A twin study of early cannabis use and subsequent
use and abuse/dependence of other illicit drugs
ARPANA AGRAWAL*, MICHAEL C. NEALE, CAROL A. PRESCOTT
AND KENNETH S. KENDLER
Virginia Institute of Psychiatric and Behavioral Genetics, Medical College of Virginia at Virginia
Commonwealth University, Department of Human Genetics and Department of Psychiatry, Richmond,
VA 23298, USA
Introduction. Cannabis use is strongly associated with the use and abuse/dependence of other illicit
drugs. Gateway and common liabilities models have been employed to explain this relationship. We
sought to examine this association using a combination of the discordant twin design and modeling
Method. We assess the relationship between early cannabis use and the subsequent use and abuse/
dependence of other illicit drugs in a population-based sample of male and female twin pairs using
four analyses: (i) analysis of the association between early cannabis use and other illicit drug use
and abuse/dependence in the entire sample of twins, (ii) assessment of the influence of early can-
nabis use in twin 1 on twin 2’s use or abuse/dependence of other illicit drugs, (iii) use of twin pairs
discordant for early cannabis use in a discordant twin design and (iv) a model-fitting procedure.
Results. We found: (i) a strong association between early cannabis use and use and abuse/depen-
dence of other illicit drugs in the sample, (ii) twin 1’s early cannabis use is significantly associated
with the twin 2’s other illicit drug use, (iii) the role of correlated genetic factors with some evidence
for a causal influence, and (iv) the correlated liabilities model fits the data well.
Conclusions. Early cannabis use is strongly associated with other illicit drug use and abuse/depen-
dence. The relationship arises largely due to correlated genetic and environmental influences with
persisting evidence for some causal influences.
Epidemiological studies report a substantial
relationship between the use of cannabis and
other illicit drugs (Kandel et al. 1992; Golub &
Johnson, 1994; Fergusson & Horwood, 2000;
Morral et al. 2002a; Lynskey et al. 2003).
Cannabis has been proposed as a gateway drug
for subsequent use and abuse/dependence of
other illicit drugs such that cannabis use has
a significant causal influence on the subsequent
use of other illicit drugs. In contrast, the
correlated liabilities model proposes that can-
nabis use and other illicit drug use are associated
because both drugs are influenced by a single
The relative validity of the gateway and
common liability models has been the topic of
intense debate since the early 1970s. Epidemi-
liability model (Huba et al. 1981; Donovan &
Jessor, 1985; Hays et al. 1987; Ellickson et al.
1992; MacCoun, 1998; Morral et al. 2002a,b)
and the gateway model (Adler & Kandel, 1981;
O’Donnell & Clayton, 1982; Yamaguchi &
et al. 1992; Kandel & Yamaguchi, 1993). While
* Address for correspondence: Dr Arpana Agrawal, Virginia
Commonwealth University, Virginia Institute for Psychiatric and
Behavioral Genetics, Department of Human Genetics, Box no.
980003 Suite 1-154, Richmond, VA 23298-0003, USA.
Psychological Medicine, 2004, 34, 1227–1237.
f 2004 Cambridge University Press
Printed in the United Kingdom
these studies show a high degree of association
between cannabis and other illicit drug use, they
do not definitively address the nature of the
association (MacCoun, 1998). Hence, it is still
unknown whether cannabis use has a direct
causal influence on illicit drug use or whether
the two are related by a common liability, or
if the association results from a combination
of correlated and causal processes.
The gateway model and the correlated liab-
ilities models are not antithetic to each other.
Information from twin pairs can be helpful in
examining these models. In particular, the dis-
cordant twin design can compare the risk of
subsequent other illicit drug use in twin pairs
discordant for cannabis use. A recent study by
Lynskey and colleagues analyzed this relation-
ship and found that, compared to their non-user
co-twins, twins with early cannabis use had a
2.6–5.2 times higher risk of using other illicit
drugs (Lynskey et al. 2003).
Yet another unexplored option for examining
the relationship between drug categories is
model-fitting. This method, though limited by
certain assumptions, employs all the available
data in an efficient manner. This combination
of methods, each with strengths and limitations,
allows for a more complete examination of the
nature of the relationship between early canna-
bis use and the subsequent use of other illicit
The present manuscript has the following goals:
(1) To test the association between early
cannabis use and the subsequent use and abuse/
dependence of other illicit drugs, within an indi-
vidual, in a population-based sample of twins.
(2) To assess the extent to which early can-
nabis use in twin 1 predicts the subsequent use
or abuse/dependence of other illicit drugs in
twin 2 after controlling for twin 2’s early can-
(3) To test the association between early
cannabis use and use and abuse/dependence of
other illicit drugs in a discordant twin design,
examining the MZ (monozygotic) and DZ (di-
zygotic) twin pairs separately.
(4) To test two models: a common liab-
ilities model and a random multiformity model,
which captures some features of the gateway
Data for these analyses come from male and
female same-sex twin pairs from the Virginia
Twin Registry. Twins were eligible for partici-
pation if they were born between 1940 and 1974
and were Caucasian. Respondents who agreed
to participate were interviewed in the first wave
of personal interviews. This included 72.8% of
the eligible males and 91.9% of the eligible
females. Zygosity was initially determined by
standard questions and zygosity was later con-
firmed for a subsample of twins by molecular
analyses. Three follow-up telephone interviews
were completed in the females and one follow-
up interview was conducted for the male twins.
