & G. C. PATTON1
1
Centre for Adolescent Health, Department of Paediatrics, University of Melbourne, 2National
Drug and Alcohol Research Centre, University of New South Wales, Sydney & 3Clinical
Epidemiology and Biostatistics Unit, Royal Children’s Hospital and Department of Paediatrics,
University of Melbourne, Australia
Abstract
Aims. To examine predictors of cannabis use initiation, continuity and progression to daily use in
adolescents. Design. Population-based cohort study over 3 years with 6 waves of data collection. Participants. 2032 students, initially aged 14– 15 years, from 44 secondary schools in the state of Victoria,
Australia. Measurements. Self-report cannabis use was categorized on four levels (none, any, weekly,
daily) and summarized as mid-school (waves 2/3) and late-school (waves 4/5/6) use. Background, school
environment, mid-school peer use and individual characteristics were assessed. Findings. Peer cannabis use,
daily smoking, alcohol use, antisocial behaviour and high rates of school-level cannabis use were associated
with mid-school cannabis use and independently predicted late-school uptake. Cannabis use persisted into
late-school use in 80% of all mid-school users. Persisting cannabis use from mid- to late-school was more likely
in regular users (odds ratio (OR) 3.4), cigarette smokers (OR any smoking: 2.0, daily smoking: 3.3) and
those reporting peer use (OR 2.1). Mid-school peer use independently predicted incident late-school daily use
in males (OR 6.5) while high-dose alcohol use (OR 6.1) and antisocial behaviour (OR 6.6) predicted
incident late-school daily use in females. Conclusions. Most cannabis use remained occasional during
adolescence but escalation to potentially harmful daily use in the late-school period occurred in 12% of early
users. Transition was more likely in males, for whom availability and peer use were determinants. In contrast,
females with multiple extreme behaviours were more likely to become daily users. Cigarette smoking was an
important predictor of both initiation and persisting cannabis use.
Introduction
There is concern about cannabis use by young
people in most developed countries (Adlaf &
Smart, 1991; Fergusson, Lynskey & Horwood,
1993; Johnston, OMalley & Bachman 1998;
Hall, Johnston & Donnelly, 1999; Lynskey &
Hall, 1999). Cannabis use is typically initiated
during adolescence with patterns of heaviest use
usually occurring during late adolescence and
young adulthood (Chen & Kandel, 1995).
Correspondence to: Ms Carolyn Coffey, Centre for Adolescent Health, 2 Gatehouse Street, Parkville 3052,
Australia
Submitted 31st January 2000; initial review completed 7th April 2000; nal version accepted 9th June 2000.
ISSN 0965– 2140 print/ISSN 1360-0443 online/00/111679 – 12 Ó Society for the Study of Addiction to Alcohol and Other Drugs
Carfax Publishing, Taylor & Francis Ltd
DOI: 10.1080/01439680020000911
1680
C. Coffey et al.
1st sample
N1 =1037
2nd sample
N2 =995
Total intended sample = N1 + N2 = 2032
Total achieved sample = 1947 (96%)
Wave 1
n1 =898
(87%)
late 1992
Wave 2
n2 =1728
(85%)
early 1993
Wave 3
n3 =1699
(84%)
late 1993
Wave 4
n4 =1629
(80%)
early 1994
Wave 5
n5 =1576
(78%)
late 1994
Wave 6
n6 =1530
(75%)
early 1995
Figure 1. Participation rates of 2032 secondary school students in the adolescent health cohort study in Victoria, Australia.
Controversy remains about the extent of the
harmful social and health consequences of
occasional use of this drug. Debate has been
polarized between those who argue that adolescent cannabis use is essentially a benign, transient practice with few social and health
consequences for the great majority of young
people (Shedler & Block, 1990; Robins, 1995)
and those who view cannabis as having the
potential to lead to escalating drug use and its
attendant problems (Kandel et al., 1986; Newcomb & Bentler, 1988; Fergusson, Lynskey &
Horwood, 1996; Hall, 1997). Its peak use also
coincides with the time of greatest risk for
adverse effects of substance use such as accidental injury, educational and legal difculties (Hall,
1995).
Most information on the risk factors for cannabis use derive from cross-sectional and retrospective studies. These studies have generated
useful hypotheses but the processes involved can
only be explored longitudinally, that is, with
prospective measurement at multiple time-points
of drug use and putative risk factors (Kandel,
1980; Farrington, 1991; Cicchetti & Rogosch,
1999). Longitudinal studies beginning early in
life have identied childhood and early adolescent risk factors for cannabis use, but
infrequent observations during the adolescent
years have limited the ability of these studies to
clarify risk processes around mid- to late teens, a
period of rapid change in drug use behaviour.
Well documented risk factors for licit and illicit
substance use include ready substance availability together with afliation with drug-using
peers (Dembo et al., 1979; Kandel & Andrews,
1987; Maddahian, Newcomb & Bentler, 1988),
but predictors of more regular use have been less
explored than those for initial uptake. Further,
few investigators have distinguished between
occasional/experimental use and more regular
use, thereby being insensitive to the possibility
that risk factors for the two levels may differ.
The aims of this report are to use data from a
3-year prospective study of a representative sample of Australian adolescents to quantify the
correlates of early cannabis use and to quantify
risk factors for incident use, continuation and
progression in use.
