See the original paper here at source

Page 1
319
American Journal of Epidemiology
Copyright © 2002 by the Johns Hopkins Bloomberg School of Public Health
All rights reserved
Vol. 156, No. 4
Printed in U.S.A.
DOI: 10.1093/aje/kwf043
ORIGINAL CONTRIBUTIONS
Cannabis Us Cannabis use and Psychosis: A Longitudinal Population-based Study
J. van Os
1,2
, M. Bak
1
, M. Hanssen
1
, R. V. Bijl
3,4
, R. de Graaf
3
, and H. Verdoux
5
1
Department of Psychiatry and Neuropsychology, European Graduate School of Neuroscience, Maastricht University,
Maastricht, the Netherlands.
2
Division of Psychological Medicine, Institute of Psychiatry, London, United Kingdom.
3
The Netherlands Institute of Mental Health and Addiction, Trimbos-Instituut, Utrecht, the Netherlands.
4
Research and Documentation Centre (WODC), Ministry of Justice, The Hague, the Netherlands.
5
Department of Psychiatry, Victor Segalen Bordeaux 2 University, and Hôpital Charles Perrens, Bordeaux, France.
Received for publication October 1, 2001; accepted for publication April 17, 2002.
Cannabis use may increase the risk of psychotic disorders and result in a poor prognosis for those with an
established vulnerability to psychosis. A 3-year follow-up (1997­1999) is reported of a general population of
4,045 psychosis-free persons and of 59 subjects in the Netherlands with a baseline diagnosis of psychotic
disorder. Substance use was assessed at baseline, 1-year follow-up, and 3-year follow-up. Baseline cannabis
use predicted the presence at follow-up of any level of psychotic symptoms (adjusted odds ratio (OR) = 2.76, 95%
confidence interval (CI): 1.18, 6.47), as well as a severe level of psychotic symptoms (OR = 24.17, 95% CI: 5.44,
107.46), and clinician assessment of the need for care for psychotic symptoms (OR = 12.01, 95% CI: 2.24,
64.34). The effect of baseline cannabis use was stronger than the effect at 1-year and 3-year follow-up, and more
than 50% of the psychosis diagnoses could be attributed to cannabis use. On the additive scale, the effect of
cannabis use was much stronger in those with a baseline diagnosis of psychotic disorder (risk difference, 54.7%)
than in those without (risk difference, 2.2%;
p
for interaction = 0.001). Results confirm previous suggestions that
cannabis use increases the risk of both the incidence of psychosis in psychosis-free persons and a poor
prognosis for those with an established vulnerability to psychotic disorder.
Am J Epidemiol
2002;156:319­27.
cannabis; drug utilization; psychoses, substance-induced; psychotic disorders; schizophrenia
Abbreviations: BPRS, Brief Psychiatric Rating Scale; CI, confidence interval; CIDI, Composite International Diagnostic
Interview; DSM-III-R,
Diagnostic and Statistical Manual of Mental Disorders
, Third Edition, Revised; OR, odds ratio; T1, time 1
(between baseline and 1997); T2, time 2 (between 1997 (T1) and 1999).
Although converging epidemiologic findings demonstrate
that the prevalence of cannabis use is higher among subjects
with psychosis than among subjects from the general popu-
lation (1­3), the mechanisms underlying this comorbid asso-
ciation have not been identified fully (4, 5). It is unclear
whether subjects with incipient psychosis use cannabis to
self-medicate their psychotic symptoms or, conversely,
whether exposure to cannabis is a risk factor for onset of
psychosis (6, 7). Because this issue cannot be disentangled
by using cross-sectional or retrospective studies, prospective
studies are essential to examine the temporal relation
between cannabis use and emergence of psychosis.