Data for the present analysis come from the
second wave of interviews in the males and the
fourth wave in the females and consists of 1196
male same-sex pairs and 934 female same-sex
twin pairs, including MZ and DZ twin pairs. As
approved by the institutional review board, all
participants were informed about the objectives
of the study prior to participation and informed
consent was obtained prior to participation.
Additional details regarding data collection and
ascertainment are available elsewhere (Kendler
et al. 1992, 2002).
Other illicit drug use was coded as a binary
variable for lifetime use of cocaine, sedative,
stimulants, opiates and hallucinogens. The
combined category for the use of ‘other illicit
drug’ includes use of cocaine, sedatives, stimu-
lants, opiates or hallucinogens (but not includ-
ing cannabis). The early cannabis use variable
who reported use of cannabis before or at the
age of 18 years and a ‘0’ if the respondent was
a non-user or used cannabis after the age of
18 years. For the categories of drugs that may
be obtained legally (sedatives, stimulants and
opiates) use was defined by any of the following
criteria: (a) use without a prescription, (b) in
greater amounts than prescribed, (c) more fre-
quently than prescribed or (d) for reasons other
than those for which it was prescribed. Abuse
and dependence was assessed using DSM-IV
criteria (APA, 1994). If the participant was di-
agnosed with abuse or dependence for cocaine,
1228A. Agrawal et al.
sedatives, stimulants, hallucinogens or opiates,
they were diagnosed with abuse/dependence for
‘any illicit drug’. Individuals who reported use
of other illicit drugs before the age of 18 years
were not included in the analysis.
In an attempt to replicate the methods used
by Lynskey and colleagues, we used six covari-
ates to adjust our odds ratios (ORs) as well as
sex and zygosity. Regular alcohol use referred to
drinking for at least once a month for 6 months
or more, before or at the age of 18 years. Simi-
larly, regular nicotine use reflects smokers who
smoked regularly for at least a month, before or
at the age of 18 years. The psychiatric disorders
Disorder (GAD) and Conduct Disorder (CD)
were diagnosed using an adaptation of SCID
(Spitzer et al. 1987) and the DSM-III-R criteria
(APA, 1994). MD was based on a lifetime di-
agnosis, GAD utilized a lifetime report of a
1-month rather than a minimum 6-month dur-
ation of illness and CD was created based on
the presence of three or more CD behaviors
prior to the age of 18 years. Years of education
corresponds to an interview item regarding
the highest grade of school or college year
We compared the Akaike Information Cri-
terion (AIC) for models that imposed a linear
or quadratic or logarithmic relationship, or
spline functions that had knots at 12–25 years
(SAS Institute, 1999) to determine early onset
of cannabis use. The linear function provided
the best fit to the data. The choice of 18 years
to distinguish early and late onset is purely
based on the median age of onset of cannabis
PROC GENMOD in SAS was used to examine the
association between early cannabis use and
the use or abuse/dependence of other illicit
drugs (SAS Institute, 1999). Data from all twin
pairs was pooled and independent estimating
equations (IEE) were employed to account for
the clustering of twin data (SAS Institute, 1999).
In an attempt to replicate the methods used by
Lynskey et al. the ORs were adjusted for sex,
years of education, CD, MD, GAD and regular
alcohol and nicotine use.
We also used PROC GENMOD to assess the
relationship between early cannabis use in one
twin and subsequent other illicit drug use or
abuse/dependence in the co-twin after control-
ling for the co-twin’s early cannabis use. ORs
were calculated separately for MZ and DZ twin
pairs and the statistical difference between the
MZ and DZ ORs was tested by interacting early
cannabis use with zygosity. A similar interaction
was employed to test for sex differences.
Our next step was to employ the discordant
twin design to study the association between
early cannabis use and other illicit drug use or
abuse/dependence. Following the approach of
Lynskey and colleagues, we selected twin pairs
discordant for early cannabis use (Lynskey et al.
2003). The association between early cannabis
use and use or abuse/dependence of other illicit
drugs may be due to the following relationships:
(i) early cannabis use has a causal influence
on subsequent other illicit drug use or abuse/
dependence or (ii) early cannabis use and other
illicit drug use and abuse/dependence are related
due to correlated genetic influences or (iii) a
combination of causal and correlative influences
are responsible for the relationship. To dis-
tinguish between these models, ORs from MZ
and DZ twins were compared to each other and
to the population ORs. If the association were
purely causal, then early cannabis use would
be strongly associated as a risk factor for sub-
sequent use or abuse/dependence of other illicit
drugs in the population and in the discordant
MZ and DZ twin pairs. In other words, the
extent of genetic overlap (100% in MZ and
50% in DZ twin pairs) would not influence the
ORs. This definition of causality is fairly narrow
and refers to the direct influence of early can-
nabis use on the use of other illicit drugs. On the
other hand, if genetic factors common to early
cannabis use and use or abuse/dependence of
other illicit drugs were exclusively responsible
for the association between the two drug cat-
egories, we would expect a different pattern of
ORs. Early cannabis use would still be strongly
associated with future other illicit drug use or
abuse/dependence in the entire population. MZ
twins who share all their genes and are discor-
dant for early cannabis use would have ORs
equal to 1.0. DZ twins, on the other hand only
share 50% of their genes and would have ORs
greater than 1.0 but less than the population
OR. Fig. 1 presents a graphical description of
a comparison of population, MZ and DZ ORs
Early cannabis use and other illicit drugs1229
expected under a purely causal or a purely
correlated liabilities model.