Method
Procedure and sample
Data were collected from subjects in a 6-wave
cohort study of adolescent health performed
throughout the state of Victoria, Australia
between August 1992 and July 1995. The cohort
was dened using a two-stage sampling procedure. At stage 1, 45 schools were selected from
a stratied frame of government, catholic and
independent schools (total number of students
60 905). One school from the initial cross-sectional survey was unavailable for the cohort
study leaving a total of 44 schools. At the second
stage, a single intact class was randomly selected
from each school and these students were measured in wave 1. At the second wave of data
collection, 6 months later, when the cohort had
moved into year 10, a second intact class from
the same grade at each participating school was
selected at random (Fig. 1). Thus half the participants had been interviewed once before wave 2.
The entire sample was followed-up from wave 2
to completion of the study.
The study was presented as dealing with
important adolescent health issues and covered
both adolescent mental health and life-style.
Written parental permission was sought at entry
into the study. Subjects completed the questionnaire at intervals of 6 months between year levels
Natural history of adolescent cannabis use
9 and 12 (6 waves). The mean age at wave 1 was
14.5 (SD 0.5) years and at wave 6, 17.4 years
(SD 0.4). The survey was administered at school
using 28 laptop computers which allowed the
collection of detailed self-report data through the
use of branched questionnaires (Paperny et al.,
1990). Subjects who were unavailable for followup at school were interviewed by telephone. The
proportion of interviews conducted by telephone
increased from 2% in wave 2 to 14% in wave 6.
Measures
Cannabis use
Assessment of cannabis use was based on self-reported frequency. Participants described their
cannabis use during the past 6 months using the
following rating scale: (1) never used, (2) not
used in the past 6 months, (3) a few times, (4)
monthly, (5) weekly and (6) daily. Cannabis use
was summarized over two periods of the study:
the highest reported level of cannabis use in
waves 2 and 3, and similarly in waves 4, 5 and 6.
These intervals correspond to the third last year
at school, and the last 2 years of school. For
convenience, these intervals are referred to as
“mid-school” and “late-school”, respectively, although the second interval contained data from
219 (11%) participants who had left school before their nal year.
Background and putative risk factors
A wide range of social, demographic, peer and
individual factors were examined as possible predictors of cannabis use. These were selected on
the basis of prior review of the literature which
identied factors most likely to be related to
cannabis use and subject to availability within
our data. The factors included were:
Demographic variables
These were assessed at study entry and included
gender, place of birth, metropolitan or rural location of school and parental separation or divorce. However, rural school location was not
associated with any cannabis use variable and so
was dropped from all outcome analyses.
Peer cannabis use
At each wave, participants reported whether (1)
1681
none, (2) some or (3) most of their friends used
cannabis. This variable was summarized over the
mid-school period so that those reporting in at
least one wave that most of their friends used
cannabis were characterized accordingly.
School level of cannabis use
In order to examine early exposure to regular
cannabis use at school, the proportion of students within each school using cannabis at least
weekly was calculated at wave 2. The schools
were then divided into tertiles on the basis of
these proportions. In all analyses of late-school
cannabis use with the three-level variable describing school-level exposure, only the highest
category held a univariate association (if at all)
with the outcome variable. Therefore the binary
variable, top tertile vs. middle or bottom tertile,
was used in each analysis.
Cigarette smoking
Participants reporting that they had smoked on 6
or 7 days in the previous week were categorized
as daily smokers. If daily smoking was recorded
in either waves 2 or 3 then the individual was
characterized as a daily smoker during the midschool period (291 of the 1890 participants). For
more detailed analysis of the effects of smoking,
occasional smoking was dened as reporting
smoking in the last month, but less than 6 days
in the past week. Non-smoking was dened as
not having smoked in the past month.
Alcohol consumption
Subjects reporting that they had drunk alcohol in
the week before the survey were asked to complete a 1-week retrospective alcohol diary (beverage- and quantity-specic). Two measures of
alcohol consumption were derived from the diary
in waves 2 and 3:
(1) Those who reported drinking on three or
more days in the previous week in either
wave 2 or 3 were classied as frequent
drinkers in the mid-school period (123 of
1890 participants).
(2) Subjects were characterized by their average
consumption of ethanol per drinking day
(one unit equivalent to one standard drink,
1682
C. Coffey et al.
9 g ethanol). Those with an average of ve
units or greater were classied as high dose
drinkers (312 of 1890 participants).
Antisocial behaviour
Antisocial behaviours were evaluated with 10
items from the Moftt & Silva (1988) selfreport early delinquency scale. Items included
antisocial behaviour relating to property damage
(vandalism, car damage, making grafti), interpersonal conict (ghting, carrying weapons,
running away from home, expulsion from
school) and theft (stealing property from parents, or other, stealing cars). Items concerning
alcohol or other substance use were not
included. The reference period was 6 months.
Antisocial behaviours were categorized according to whether more than one behaviour was
endorsed “more than once” in order to distinguish participants with more global antisocial
behaviours. If this occurred in either wave 2 or
wave 3, individuals were characterized as displaying antisocial behaviour in the mid-school
period (240 of 1890 participants).