A previous Swedish study (8) showed that young men
using large quantities of cannabis at conscription were at
increased risk of being admitted for schizophrenia over the
subsequent 15-year period. However, to our knowledge,
these findings have not yet been replicated in another popu-
lation-based sample. Furthermore, the Swedish study was
hampered by several limitations, such as the lack of informa-
tion on drug use over the follow-up period and the fact that
Correspondence to Prof. Jim van Os, Department of Psychiatry and Neuropsychology, Maastricht University, P.O. Box 616 (DRT
10
), 6200 MD
Maastricht, the Netherlands (e-mail: j.vanos@sp.unimaas.nl).

Page 2
320
van Os et al.
Am J Epidemiol
Vol. 156, No. 4, 2002
psychosis cases were restricted to persons requiring hospital
admission.
Another unsolved question regarding the link between
cannabis use and psychosis is whether the impact of
cannabis use on subsequent psychosis, if any, is stronger in
subjects with a preexisting vulnerability to psychosis (9, 10).
Several studies of subjects with a hospital diagnosis of
psychosis have found that cannabis users have a poorer prog-
nosis than nonusers do (6, 11­13). Since these findings were
obtained in clinical samples, several factors linked to
hospital-based recruitment of subjects may confound the
association between cannabis use and poor outcome in
subjects with an established vulnerability to psychosis.
The aims of the present study were to test the following
hypotheses in a population-based sample of subjects
followed up over 3 years. First, cannabis use increases the
risk of psychosis, independent of the use of other drugs;
second, those with an established vulnerability to psychotic
disorder are more susceptible to this risk-increasing effect.
MATERIALS AND METHODS
Sample
This study was part of the Netherlands Mental Health
Survey and Incidence Study (NEMESIS), a longitudinal
study of the prevalence, incidence, course, and consequences
of psychiatric disorders in the Dutch general population. The
local ethics committee approved the study proposal as well
as the manner in which informed consent was obtained from
subjects. Subjects were contacted three times: in 1996 (base-
line), in 1997 (T1-assessing the period between baseline
and T1), and in 1999 (T2-assessing the period between T1
and T2) (14, 15).
A multistage, stratified, random sampling procedure was
used to identify the sample. First, 90 municipalities were
sampled randomly. Second, addresses of private households
were selected randomly. Third, the person in the private
household who had had the most recent birthday and was
aged 18­64 years was asked to participate. Persons living in
institutions, including those residing in psychiatric hospitals,
were not included in the sampling frame. A total of 7,076
subjects were enlisted at baseline. The response rate was
69.7 percent. No difference in psychiatric morbidity based
on the 12-item General Health Questionnaire was found
between responders and nonresponders (14, 15). At T1,
5,618 subjects participated for the second time; at T2, 4,848
subjects participated.
Instruments
The study sample was interviewed at home by using the
Composite International Diagnostic Interview (CIDI),
version 1.1 (16) for all three measurements. The CIDI gener-
ates
Diagnostic and Statistical Manual of Mental Disorders
,
Third Edition, Revised (DSM-III-R) diagnoses and is
designed for trained interviewers who are not clinicians.
Interviewers read the questions and record respondents'
answers, making the CIDI essentially a self-report instru-
ment (17). The CIDI psychosis section (G) consists of 17
core psychosis items (G1­G13, G15, G16, G20, and G21) on
delusions (13 items) and hallucinations (four items). These
psychosis items correspond to classic psychotic symptoms
such as persecution, thought interference, auditory halluci-
nations, and passivity phenomena.