The discordant twin analyses were performed
using conditional logistic regression in PROC
PHREG in SAS. The ORs obtained from the con-
ditional logistic regression model reflect the risk
of other illicit drug use or abuse/dependence in
an early cannabis using twin versus a non-user
co-twin. The information in the analysis arises
from the within-twin pair differences. In an at-
tempt to replicate results presented by Lynskey
and colleagues we first fit a model that pooled
MZ and DZ twins. Covariates used for the
population ORs were used for adjustment in
the discordant twin design (Lynskey et al. 2003).
The same analyses were performed by separat-
ing the MZ and DZ twins and controlling for
sex. We assessed the statistical significance of
different ORs in MZ and DZ discordant twin
pairs by comparing a model that allowed for
separate parameter estimates in MZ and DZ
estimate for both zygosities. Additionally, we
tested for the effect of sex on this association.
We also used two models from the Neale–
Kendler co-morbidity models to examine the
association between early cannabis use and the
subsequent use or abuse/dependence of other
illicit drugs. To reduce the possible number
of combinations of models and to maximize
power, we only used the ‘other illicit drug use’
category (any illicit drug, except cannabis) for
model-fitting. We tested three putative models
to explain the relationship between early can-
nabis use and the use of ‘other illicit drugs’. The
first model fitted to the data was the correlated
liability model where early cannabis use and the
use of ‘other illicit drugs’ each have liabilities
influenced by correlated genetic (A), shared
environmental (C) and unique environmental
(E) influences. We fitted a sub-model to the
correlated liabilities model that only allowed
for the unique environmental influences to be
correlated (Re only). The correlated E model
allows for twins to have correlated unique en-
vironmental influences alone and independent
genetic and shared environmental
Additionally, we fitted a sub-model where only
genetic (Ra) and shared environmental (Rc)
cross-drug correlations were allowed and the
correlation between unique environmental fac-
tors (Re) was constrained to zero.
The final model was the random multiformity
of cannabis model, which captures some aspects
of the gateway hypothesis. In this model, early
cannabis use and use of ‘other illicit drugs’ is
described by unique and independent liability
distributions that are influenced by non-over-
lapping A, C and E. However, individuals above
the threshold on the distribution for early
cannabis use are users of cannabis and also at
increased risk for using ‘other illicit drug’. This
increased risk is independent of the individual’s
liability to use ‘other illicit drugs’ itself. Thus,
while some proportion of individuals use ‘other
illicit drugs’ because they are above the thres-
hold for ‘other illicit drug’ use, another pro-
portion of individuals use ‘other illicit drugs’
only due to their prior early cannabis use. A
significant limitation of this model is that it
does not allow for correlations between A, C
Population Disc. DZDisc. MZ
Population and discordant twins
PopulationDisc. DZ Disc. MZ
Population and discordant twins
subsequent use of other illicit drugs in the discordant twin design. (a)
The pattern of odds ratios (ORs) seen if the relationship between
early cannabis use and subsequent other illicit drug use is explained
by causal influences alone. A fairly narrow definition of causality
implied by this model. In this case, the MZ (%) and DZ ( ) discor-
dant twin ORs (Disc. MZ and Disc. DZ respectively) are equal to
each other and equal to the population (&) OR. Therefore, the risk
of other illicit drug use is elevated, irrespective of genetic relation-
ship. (b) The pattern of ORs that would be obtained if correlated
genetic influences were solely responsible for the association between
early cannabis use and the use of other illicit drugs. MZ twins who
share all their genes have an OR (%) equal to unity while DZ twins
who only share half their genes have an OR greater than unity ( )
but significantly lower than the population OR (&) where indi-
viduals are unrelated.
Two types of associations between early cannabis use and
1230A. Agrawal et al.
and E influencing each drug category. Although
that combines aspects of the correlated liabilities
model (with Ra, Rc and Re) and models causal
paths, this model is not identified and therefore,
cannot be tested.
Further descriptions and mathematical in-
terpretations of the correlated liabilities and
random multiformity model are available else-
where (Neale & Kendler, 1995).
All structural equation models were fitted
using AIC, which is an index of fit and parsi-
mony and is calculated by subtracting two times
the degrees of freedom from the x2fit. The
model with the lowest AIC (largest negative)
was the best-fitting model (Akaike, 1987;
This study utilized data from 2402 male and
1942 female same-sex twins. The mean age at
the time of the interview was 35.5 years (range
20–58 years) for the males and 35.8 years (range
21–62 years) for the females. The mean of years
of education reported by the twins was 13.6 and
14.3 years in the males and females respectively.
Of these twins, 46.6% females and 52.9% males
reported lifetime cannabis use. Of the cannabis
users, 61.2% females and 64.5% males reported
the early use of cannabis (i.e. 18 years or before).