Mental health
A computerized form of the Clinical Interview
Schedule (CIS-R) was used to rate psychiatric
morbidity (Lewis & Williams, 1989; Lewis et
al., 1992). This is a structured psychiatric interview designed for assessing symptoms of general
psychiatric morbidity in non-clinical populations
and includes indicators of depression and anxiety. The instrument generates 14 subscales
which can then be added to form a scale indicating the degree of psychiatric morbidity.
Mean scores for waves 2 and 3 were calculated
and then dichotomized at the 11/12 cut-point,
corresponding to the level at which a general
practitioner might be concerned about a subject’s mental health (Lewis & Williams, 1989;
Lewis et al., 1992). Thirty-two per cent of
females and 15% of males scored above this
threshold.
Data analysis
Data analysis was undertaken using Stata (StataCorp, 1999). Initially, cannabis use was
assessed using a three-category ordinal scale:
(1) not used in previous 6 months, (2) used in
the last 6 months but less often than weekly
and (3) weekly or more regular use. We considered two alternative ways of analysing this
data. The rst alternative was to dichotomize
cannabis use as: (1) versus (2)– (3); or (1)– (2)
versus (3), and then to examine separate logistic
regression models tted to these dichotomous
outcomes. This approach would have resulted
in two different odds ratio (OR) estimates of
the association of a factor with cannabis use. A
marked difference between these OR would
indicate that the association was different at different parts of the ordinal scale. If the underlying association with cannabis use that we were
trying to estimate was, in fact, the same across
the ordinal scale (i.e. the underlying OR were
equal) then this analysis method would be
inefcient and would ignore some of the information from the three-category scale. To
optimize efciency we used the alternative
strategy of tting ordinal logistic regression
models. Within these models, it was possible to
perform likelihood-ratio (LR) tests (Peterson &
Harrell, 1990) of the assumption of a factor’s
association with cannabis use being constant
across the ordinal scale (the proportional odds
(PO) assumption (McCullagh, 1980)). All variables in the multivariable ordinal models
included in this report complied with the proportional odds assumption at the 0.05 level of
signicance.
Exploratory univariate analyses were performed followed by multivariable ordinal logistic regression modelling. First-order interactions
with gender were tested in all models using the
LR test comparing the more complex model
with the simpler model. All reported condence
intervals (CI) are based on a 95% condence
level.
Other analyses performed were on the binary
outcomes: poor survey completion, late-school
daily use and persistence from early to lateschool use. These analyses used multivariate
logistic regression. In the case of the predictive
model for daily cannabis use, backwards stepwise selection was used to examine interaction
terms with gender, keeping all main terms in
the model. Items were dropped if p . 0.2 and
reincluded if p , 0.1. A similar process was
then used in the selected model in order to
examine the main terms, dropping terms if
p . 0.1, and reincluding if p , 0.05.
Natural history of adolescent cannabis use
Results
Sample characteristics
From the total sample of 2032 students on class
registers, 1947 (95.8%) completed the questionnaire at least once in the course of the study.
Based on the intended sample, response rates
across waves were as follows: wave 1, 87%; wave
2, 85%; wave 3, 84%; wave 4, 80%; wave 5,
78%; and wave 6, 75%. The gender ratio of the
cohort (males 47.0%) was similar to that in
Victorian schools at the time of sampling (Australian Bureau of Statistics, 1993). A total of
1890 (93%) young people participated in waves
2– 6. The mean age at wave 2 was 15.4 (SD 0.5)
years and at completion of the follow-up was
17.4 years (SD 0.4).
Two hundred and three subjects (11%) completed only one or two waves between waves 2
and 6. Characteristics of these low completers
were examined in a logistic regression model.
Males were over-represented (OR 1.8, 95% CI
1.3– 2.5), as were non-Australian-born subjects
(OR 2.0, CI 1.3– 3.1), those who had experienced parental divorce or separation (OR 2.6, CI
1.8– 3.7) and those who reported using cannabis
at least weekly at study inception (OR 1.9, CI
1.0– 3.5).
Four major outcome analyses were performed
and are shown in Fig. 2. This gure illustrates
one cross-sectional analysis and three prospective analyses that are the subject of this report.
Table 1 shows the frequency of mid-school cannabis users by late-school users, and denes the
observations included in the prospective analyses
(2) to (4) illustrated in Fig. 2.
(1) Mid-school cannabis use
Twenty-one per cent of the 1864 participants in
waves 2 and 3 (24% of males and 18% of
females) reported using cannabis in the midschool period of follow-up (Fig. 2). As daily use
was infrequent we combined this category with
weekly use to generate a three-level variable describing cannabis use: (1) none, (2) less often
than weekly ( , weekly), (3) weekly or more
often (weekly 1 ). Male gender held a modest
univariate association with mid-school cannabis
use, but this association was not sustained after
adjustment for covariates (Table 2). Reported
peer use held the strongest independent association with cannabis use with a greater than 10fold increase in odds. Antisocial behaviours,
1683
daily smoking and high-dose alcohol use were
markedly associated with cannabis use, showing
between three- and ve-fold increases in odds,
while alcohol use on three or more days was only
modestly associated. Having divorced or separated parents showed a slightly elevated univariate risk, which was still evident after adjustment
for possible confounders. There was no evidence
of an association with either psychiatric morbidity or Australian birth after adjustment for confounders.