Clinical reinterviews for psychotic symptoms
Because psychotic symptoms may be difficult to diagnose
by lay interviewers, especially the clinical relevance of such
symptoms (18­20), clinical reinterviews were conducted by
telephone by an experienced clinician (psychiatrist, senior
psychiatric trainee, or psychologist) for all subjects who had
evidence of significant psychosis at baseline and at T2 (21,
22). Since the yearly incidence of psychosis is very low, no
attempt was made to conduct clinical reinterviews at T1, just
1 year after the baseline interview. However, cases of
psychosis that were incident between baseline and T1 would
still have been identified at the T2 interview if the subjects
continued to experience symptoms between T1 and T2,
which is likely for the majority of cases who had any signif-
icant level of psychotic symptoms. The proportions of
eligible subjects who were reinterviewed successfully by the
clinician were 47.2 percent at baseline and 74.4 percent at
T2.The reinterviews were conducted by using questions from
the Structured Clinical Interview for DSM-III-R (SCID), an
instrument with proven reliability and validity in diagnosing
schizophrenia (21). If the clinician's CIDI psychosis
symptom rating did not coincide with that of the lay inter-
viewer, the rating of the lay interviewer was replaced with
that of the clinician. These corrected CIDI ratings were then
entered into the CIDI diagnostic program. The DSM-III-R
diagnoses of psychotic disorder at baseline were thus based
on the Structured Clinical Interview for DSM-III-R data
from these clinical reinterviews. Since no assessment of the
need for treatment was made at baseline for subjects with a
DSM-III-R diagnosis of psychosis (see below), this group
included subjects who were clinical psychosis cases but also
subjects whose psychotic experiences were not associated
with the need for treatment. Thus, these subjects are here-
after referred to in this paper as persons with an established
vulnerability to psychosis.
T2 assessment of incident psychotic symptoms
At T2, three Brief Psychiatric Rating Scale (BPRS) items
(22)-"unusual thought content," "hallucinations," and
"conceptual disorganization"-were additionally scored by
the clinician who conducted the telephone clinical reinter-
view. The scores for each symptom ranged from 1, "absent"
to 7, "very severe." Ratings of 2­3 indicate nonpathologic
intensities of symptoms, and 4­7 indicate pathologic intensi-
ties of symptoms (23). The BPRS items "unusual thought
content" and "hallucinations" represent the positive symp-
toms for psychosis and were used in the analyses in two
ways: 1) any rating of more than 1 for either of these items
(hereafter referred to as BPRS any psychosis) and 2) any
rating of 4 or more on either of these items (hereafter
referred to as BPRS pathology-level psychosis). To skew the

Page 3
Cannabis Use and Psychosis
321
Am J Epidemiol
Vol. 156, No. 4, 2002
sample toward subjects with a lifetime first-ever occurrence
of symptoms, we included only those subjects who, at the
baseline interview, had a rating of lifetime
absence
for all of
the individual items in the CIDI psychosis section.
T2 assessment of incident cases of psychosis
Since the most widely used system of classifying psychi-
atric disorders, the DSM-III-R (24), allocates "patient
status" on the basis of disability and distress rather than clin-
ical need, and persons in need of treatment may not be iden-
tified reliably, especially in population-based research (25),
we used a procedure yielding a needs-based diagnosis to
identify incident cases of psychosis at T2. Following each
interview by clinicians at T2, a consensus meeting was
organized that was attended by two psychologists and two
psychiatrists, during which all available information
regarding the case of psychosis was presented by the person
who had conducted the telephone interview. The four clini-
cians reached a consensus regarding whether there was a
need for mental health care in the context of psychotic symp-
toms, as defined in the Camberwell Assessment of Need
(CAN) (26). The diagnoses that resulted from the consensus
meeting were categorized as 1) no need for care in relation to
psychotic symptoms or 2) probable or definite need for care
in relation to psychotic symptoms. Validation of the needs-
based diagnosis has been described in previous papers (M.
Hanssen, Maastricht University, unpublished manuscript;
M. Bak, Maastricht University, unpublished manuscript).
Drug use assessment
The L section of CIDI assesses use of a variety of
substances. Included in the current analyses were questions
about the following substances: cannabis, psychostimulants,
cocaine, phencyclidine (PCP), and psychedelics. Psycho-
stimulants, cocaine, phencyclidine, and psychedelics were
combined into one group of "other drugs." Two types of
exposure variables were constructed: 1) any use (hereafter
referred to as any use) and 2) frequency of use over all three
assessment periods (hereafter referred to as cumulative
frequency). Any use at baseline, T1, or T2 was defined as
any frequency of use (lifetime any use at baseline, use over
the previous period at T1 and T2). Frequency of use during
the period of heaviest use was expressed on a 1­5 scale
(nearly every day, 3­4 days per week, 1­2 days per week, 1­
3 days per month, less than once a month). Cumulative
frequency was the overall longitudinal exposure defined as
the sum of the five-level frequency ratings at baseline, T1,
and T2. This score, with a range of 1­15, was also divided
into three groups (1­5, 6­10, 11­15).