Forty-two per cent of the females and 62% of
the males reported having used other illicit
Table 1 shows the ORs obtained when exam-
ining the association between early cannabis
use and the use or abuse/dependence of other
illicit drugs, within an individual. The ORs were
calculated using the twin data as a population
sample and unadjusted as well as covariate-
adjusted ORs are presented. The association
between early cannabis use and use and abuse/
dependence of other illicit drugs, in the popu-
lation, was robust. Early cannabis use was
associated with other illicit drug use with an OR
of 6.9 [95% confidence interval (CI) 5.8–8.1]
and with other illicit drug abuse/dependence
with an OR of 4.8 (95% CI 3.7–6.2).
The ORs presented in Table 2 refer to the
extent to which twin 1’s early cannabis use
influenced subsequent other illicit drug use or
abuse/dependence in twin 2, after controlling
for early cannabis use in twin 2. The interaction
of twin 2’s other illicit drugs use with sex was
(ORs) and 95% confidence interval (CI) for risk
predicted by cannabis use before 18 years of age
Within-person population odds ratios
Drug Unadjusted (95% CI) Adjusted (95% CI)
interval (CI) where twin 1’s cannabis use before
18 years predicts twin 2’s illicit drug use and
abuse/dependence after controlling for twin 2’s
own early cannabis use. Results based on logistic
regression controlling for sex
Odds ratios (ORs) and 95% confidence
significance of MZ
and DZ ORs*
5.1 (2.2–11.5) 1.2 (0.6–2.6)
* Analyses adjusted for sex. Values presented are two-tailed
Early cannabis use and other illicit drugs 1231
not significant (p>0.1), which allowed us to
pool male and female twins for further analyses.
The ORs for the association between early
cannabis use and use and abuse/dependence of
other illicit drugs were strong and significant.
tives and opiates, the DZ ORs were consist-
ently lower than the MZ ORs. Overall, the ORs
in MZ twins were 1.5–4.0 times greater than
the ORs in DZ twins and these differences were
Males and females were pooled for the dis-
cordant twin analyses. However, MZ and DZ
twin pairs were analyzed separately. Even after
adjusting for potential covariates, there was a
significant increase in risk of other illicit drug
use and abuse/dependence in early cannabis
users compared to their non-user co-twins. Re-
sults for the pooled sample are available upon
request. Table 3 presents ORs for the discordant
twin design, separately for MZ and DZ twin
pairs. The difference between the MZ and DZ
ORs was significant for other illicit drug use
and other illicit drug abuse/dependence but not
for the individual drugs.
Fig. 2 depicts our results for the ORs ob-
served in the discordant twin analyses. The
pattern was similar to the correlated genetic
model in Fig. 1. The population OR was sig-
nificantly higher than the discordant twin ORs.
Within the discordant twin design, the DZ
OR for early cannabis use predicting other
illicit drug use or abuse/dependence was signifi-
cantly greater than the MZ OR but considerably
lower than the population OR. However,
our observed pattern of ORs differs from the
pattern predicted by a model where correlated
genetic factors alone explain the relationship
between the two drug categories in one import-
ant way: the MZ OR was significantly greater
The model-fitting procedure provided some
interesting insights as well (Table 4). The best-
fitting correlated liabilities model in males and
females allowed for genetic, shared and unique
environmental correlations across early canna-
bis use and use of ‘other illicit drugs’. The cor-
relation between the genetic factors (Ra) was
substantial (0.91 in males and 1.0 in females)
while the shared environmental factors (Rc)
were correlated between 0.68–1.00. The unique
environmental factors (Re) showed low corre-
lations with greater evidence for drug-specific
environmental influences. The proportion of the
total covariance due to genetic factors was 46%
in males and 53% in females. Correlated unique
environmental influences contribute to 13%
and 10% of the total covariance in males and
females respectively, with the remaining covari-
ance accounted for by shared environmental
influences. Furthermore, a submodel where Re
twins in the discordant twin design. ORs reflect
risk of subsequent other illicit drug use or abuse/
dependence in an early cannabis user twin versus
the non-user co-twin
Odds ratios (ORs) for MZ and DZ
and DZ OR*
* Values presented are two-tailed p values.
Population and discordant twins
Disc. DZDisc. MZ
our sample of discordant twin pairs from the Virgnia Twin Registry.
The ORs presented in the diagram depict ORs for early cannabis use
and subsequent use of other illicit drugs. Overall, the pattern of ORs
is very similar to the pattern of ORs presented in the correlated
genetic factors model in panel (b) of Fig. 1. The MZ OR (%) is lower
than the DZ OR ( ), which is lower than the population OR (&).
However, unlike the expected pattern in Fig. 1, the MZ OR is greater
than 1.0. This suggests that the relationship between early cannabis
use and other illicit drugs is partly due to correlated genetic factors
and partly due to causal influences.
The pattern of MZ and DZ odds ratios (ORs) observed in
1232A. Agrawal et al.
was set to zero led to a significant increase in
AIC (AICmales=2.24, AICfemales=x1.27). This
increase in AIC indicated that while the unique
environmental influences were only modestly
correlated, their influence on early cannabis use
and use of ‘other illicit drugs’ via a correlated
pathway could not be ignored. The model with
correlations only between unique environmen-
tal factors (correlated E model) fitted poorly.