Analysis Mid-school
level of use
Late-school
level of use
(1) Mid-school cannabis use (cross-sectional)
outcome
no use
no use
< weekly
< weekly
weekly
weekly
daily
daily
(2) Cannabis use initiation
no use
outcome
no use
< weekly
< weekly
weekly
weekly
daily
daily
(3) Continuity of cannabis use
no use
outcome
no use
< weekly
< weekly
weekly
weekly
daily
daily
(4) Daily use initiation
no use
outcome
no use
< weekly
< weekly
weekly
weekly
daily
daily
Figure 2. Description of analyses. Shaded areas indicate
data included in analysis, borders indicate boundaries
between categories, gaps between categories indicate levels of
outcome, and arrows indicate path of transition.
1684
C. Coffey et al.
Table 1. Frequency of mid-school cannabis use by late-school cannabis use. Figures in brackets
are row percentages
Late-school cannabis use
Mid-school
cannabis use
None
, Weekly
Weekly
Daily
Total
None
1153
(85.6)
163
(12.1)
26
( 1.9)
5
( 0.4)
1347
(100)
, Weekly
63
(24.5)
123
(47.9)
61
(23.7)
10
( 3.9)
257
(100)
Weekly
3
( 4.0)
22
(29.3)
28
(37.3)
22
(29.3)
75
(100)
Daily
3
(10.0)
2
(40.0)
8
(35.0)
7
(100)
20
(15.0)
Total
1222
(71.9)
310
(18.3)
123
( 7.2)
44
( 2.6)
1699
(100)
There were 123 non-users, 25 , weekly, 10 weekly and seven daily cannabis users from
the mid-school period who had no late-school observations.
Table 2. Associations with mid-school cannabis use measured on three levels*: OR from
ordinal logistic regression models (n 5 1864)
Univariate
Multivariate
Explanatory variable
OR
95% CI
OR
95% CI
Gender (male vs. female)
Australian birth
Divorced/separated parents
Peer cannabis use
Daily smoking
Alcohol . 2 days per week
High dose drinker
Antisocial behaviours
Psychiatric morbidity
1.4
1.7
2.3
26
11
6.0
8.7
8.6
2.1
1.2– 1.8
1.2– 2.4
1.8– 3.0
19– 35
8.3– 14
4.2– 8.7
6.7– 11
6.5– 11
1.7– 2.7
1.2
1.3
1.5
12
4.7
1.6
3.2
3.9
1.0
0.86– 1.5
0.85– 2.1
1.1– 2.1
8.6– 17
3.5– 6.4
1.0– 2.5
2.3– 4.3
2.8– 5.5
0.76– 1.4
* Levels of cannabis use: none (79%), less than weekly (15%), weekly or more often
(6%). 1. Proportional odds (PO) assumed for all variables and interaction terms. 2.
Overall likelihood-ratio test of PO assumption for multivariable model: v2 (8) 5 7.8;
p 5 0.45.
(2) Prediction of rst cannabis use
Four hundred and forty-four of 1725 late-school
participants (34% of males, 24% of females)
reported cannabis use in the late-school period.
Eighteen per cent reported using less than
weekly, 7% weekly and 2.6% daily. Incident
late-school cannabis use was examined in 1347
individuals who had not reported using cannabis
in the mid-school period and had observations
available in the late-school period (Fig. 2). In the
multivariate ordinal model, peer use, daily smoking, frequent and high-dose alcohol use and antisocial behaviours all predicted cannabis uptake
in the late-school period with between a two-
and three-fold increase in odds (Table 3). Early
exposure to a high level of school cannabis use
was also predictive of subsequent cannabis
initiation. Gender was not associated with late
school initiation. There were no rst order interactions with gender.
(3) Continuity between mid- and late-school any
cannabis use
We dened participants who reported any level
of use in both mid- and late-school as continuing
users. Continuing users (N 5 283, 57% male)
were compared with those reporting mid-school
Natural history of adolescent cannabis use
1685
Table 3. Prediction of late-school cannabis use measured on three levels* for adolescents with no earlier
reports of cannabis use (n 5 1347): OR from ordinal logistic regression models.
Univariate
Multivariate
Explanatory variable
OR
95% CI
OR
95% CI
Gender (male)
Australian birth
Divorced/separated parents
High level of weekly cannabis use in
school at study inception
Mid-school: most peers used cannabis
Mid-school: daily smoker
Mid-school: alcohol . 2 days/week
Mid-school: high dose drinker
Mid-school: antisocial behaviours
Mid-school: psychiatric morbidity
1.4
1.9
1.6
1.8
0.52– 1.0
1.1– 3.3
1.1– 2.5
1.3– 2.4
1.3
1.6
1.4
1.7
0.94– 1.8
0.91– 2.7
0.88– 2.1
1.2– 2.4
2.5
2.9
4.1
3.9
3.4
1.6
1.2– 4.8
1.8– 4.8
2.3– 7.3
2.6– 5.8
2.1– 5.5
1.1– 2.2
2.0
2.3
2.1
2.6
2.3
1.5
1.0– 4.2
1.3– 3.9
1.1– 3.9
1.7– 4.1
1.4– 3.8
1.0– 2.1
* Levels of cannabis use: no use (83%), less than weekly (13%), weekly or more often (4%). 1.