Analyses
Associations between cannabis use and psychosis
outcome.
Associations between drug use and psychosis
outcome were expressed as odds ratios. The odds ratio
expressed the risk of developing the psychosis outcome for
those using drugs relative to those not using drugs. Adjusted
odds ratios were computed by using logistic regression (27)
in the STATA statistical software program (28). All analyses
were a priori adjusted for age (10-year groups), sex, ethnic
group (0, subject and both parents born in the Netherlands; 1,
other), level of education (four levels), unemployment, and
single marital status. In addition, we also controlled for two
previous risk factors for psychosis identified in the cohort:
urbanicity (three levels) and experience with discrimination
(four levels) (29).
To examine whether the effect of cannabis use at baseline
on psychosis at follow-up was a proxy effect of cannabis use
at follow-up, we entered cannabis any use at baseline,
cannabis any use at T1 follow-up, and cannabis any use at T2
follow-up simultaneously in the adjusted models of the two
psychosis outcomes. To assess whether any effect of
cannabis was independent of use of other drugs associated
with psychosis, models in which cannabis and other drugs
were entered jointly were compared with models in which
cannabis and other drugs were entered separately.
Risk set and sensitivity analyses.
All of the analyses,
with the exception of the interaction effects (see below),
were conducted in the group of subjects who 1) at baseline,
had a score of lifetime absence for all individual items in the
CIDI psychosis section (5,838 of 7,076 subjects, 82.5
percent (30)); 2) had had a CIDI interview at T2 (
n
= 4,848);
and 3) at T2, had not missed being reinterviewed by clini-
cians about the presence of psychotic symptoms if they had
been eligible for this clinical reinterview. This risk set
included 4,045 subjects. To check for possible fluctuations
in sample demographic stability occasioned by conditions 1
and 2, the age and sex distributions of the groups identified
by conditions 1 and 2 were compared with the risk set of
4,045 subjects. The proportion of men across the groups of
5,838 (condition 1), 4,848 (condition 2), and 4,045 (risk set)
was 52.6, 53.5, and 52.7 percent, respectively, and mean age
was 41.6, 41.2, and 41.5 years, respectively.
Because we did not reinterview all of the eligible subjects
at T2 who had shown evidence of psychosis (the clinical
reinterview rate at T2 was 74.4 percent, as noted above), we
conducted sensitivity analyses to examine whether differen-
tial attrition could have biased any findings. These analyses
were performed by substituting missing data, for the subjects
who had missed clinical reinterview, in such a way that the
extremes of any bias could be quantified. In the first sensi-
tivity analysis, all missing subjects were allocated to the
category of caseness of BPRS pathology-level psychosis; in
the second type of analysis, all missing subjects were allo-
cated to the category of noncaseness. In the same manner, we
tested whether attrition in the sample as a whole (
n =
7,076
at baseline,
n =
4,848 at T2) could have biased the findings.
Population attributable fraction.
The population attribut-
able risk fraction, or the proportion of psychosis outcomes
that could have been prevented if cannabis use 1) were a
causal risk factor and 2) were eliminated completely from
the population, is a measure of the public health importance
of an exposure. It was derived from the adjusted associations
between cannabis use and psychosis in the logistic regres-
sion models by using the AFLOGIT procedure in STATA
software (28), which allows for estimating the population
attributable fraction from within a logistic regression frame-
work, thus enabling confounders to be taken into account.

Page 4
322
van Os et al.
Am J Epidemiol
Vol. 156, No. 4, 2002
Interaction between cannabis use and vulnerability to
psychosis.