Finally, the random multiformity of cannabis
model did not fit the data well.
Cannabis is the most commonly used illicit
psychoactive drug in the US population and has
been proposed as a gateway drug (Kandel &
Yamaguchi, 1993; Golub & Johnson, 1994;
Kandel, 2003). A prior twin study provides evi-
dence for the causal influence of early cannabis
use on the use and abuse/dependence of other
illicit drugs (Lynskey et al. 2003). The primary
goal of our study was to elucidate the relation-
ship between early cannabis use and use and
abuse/dependence of other illicit drugs in
another genetically informative sample. We
used four related methods to study the associ-
ation between the two drug categories beginning
with an assessment of the within-person associ-
ation between early cannabis use and use and
abuse/dependence of other illicit drugs in the
entire twin sample. This was followed by a test
for cross-twin prediction of risk of other illicit
drug use and abuse/dependence from the co-
twin’s early cannabis use. Next, we used the
discordant twin design to identify the nature
of the association between early cannabis use
and other illicit drug use and abuse/dependence.
Finally, we examined models that test the im-
pact of correlated liabilities or random multi-
formity on the relationship between the two
We documented a strong and significant
association between early cannabis use and use
and abuse/dependence of other illicit drugs. The
cross-twin analyses, showed that early cannabis
use in one twin was a significant predictor of
other illicit drug use and abuse/dependence in
the co-twin even after controlling for the co-
twin’s own early cannabis use. Furthermore, the
cross-twin prediction was substantially stronger
in MZ twins who share all their genes than in
DZ twin pairs, who share only 50% of their
genes. This suggested that genetic factors that
influence both early cannabisuseandother illicit
drug use and abuse/dependence may be partly
responsible for the association between the two
The discordant twin analyses yielded a very
Table 4.Model-fit and parameter estimates for model-fitting procedure
Early cannabis useOther illicit drug use
covariance % A % C% Ea2
x7.0 0.230.50 0.270.57 0.170.260.911.00 0.340.714641 13
265.6 0.05 0.210.740.230.00 0.77—— 0.87————
77.5 0.160.520.32 0.69 0.00 0.31———————
x7.7 0.44 0.360.200.30 0.39 0.311.000.68 0.26 0.685337 10
210.80.05 0.34 0.61 0.000.240.76—— 0.81————
a2, Additive genetic influence; c2, shared environmental influence; e2, unique environmental influence.
Ra, Genetic correlation between early cannabis use and use of any illicit drug; Rc, shared environmental correlation between early cannabis
use and use of any illicit drug; Re, unique environmental correlation between early cannabis use and use of any illicit drug.
% A, % C and % E refer to the percentage of the total covariance due to additive genetic, shared environmental and unique environmental
The best-fitting model is highlighted in gray.
Early cannabis use and other illicit drugs 1233
DZ twins had ORs significantly greater than
discordant MZ twins but lower than the popu-
lation ORs. The MZ ORs were greater than 1.0
and this suggests a true causal influence of early
cannabis use on the use and abuse/dependence
of other illicit drugs. However, this elevation of
ments. Overall, our results suggest that a signifi-
cant proportion of the relationship between
early cannabis use and subsequent use or abuse/
dependence of other illicit drugs is due to cor-
related genetic and environmental factors but
we cannot rule out the role of direct causal in-
fluences in this relationship.
Issues of causality pose a challenge in the
study of genetically informative data. Analysis
of data from discordant twin pairs cannot
distinguish a truly causal model from a model
where the overlap of unique environmental
influences across the two drug categories leads
to uniformly elevated ORs. For example, im-
agine that a certain environmental factor, such
as childhood head trauma, causes early canna-
bis use and subsequent use of other illicit drugs.
Now, consider an MZ twin pair where one
member suffers the head trauma but the other
member of the twin pair does not. As a conse-
quence of the head trauma, the twin with the
trauma is predisposed to cannabis use before
the age of 18 years as well as other illicit drug
use. The member of the twin pair that did not
experience the trauma uses neither cannabis
nor other illicit drugs. Even though the twins
share all their genetic factors, one member of the
twin pair did not experience the environmental
factor – the head trauma and hence, this MZ
twin pair is discordant for early cannabis use.
However, when the early cannabis user twin
is compared to the non-user co-twin for sub-
sequent other illicit drug use, it would seem like
the cannabis-using twin has a greater risk for
subsequent other illicit drug use than the
non-user twin. However, the reason for this in-
creased risk is because the cannabis-using twin
experienced the head trauma, which influenced
both his early cannabis and other illicit drug use
(or a common liability to illicit drug use) and
not because early cannabis use has a causal
impact on subsequent other illicit drug use. This
correlated environmental influence cannot be
discriminated from a true causal effect within
the discordant twin design.
However, it is also possible that the elevation
in the MZ OR is due to the causal influence
of early cannabis use on the subsequent use or
abuse/dependence of other illicit drugs. Theor-
etically, the only method that could truly assess
the causal impact of early cannabis use on later
other illicit drug use is a prospective study that
measures environmental factors. Such a study
fic environmental influences on each drug and
control for them. However, choice of which en-
vironmental factors to measure, how to measure
them and the possibility that they are elicited
or selected by certain individuals are serious
obstacles to such a study.