Proportional odds (PO) assumed for all variables. 2. Overall likelihood-ratio test of PO
assumption for nal multivariable model: v2 (10) 5 11.7; p 5 0.31
cannabis use but who reported no subsequent
use (N 5 69, 46% male) (Fig. 2). Seventy-ve
per cent of the 257 , weekly mid-school users
and 94% of 95 weekly 1 mid-school users continued (Table 4). In the initial analysis it was
clear that daily smoking was an important predictor of continued use. In order to examine this
effect further we included mid-school smoking in
the model on three levels: non-smoker (60/83
continued), smoked in the last month (104/129
continued) and daily smoking (161/182 continued). Compared with non-smokers, occasional
smokers were at double the risk of continuation
and daily smokers were at over three times elevated risk, with evidence of a dose effect with
increasing frequency of smoking. More frequent
mid-school cannabis use and peer use were associated with a three-fold and two-fold elevation in
risk, respectively. Although there was evidence of
an interaction between parental divorce and gender (likelihood ratio v2 (1) 5 5.9, p 5 0.015), the
effect of divorce within each gender was not
substantial. The residual gender effect showed
that males were at increased risk of continuing
after allowing for this interaction (males to
females adjusted OR 2.6, 1.3– 5.6). Interaction
between gender and mid-school level of cannabis
use could not be tested due to the small number
of weekly 1 users who discontinued. There were
no other signicant rst order interactions with
gender.
(4) Daily cannabis use
Young people reporting daily cannabis use were
considered to be at high risk of harmful and
dependent patterns of use so we were particularly interested in patterns of continuity and progression to daily use. Forty-four young people
(3.7% of males and 1.7% of females) of the 1699
with observations in both periods reported using
cannabis daily in late-school (another two had
late-school but no mid-school observations)
(Table 1). Only ve of these had not reported
some mid-school use. Twelve per cent of all
mid-school users (25/192 males and 14/146
females) reported late-school daily use, constituting 4% of , weekly mid-school users and
31% of weekly 1 mid-school users. There was
strong evidence of a dose– response relationship
between late-school daily use and level of midschool use after adjustment for confounders
(adjusted OR: less than weekly use mid-school
4.4, 1.3– 15; weekly use mid-school 27, 7.0– 1.5;
daily use mid-school 25, 4.3– 142).
Prediction of initiation into late-school
daily cannabis use
The onset of daily cannabis use was examined in
those participants not previously reporting daily
cannabis use in the mid-school period (Fig. 2).
There were 37 reports (24 males) of incident
late-school daily cannabis use (male versus.
1686
C. Coffey et al.
Table 4. Prediction of continuation of cannabis use from mid-school into late-school (n 5 283) for those
adolescents reporting earlier cannabis use (n 5 352): OR from logistic regression models
Univariate
Multivariate
Explanatory variable
OR
95% CI
OR
95% CI
Australian birth
Parental divorce
females
males
High level of weekly cannabis use
in school at study inception
Mid-school: cannabis use weekly 1
Mid-school: most peers used cannabis
Mid-school:
Non-smoker
Smoked in the last month
Daily smoker
Mid-school: alcohol . 2 days/week
Mid-school: high dose drinker
Mid-school: antisocial behaviours
Mid-school: psychiatric morbidity
1.6
0.66– 3.7
2.4
0.92– 6.1
2.1
0.63
1.1
0.87– 5.3
0.26– 1.6
0.66– 1.9
2.1
0.47
0.87
0.82– 5.6
0.14– 1.6
0.49– 1.6
4.8
2.5
2.0– 12
1.4– 4.4
3.4
2.1
1.3– 9.0
1.1– 4.0
1
1.7
2.9
1.1
1.3
2.0
1.2
0.86– 3.2
1.5– 5.7
0.55– 2.3
0.75– 2.2
1.1– 3.7
0.68– 2.1
1
2.0
3.3
0.61
0.62
1.5
1.0
1.0– 4.2
1.6– 7.2
0.27– 1.4
0.33– 1.2
0.74– 3.0
0.55– 2.0
female OR 2.2, 1.1– 4.3). All main effects and
interactions between gender and the explanatory
variables were examined using backwards stepwise regression. As all incident cases of daily
cannabis use were participants born in Australia,
this variable was not included in the analysis.
There was evidence of important interactions
between gender and three mid-school predictors
(Table 5). Males who reported that most of their
peers used cannabis were at six-fold increased
risk, in contrast to females for whom this effect
was negligible. Conversely, females, unlike
males, were at around six-fold elevated risk if
they reported earlier high dose drinking or antisocial behaviours. There was a trend for schoollevel exposure to cannabis use to predict incident
daily cannabis use in late-school, independent of
gender. The residual effect for gender was not
signicantly predictive of daily use at p 5 0.05
(OR 3.8, 0.82– 18). Parental divorce or separation (univariate OR: 3.2, 1.6– 6.3), mid-school
daily smoking (univariate OR: 5.5, 2.4– 13), midschool frequent alcohol use (univariate OR: 3.8,
1.6– 8.9) and mid-school psychiatric morbidity
(univariate OR: 2.0, 1.0– 3.9) were removed
from the model during the selection process as
they were not predictive of initiation into daily
cannabis use in the multivariate model.
Discussion
One in ve Australian adolescents used cannabis
during the mid-teens. For the great majority the
frequency of cannabis use remained at low levels
with around two-thirds of all users in both midand late-school periods reporting less than
weekly use. By examining progression to daily
use we were able to delineate a group who were
at unequivocal risk of harmful use. The mid- to
late teens was an important period for progression in use with 13% of male and 9% of
female mid-school users going on to daily cannabis use.
This study differs from earlier work in that it is
based on the repeated measurement of cannabis
use at multiple points. It is therefore able to
address questions of both initiation of use and
progression to higher levels of use. As school
retention rates were 98% in this state in the year
of initial sampling, the sample frame provided an
almost representative adolescent study population (Australian Bureau of Statistics, 1993).
The age range is around the previously reported
peak age for initiation of cannabis use (Chen &
Kandel, 1995). One issue of importance is that
of the validity of self-report of cannabis use.
Self-report of cannabis use has been demonstrated to have good construct validity, to have
reasonable stability and to be no worse in this
regard than other self-report
measures
(O’Malley, Bachman & Johnstone, 1983). Stability has been shown to be related to the recall
period so we can expect that the daily and
weekly response categories were reasonably
Natural history of adolescent cannabis use
1687
Table 5. Prediction of initiation into late-school daily cannabis use (n 5 37) by adolescents who reported
none or less than daily mid-school cannabis use (n 5 1679): OR from logistic regression models
Univariate
Explanatory variable
High level of weekly cannabis use in
school at study inception
Mid-school: cannabis use (weekly or
less often)
Mid-school: most peers used cannabis
Females
Males
Mid-school: high dose drinker
Females
Males
Mid-school: antisocial behaviours
Females
Males
reliable. Although the occasional category used a
6-month reference period, enhanced ability to
remember unusual events could have countered
a tendency to under-report (O’Malley et al.,
1983). Another source of bias could have been
the lower participation rates noted to be associated with weekly cannabis use at study entry.
There was possibly the potential for misspecication of cannabis use in individuals
absent from waves within each study period. We
have assumed that patterns of associations
observed in the data were similar for individuals
for whom data was missing. This could have
resulted in slightly biased OR estimates.
Different mechanisms have been suggested to
explain the uptake of illicit drugs in young people. The stage theory implies that use of one
drug further down a sequence, for example
alcohol and/or nicotine, in some way facilitates
the use of drugs at higher levels, for example
cannabis (Adler & Kandel, 1981; Yamaguchi &
Kandel, 1984; Welte & Barnes, 1985; Fleming et
al., 1989; Graham et al., 1991; Ellickson, Hays
& Bell, 1992; Kandel, Yamaguchi & Chen,
1992). Evidence from these studies is also consistent with the hypothesis that drug use is determined by a single underlying dimension of
vulnerability to drug use or “transition proneness” (Jessor & Jessor, 1977) and that the use of
different drugs at different times is an opportunistic response to changing environmental
conditions such as availability. The concept of
vulnerability has been extended further to sug-
Multivariate
OR
95% CI
OR
95% CI
3.6
1.9– 7.1
2.0
0.97– 4.3
29
11– 74
8.7
2.8– 26.8
11
23
3.5– 32
9.3– 58
1.3
6.5
0.35– 4.6
2.3– 18.3
29
4.0
7.8– 107
1.7– 9.0
6.1
1.0
1.4– 25.4
0.41– 2.6
22
3.9
6.9– 69
1.7– 9.0
6.6
0.91
1.9– 23.3
0.36– 2.4
gest that drug use was one of a constellation of
deviant behaviours described collectively as a
syndrome of problem behaviours (Donovan &
Jessor, 1985). The veracity of these theories can
be informed by examining risk processes
involved in the natural history of cannabis use.
In this study, prior use of cannabis was found
to be strongly and independently predictive of
subsequent use. Overall, four-fths of those who
reported earlier cannabis use continued at some
level. Only ve of the 44 using cannabis daily in
the later period had not reported earlier use, with
strong evidence that more frequent early use
substantially increased the propensity to later,
possibly harmful, daily use. Specically, both
weekly and daily use carried around a six-fold
elevated risk of later daily use relative to
occasional use. However, it must be remembered
that escalation was far from being an inevitable
consequence of early occasional use in that only
4% of mid-school occasional users made this
transition.
Quitting and persistence in cannabis use in
adolescence has not been studied previously in
non-clinical settings. Eighty-two per cent of
those reporting cannabis in the mid-school
period continued use in the late school period.
Continued use was more common among males,
young people reporting more regular cannabis
use, smokers and those with cannabis using
friends.
The co-occurrence of tobacco use and cannabis use is well documented (Hall, 1995). We
1688
C. Coffey et al.
found that although both alcohol use and smoking were associated with cannabis uptake, only
smoking was independently predictive of persistent use by early users. This nding indicates
that it is the co-occurrence of smoking rather
than alcohol use that distinguishes between transient experimentation and entrenched behaviour, with the degree of entrenchment
apparently related to smoking frequency. It is
interesting to speculate whether the mechanism
is purely social, reecting the companionable
experience in common with smoking cigarettes
and smoking cannabis, or whether there may in
part be an underlying physiological or psychological vulnerability to both nicotine and cannabis dependency in these young people. This
vulnerability may simply be that initiation of
cannabis is unlikely in the absence of some prior
history of smoking as a method of drug ingestion. That peer use was also an independent
predictor of persistent use tends to support the
possibility of a social determinant component.