Overall, the complementary techniques em-
ployed in this study suggest a robust association
between early cannabis use and other illicit drug
use and abuse/dependence. There was also some
evidence for the causal influence of early can-
nabis use on subsequent use or abuse/depen-
dence of other illicit drugs and this causal effect
was indistinguishable for the influence of corre-
lated environmental factors. Some of earliest
evidence for the gateway suggested that age of
onset of cannabis had a significant causal impact
on later illicit drug use (O’Donnell & Clayton,
1982; Voss & Clayton, 1987; Graham et al.
1991; Kandel et al. 1992; Kandel & Yamaguchi,
1993; Hawkins et al. 1997). Morral and col-
leagues reported that use of cannabis before
15 years of age conferred a 1.60 times greater
risk for other illicit drug use than if initiation
was after 15 years (Morral et al. 2002a). Fur-
thermore, another study reported an elevated
hazard ratio of 140.0 in users of cannabis for
future use of other illicit drugs (Fergusson &
Horwood, 2000). While each of these studies
commented on the increased risk of using other
illicit drugs in early users of cannabis, they did
not formally comment on the causal nature of
We are unaware of any epidemiological
analyses that directly imply that the influence of
cannabis use on subsequent use of other illicit
drugs is causal (Kandel, 2003). One reason for
this is the inability to determine causality inde-
pendent of correlated environments. While ani-
mal models have provided some evidence for the
causal impact of cannabis use, these controlled
experiments are impossible in humans (Tanda
et al. 1997; Cadoni et al. 2001; Lamarque et al.
1234A. Agrawal et al.
2001). Twin studies allow us to control for
genetic background when assessing causal in-
ference. A recent report presented significant
evidence for elevated ORs in co-twins of early
cannabis user twins (OR 2.1–5.2) (Lynskey et al.
2003). The analyses by Lynskey et al. suggested
a 2.1–5.2 times higher odds of other illicit drug
use and abuse/dependence in early cannabis
users than non-users. Our results show similar
ORs but present an alternative explanation for
the elevated ORs in the discordant twin design.
Also, the methods used by Lynskey et al. were
somewhat different from our approach. First,
Lynskey et al. did not assess ORs in MZ and
DZ twin pairs separately. According to their
discordant twin analyses, an interaction with
zygosity was not significant and MZ and DZ
twins could be pooled within the discordant
twin design. We, on the other hand, noted
significantly higher DZ versus MZ ORs in our
discordant twin design. This divergence in the
results may reflect true differences between drug
habits and association between illicit drugs in
the United States and Australian populations.
Secondly, Lynskey and colleagues did not
employ a model-fitting approach nor did they
attempt to distinguish between true causality
and correlated environments. While our model-
fitting methods are limited by certain assump-
tions (e.g. no gene–environment correlations,
assortative mating) andare unable totest certain
key models, they are complementary to the dis-
cordant twin design and shed further light on the
complex process of poly-drug use and misuse.
In our data, the association between early
cannabis use and other illicit drug use or abuse/
dependence was partly due to the correlated
nature of the factors that influence these drugs.
Furthermore, the correlated genetic factors ac-
count for 46% and 53% of the total covariance
in males and females respectively. This common
liability model has been supported by some
Morral et al. 2002a,b). A simulation study by
Morral and colleagues showed that both se-
quencing and association can be explained by
a common liability model where drug use pro-
pensity drives both processes without imposing
a causal pathway from cannabis use to sub-
sequent use of other illicit drugs (Morral et al.
2002a). Genetic studies can partition these
common factors into genes and environment.
Most genetic studies provide evidence for a high
correlation of genetic factors (van den Bree et al.
1998; Tsuang et al. 2001; Kendler et al. 2003).
A study by Tsuang and colleagues compared
a liability-based gateway to a common pathway
model and rejected the gateway model (Tsuang
et al. 1998). Our random multiformity of can-
nabis model, which measures the impact of early
cannabis use per se, was also rejected. A theo-
retically superior model that examines the gate-
way effect would include the strong correlations
between genetic and environmental factors that
we observe, plus causal pathways. While such a
model is conceptually more informative, it is not
identified. Therefore, given the scope of the pre-
sent data and the techniques available to us, we
can only claim that strong genetic and environ-
relationship between cannabis and other illicit
drugs. While there is some evidence for causal
influences, our results suggest that these causal
influences cannot explain all the association be-
tween early cannabis use and the subsequent use
and abuse/dependence of other illicit drugs.
In conclusion, the relationship between early
cannabis use and use and abuse/dependence
of other illicit drugs is strongly influenced by
correlated genetic and environmental factors.
A proportion of these environmental factors,
which may be the effect of deviant peer groups
that allow opportunity for exposure to all kinds
of illicit drugs (Dishion & Owen, 2002; Heim
et al. 2002; Wagner & Anthony, 2002) are
common to early cannabis use and use and
abuse/dependence of other illicit drugs. These
for the relationship between the two drug cat-
egories (Lynskey et al. 2003). On the other hand,
it is possible that early cannabis use has a true
causal impact on other illicit drug use and
abuse/dependence. Future studies need to con-
sider both these sources (correlated and causal
cannabis and other illicit drugs.