The lack of independent association with other
norm-violating behaviours or with symptoms of
depression and anxiety would seem to discount
problem behaviour or psychological vulnerability
as the mechanism.
A number of previous studies have reported
that tendencies in childhood to disruptive or
norm-violating behaviours are important predictors of the development of cannabis use (Shedler
& Block, 1990; Lynskey & Fergusson, 1995). In
an extension of these ndings and in contrast to
persisting use, we found that antisocial behaviour in the mid-school period was predictive of
cannabis uptake. As is already well-documented,
we found that reported peer cannabis use held
clear and robust associations with cannabis use
and was strongly predictive of uptake. Further,
to the best of our knowledge, this is the rst
study to specically examine the inuence of the
level of cannabis use within the individual’s
school environment measured at the school
level. Elevated risk of cannabis initiation associated with environmental cannabis use is consistent with earlier reports that family, peer and
community levels of drug use are important
determinants of substance use behaviours
(Hawkins, Catalano & Miller, 1992).
The analysis method we used to examine risk
factors for cannabis initiation allowed us to infer
that the inuence of each risk factor was similar
for incident occasional use and incident regular
use. This nding must be interpreted cautiously
as the test of “proportional odds” had low
power, but it may indicate that identied factors
endowed a general blanket of risk, irrespective of
the level of uptake.
Initiation of daily cannabis use in the lateschool period differed between males and
females. Males were more than twice as likely to
make the transition to daily use but earlier
norm-violating behaviour, indicated by antisocial behaviour and high-dose drinking, was
found to predict of daily use only in females.
This observation lends credence to the existence
of a syndrome of problem behaviours described
by Donovan & Jessor (1985), but only for young
women. Males, on the other hand, appeared to
be responding more to social expectations and
opportunities indicated by their greater responsiveness to peer inuences. This nding has
important implications for the prevention of
harmful substance use and suggests that different strategies may be needed to address risks of
heavy cannabis use in young males and females.
Prevention of early cannabis use is likely to
affect rates of daily cannabis use in both sexes.
For boys preventive and early treatment interventions might sensibly address the peer social
context. In contrast, girls who become daily
users appear to lead more chaotic lives and it is
likely that intervention responses would sensibly
extend beyond a focus on cannabis alone.
Acknowledgements
The authors acknowledge the support of the
Victorian Health Promotion Foundation. We
are particularly indebted to Associate Professor
John Carlin from the Clinical Epidemiology and
Biostatistics Unit, Royal Children’s Hospital for
reviewing the paper.
References
ADLAF, E. M. & SMART, R. G. (1991) Drug use among
adolescent students in Canada and Ontario: the
past, present and future, Journal of Drug Issues, 21,
59– 72.
ADLER, I. & KANDEL, D. B. (1981) Cross-cultural perspectives on developmental stages in adolescent drug
use, Journal of Studies on Alcohol, 42, 701– 715.
AUSTRALIAN BUREAU OF STATISTICS (1993) Australia’s
Young People (Canberra, Australian Government
Publishing Service).
Natural history of adolescent cannabis use
CHEN, K. & KANDEL, D. B. (1995) The natural
history of drug use from adolescence to the midthirties in a general population sample, American
Journal of Public Health, 85, 41– 47.
CICCHETTI, D. & ROGOSCH , F. A. (1999) Psychopathology as risk for adolescent substance use disorders: a developomental psychopathology perspective,
Journal of Clinical Child Psychology, 28, 355– 365.
DEMBO, R., FARROW, D., SCHMEIDLER, J. & BURGOS,
W. (1979) Testing a causal model of environmental
inuences on early drug involvement of inner city
junior high school youths, American Journal of Drug
and Alcohol Abuse, 6, 313– 336.
DONOVAN, J. E. & JESSOR, R. (1985) Structure of problem behaviour in adolescence and young adulthood,
Journal of Consulting and Clinical Psychology, 53, 890–
904.
ELLICKSON, P. L., HAYS, R. D. & BELL, R. M. ( 1992)
Stepping through the drug use sequence: longitudinal scalogram analysis of initiation and regular use,
Journal of Abnormal Psychology, 101, 441– 451.
FARRINGTON, D. P. (1991) Longitudinal research
strategies: advantages, problems, and prospects,
Journal of American Academy of Child and Adolescent
Psychiatry, 30, 369– 374.
FERGUSSON, D. M., LYNSKEY, M. T. & HORWOOD , L. J.
(1993) Patterns of cannabis use among 13– 14 year
old New Zealanders, New Zealand Medical Journal,
106, 247– 250.
FERGUSSON, D. M., LYNSKEY, M. T. & HORWOOD , L. J.
( 1996) The short-term consequences of early onset
cannabis use, Journal of Abnormal Child Psychology,
24, 499– 512.
FLEMING, R., LEVENTHAL, H., GLYNN, K. & ERSHLER,
J. (1989) The role of cigarettes in the initiation and
progression of early substance use, Addictive Behaviours, 14, 261– 272.