The findings in this paper should be viewed with
the following limitations in mind:
(1) These results depend on the validity of
retrospectively reported data. There may be
some recall bias or telescoping when reporting
use of cannabis, age of first use and use and
Early cannabis use and other illicit drugs1235
symptoms of abuse/dependence of other illicit
drugs. While we only have data on the illicit
drugs from one wave of interviews, short-term
test–retest reliability measures (n=172 twin
pairs measured after 4 weeks of primary inter-
view) are available on a subset of the twin pairs.
Drug use was assessed with very high test–retest
reliability (r=0.98 for cannabis and r>0.90
for other illicit drugs). Abuse/dependence was
diagnosed with fair reliability with intra-class
correlations r>0.80. Additionally, the age of
onset of first cannabis use was reported with a
very high reliability (r=0.91). Therefore, it is
unlikely that the findings in this analysis have
been significantly impacted by recall bias.
(2) These analyses pool non-users of canna-
bis with those who initiated cannabis use after
the age of 18 years. We performed the popu-
lation regression and discordant twin analyses
with a different definition of early cannabis use
where only users of cannabis were included in
the analyses. Early cannabis use was defined as
a binary variable where users initiated cannabis
use before or at the age of 18 years and non-
users initiated cannabis use after 18 years (non-
users not included). This analysis found ORs
that were fairly low and we had limited ability
to distinguish discordant MZ and DZ ORs.
However, the pooled ORs for early cannabis use
predicting other illicit drug use was significantly
greater than unity for all drugs except cocaine
and sedatives. The magnitude of the ORs may
reflect a true lack of association but it is more
likely that the poor ability to discriminate the
ORs is due to the extremely reduced sample size
that results from eliminating 50% of the dataset
by ignoring the non-users. Results are available
(3) We used drug use as a binary variable
assessing lifetime use ofcannabisandotherillicit
drugs. While age of onset data was utilized to
code the early cannabis use variable, we did not
employ the frequency of use data that could
potentially distinguish experimental users from
(4) As noted in the text, we employ a fairly
narrow definition of causality for the discordant
twin design. Due to the nature of retrospec-
tive report data, certain assumptions were made
during the twin analyses and therefore, true
measures of causality are not within the scope of
(5) This study utilizes information from
Caucasian twin pairs and these results may not
apply to other ethnicities or to other socio-
This work was supported by NIH grants MH-
40828, MH/AA/DA-49492, AA-09095 and DA-
11287. We acknowledge the contribution of the
Virginia Twin Registry, now part of the Mid-
Atlantic Twin Registry (MATR), to ascertain-
ment of subjects for this study. The MATR,
directed by Dr J. Silberg, L. Corey and L. Eaves,
has received support from the National In-
stitutes of Health, the Carman Trust and the
WM Keck, John Templeton and Robert Wood
DECLARATION OF INTEREST
Adler, I. & Kandel, D. B. (1981). Cross-cultural perspectives on
developmental stages in adolescent drug use. Journal of Studies
of Alcohol 42, 701–715.
Akaike, H. (1987). Factor Analysis and AIC, 52 edn, pp. 317–332.
APA (1994). Diagnostic and Statistical Manual of Mental Disorders
(4th edn) (revised edn). American Psychiatric Association:
Blaze-Temple, D. & Lo, S. K. (1992). Stages of drug use: a com-
munity survey of Perth teenagers. British Journal of Addiction 87,
Cadoni, C., Pisanu, A., Solinas, M., Acquas, E. & Di Chiara, G.
(2001). Behavioural sensitization after repeated exposure to Delta
9-tetrahydrocannabinol and cross-sensitization with morphine.
Psychopharmacology (Berlin) 158, 259–266.
Dishion, T. J. & Owen, L. D. (2002). A longitudinal analysis
of friendships andsubstance
from adolescence to adulthood. Developmental Psychology 38,
Donovan, J. E. & Jessor, R. (1985). Structure of problem behavior 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 in-
itiation and regular use. Journal of Abnormal Psychology 101,
Fergusson, D. M. & Horwood, L. J. (2000). Does cannabis use en-
courage other forms of illicit drug use? Addiction 95, 505–520.
Golub, A. & Johnson, B. D. (1994). The shifting importance of
alcohol and marijuana as gateway substances among serious drug
abusers. Journal of Studies of Alcohol 55, 607–614.
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.
Hawkins, J. D., Graham, J. W., Maguin, E., Abbott, R., Hill, K. G. &
Catalano, R. F. (1997). Exploring the effects of age of alcohol
use initiation and psychosocial risk factors on subsequent alcohol
misuse. Journal of Studies on Alcohol 58, 280–290.
1236A. Agrawal et al.
Hays, R. D., Widaman, K. F., DiMatteo, M. R. & Stacy, A. W. Download full-text
(1987). Structural-equation models of current drug use: are
appropriate models so simple (x)? Journal of Personality and Social
Psychology 52, 134–144.
Heim, C., Newport, D. J., Wagner, D., Wilcox, M. M., Miller, A. H.
& Nemeroff, C. B. (2002). The role of early adverse experience
and adulthood stress in the prediction of neuroendocrine stress
reactivity in women: a multiple regression analysis. Depression and
Anxiety 15, 117–125.
Huba, G. J., Wingard, J. A. & Bentler, P. M. (1981). A comparison
of two latent variable causal models for adolescent drug use.
Journal of Personality and Social Psychology 40, 180–193.
Kandel, D. & Yamaguchi, K. (1993). From beer to crack: develop-
mental patterns of drug involvement. American Journal of Public
Health 83, 851–855.
Kandel, D. B. (2003). Does marijuana use cause the use of other
drugs? Journal of the American Medical Association 289, 482–483.
Kandel, D. B., Yamaguchi, K. & Chen, K. (1992). Stages of pro-
gression in drug involvement from adolescence to adulthood:
further evidence for the gateway theory. Journal of Studies on
Alcohol 53, 447–457.
Kendler, K. S., Jacobson, K. C., Prescott, C. A. & Neale, M. C.
(2003). Specificity of genetic and environmental risk factors for
use and abuse/dependence of cannabis, cocaine, hallucinogens,
sedatives, stimulants, and opiates in male twins. American Journal
of Psychiatry 160, 687–695.
Kendler, K. S., Myers, J. & Prescott, C. A. (2002). The etiology of
phobias: an evaluation of the stress-diathesis model. Archives
of GeneralPsychiatry 59, 242–248.
Kendler, K. S., Neale, M. C., Kessler, R. C., Heath, A. C. & Eaves,
L. J. (1992). A population-based twin study of major depression
in women. The impact of varying definitions of illness. Archives of
General Psychiatry 49, 257–266.
Lamarque, S., Taghzouti, K. & Simon, H. (2001). Chronic
locomotor response to amphetamine and heroin. Implications
for vulnerability to drug addiction. Neuropharmacology 41,
Lynskey, M. T., Heath, A. C., Bucholz, K. K., Slutske, W. S.,
Madden, P. A., Nelson, E. C., Statham, D. J. & Martin, N. G.
(2003). Escalation of drug use in early-onset cannabis users vs
co-twin controls. Journal of the American Medical Association 289,
MacCoun, R. (1998). In what sense (if any) is marijuana a gateway
drug? FAS Drug Policy Analysis Bulletin 4 (html: www.fas.org/
Morral, A. R., McCaffrey, D. F. & Paddock, S. M. (2002a). Re-
assessing the marijuana gateway effect. Addiction 97, 1493–1504.
Morral, A. R., McCaffrey, D. F. & Paddock, S. M. (2002b).
Evidence does not favor marijuana gateway effects over a
common-factor interpretation of drug use initiation: responses
to Anthony, Kenkel & Mathios and Lynskey. Addiction 97,
Neale, M. C. (1990). Statistical Modeling with Mx. Dept. of
Psychiatry, Box no. 980710, Richmond VA 23298, USA.
Neale, M. C. & Kendler, K. S. (1995). Models of comorbidity for
multifactorial disorders. American Journal of Human Genetics 57,
O’Donnell, J. A. & Clayton, R. R. (1982). The stepping-stone
hypothesis – marijuana, heroin, and causality. Chemical Depen-
dence 4, 229–241.
SAS Institute (1999). SAS User Guide, Version 8.2. SAS Institute
Inc., Cary, NC.
Spitzer, R. L., Williams, J. B. & Gibbon, J. (1987). Structured Clinical
Interview for DSM-III-R: Patient Version (SCID-P). New York
State Psychiatric Institute: New York.
Tanda, G., Pontieri, F. E. & Di Chiara, G. (1997). Cannabinoid and
heroin activation of mesolimbic dopamine transmission by a
common mu1 opioid receptor mechanism. Science 276, 2048–2050.
Tsuang, M. T., Bar, J. L., Harley, R. M. & Lyons, M. J. (2001). The
Harvard Twin Study of Substance Abuse: what we have learned.
Harvard Review of Psychiatry 9, 267–279.
Tsuang, M. T., Lyons, M. J., Meyer, J. M., Doyle, T., Eisen, S. A.,
Goldberg, J., True, W., Lin, N., Toomey, R. & Eaves, L. (1998).
Co-occurrence of abuse of different drugs in men: the role of drug-
specific and shared vulnerabilities. Archives of General Psychiatry
van den Bree, M. B., Johnson, E. O., Neale, M. C. & Pickens, R. W.
(1998). Genetic and environmental influences on drug use and
abuse/dependence in male and female twins. Drug and Alcohol
Dependence 52, 231–241.
Voss, H. L. & Clayton, R. R. (1987). Stages in involvement with
drugs. Pediatrician 14, 25–31.
Wagner, F. A. & Anthony, J. C. (2002). Into the world of illegal drug
use: exposure opportunity and other mechanisms linking the use
of alcohol, tobacco, marijuana, and cocaine. American Journal of
Epidemiology 155, 918–925.
Williams, L. J. H. P. J. (1994). Parsimony-based Fit Indices for
Multiple-indicator Models: Do They Work? (1st edn), pp. 161–189.
Yamaguchi, K. & Kandel, D. B. (1984). Patterns of drug use from
adolescence to young adulthood: III. Predictors of progression.
American Journal of Public Health 74, 673–681.
Early cannabis use and other illicit drugs1237