GRAHAM, J. W., COLLINS, L. M., WUGALTER, S. E.,
CHUNG, N. K. & HANSEN, W. B. (1991) Modeling
transitions in latent stage-sequential processes: a
substance use prevention example, Journal of Consulting and Clinical Psychology, 59, 48– 57.
HALL, W. (1995) The public health signicance of
cannabis use in Australia, Australian Journal of Public
Health, 19, 235– 242.
HALL, W. (1997) The recent Australian debate about
the prohibition on cannabis use, Addiction, 92,
1109– 1115.
HALL, W., JOHNSTON, L. & DONNELLY, N. (1999) The
epidemiology of cannabis use and its consequences,
in: KALANT, H. et al. (Eds) The Health Effects of
Cannabis, pp. 71– 125 (Toronto, Addiction Research
Foundation).
HAWKINS, J. D., CATALANO, R. F. & MILLER, J. Y.
(1992) Risk and protective factors for alcohol and
other drug problems in adolescence and early adulthood: implications for substance abuse prevention,
Psychological Bulletin, 112, 64– 105.
JESSOR, R. & JESSOR, S. L. (1977) Problem Behaviour
and Psychosocial Adjustment: a longitudinal study of
youth (San Diego, CA, Academic Press).
JOHNSTON, L. D., O’MALLEY, P. M. & BACHMAN, J. G.
(1998) National Survey Results on Drug Use from the
1689
Monitoring the Future Study, 1975– 1997: vol. 1. Secondary school students (Washington DC, US Department of Health and Human Services).
KANDEL, D. B. (1980) Drug and drinking behaviour
among youth, Annual Review Sociology, 6, 235– 285.
KANDEL, D. B. & ANDREWS, K. (1987) Processes of
adolescent socialisation by parents and peers, International Journal of the Addictions, 22, 319– 342.
KANDEL, D. B., DAVIES, M., KARUS, D. & YAMAGUCHI,
K. (1986) The consequences in young adulthood of
adolescent drug involvement, Archives of General
Psychiatry, 43, 746– 754.
KANDEL, D. B., YAMAGUCHI, K. & CHEN, K. (1992)
Stages of progression in drug involvement from adolescence to adulthood: further evidence for the gateway theory, Journal of Studies on Alcohol, 53,
447– 457.
LEWIS, G., PELOSI, A. J., ARAYA, R. & DUNN, G. (1992)
Measuring psychiatric disorder in the community: a
standardized assessment for use by lay interviewers,
Psychological Medicine, 22, 465– 486.
LEWIS, G. & WILLIAMS, P. (1989) Clinical judgement
and the standardized interview in psychiatry, Psychological Medicine, 19, 971– 979.
LYNSKEY, M. & HALL, W. (1999) The prevalence of
illicit drug use among youth: results from the Australian School Students’ Alochol and Drugs Survey,
Australian and New Zealand Journal of Public Health,
23, 519– 524.
LYNSKEY, M. T. & FERGUSSON, D. M. (1995) Childhood conduct problems and attention decit behaviours and adolescent alcohol, tobacco and illicit
drug use, Journal of Abnormal Child Psychology, 23,
281– 302.
MADDAHIAN, E., NEWCOMB, M. D. & BENTLER, P. M.
(1988) Adolescent drug use and intention to use
drugs: concurrent and longitudinal analyses of four
ethnic groups, Addictive Behaviours, 13, 191– 195.
MCCULLAGH, P. (1980) Regression models for ordinal
data (with discussion), Journal of the Royal Statistical
Society B, 42, 109– 142.
MOFFITT , T. E. & SILVA, P. A. (1988) Self-reported
delinquency: results from an instrument for New
Zealand, Australian and New Zealand Journal of
Criminology, 21, 227– 240.
NEWCOMB, M. D. & BENTLER, P. M. (1988) Consequences of Adolescent Drug Use: impact on the lives of
young adults (Beverly Hills, Sage).
O’MALLEY, P. M., BACHMAN, J. G. & JOHNSTONE, L. D.
(1983) Reliability and consistency of self-reports of
drug use, International Journal of the Addictions, 18,
805– 824.
PAPERNY, D. M., AONO, J. Y., LEHMAN, R. M., HAMMAR, S. L. & RISSER, J. (1990) Computer-assisted
detection and intervention in adolescent high-risk
health behaviours, Journal of Paediatrics, 116, 456–
462.
PETERSON, B. & HARRELL, F. E. (1990) Partial proportional odds models for ordinal response variables,
Applied Statistics, 39, 205– 217.
ROBINS , L. N. (1995) The natural history of substance
use as a guide to setting drug policy, American
Journal of Public Health, 85, 12– 13.
SHEDLER, J. & BLOCK , J. (1990) Adolescent drug use
1690
C. Coffey et al.
and psychological health: a longitudinal inquiry,
American Psychologist, 45, 612– 630.
STATACORP (1999) STATA (release 6.0) (Texas,
USA, Stata Corporation).
WELTE, J. W. & BARNES, G. M. (1985) Alcohol: the
gateway to other drug use among secondary-school
students, Journal of Youth and Adolescence, 14, 487–
498.
YAMAGUCHI, K. & KANDEL, D. B. (1984) Patterns of
drug use from adolescence to young adulthood: II.
Sequences of progression, American Journal of Public
Health, 74, 668– 672.