To assess the effect of cannabis in those with an
established vulnerability to psychosis, the effect of cannabis
on the psychosis outcomes at T2 was estimated in the group
of subjects who, at baseline, had had a DSM-III-R lifetime
diagnosis of any affective or nonaffective psychotic disorder
(
n
= 59 with baseline and T2 outcome data) and was then
compared with the effect of cannabis in the risk set (
n
=
4,045). The group of 59 subjects who had a psychotic
disorder and T2 outcome data came from a larger group of
107 who had a baseline DSM-III-R diagnosis of psychotic
disorder. A comparison between the 59 psychosis cases with
and the 48 without T2 outcome data revealed no differences
in terms of baseline cannabis use (
2
= 1.0, df = 1,
p
= 0.31),
age (
t
= ­0.55, df = 105,
p
= 0.58), or sex (
2
= 2.7, df = 1,
p
= 0.10). In line with recent advances in the conceptualiza-
tion of interaction, we calculated the statistical additive
interaction and estimated from that the population amount of
biologic synergism between cannabis use and psychosis
vulnerability (refer to the Appendix) (31). To calculate the
statistical interaction under an additive model, the BINREG
procedure in STATA software (28), which fits generalized
linear models for the binomial family estimating risk differ-
ences, was used to model interactions between cannabis (any
use) and psychosis vulnerability (any lifetime diagnosis of
psychosis).
RESULTS
Cannabis use and psychosis
In the risk set of 4,045 subjects who, at the baseline inter-
view, had a rating of lifetime absence for all individual items
in the CIDI psychosis section, seven subjects (0.17 percent)
at T2 had a probable/definite need for care related to
psychotic symptoms. The number of subjects at T2 with
BPRS any psychosis was 38 (0.94 percent), and the number
with BPRS pathology-level psychosis was 10 (0.25 percent).
Subjects with a psychosis outcome at T2 had higher levels
of cannabis use at baseline (table 1). Cannabis any use at
baseline and cumulative frequency were associated with a
high risk of psychosis outcome at T2 (table 2). The associa-
tions generally remained significant after adjustment for age,
sex, ethnic group, single marital status, level of education,
urbanicity, and level of discrimination. Additional adjust-
ment for the presence of any CIDI lifetime diagnosis at base-
line (to exclude the possibility that cannabis use at baseline
was secondary to a nonpsychotic DSM-III-R diagnosis,
which, in turn, was a prodrome of later psychosis) changed
the parameters by only a tiny amount (e.g., BPRS pathology-
level outcome-cannabis any use at baseline: odds ratio
(OR) = 23.32, 95 percent confidence interval (CI): 4.92,
110.50; cumulative frequency: OR = 4.17, 95 percent CI:
2.34, 7.42).
Distal versus proximal effects
For the three psychosis outcomes, the effect of baseline
cannabis use was much stronger than the effects of more
proximal cannabis use at T1 and T2. This finding indicated
that the effects of cannabis use at baseline on psychosis at
follow-up was not a proxy effect of cannabis use at follow-
up (table 3).
Cannabis and other substances
Use of other drugs, as defined in Materials and Methods,
was strongly associated with the three psychosis outcomes
when entered separately in the model. However, when
entered jointly with cannabis, the effects were greatly
reduced or even disappeared altogether, while the effect of
cannabis remained (table 4).
Population attributable fraction
The population attributable fraction was calculated for the
"any use" exposure and the three outcomes. The population
attributable fractions were 13.4 percent for BPRS any
psychosis, 67.1 percent for BPRS pathology-level psychosis,
and 50.4 percent for needs-based diagnosis of psychotic
TABLE 1. Patterns of cannabis use and psychosis outcome, the Netherlands Mental Health Survey and Incidence Study, 1996­1999
*
T2, time 2 (between 1997 and 1999); BPRS, Brief Psychiatric Rating Scale (Psychol Rep 1962;10:799­812); ­, absence of the outcome; +,
presence of the outcome.
Some percentages do not total 100 because of rounding.
Cannabis exposure
T2
*
psychosis outcome
BPRS
*
any psychosis
BPRS pathology-level psychosis
Needs-based diagnosis of psychotic disorder
Outcome­
*
(
n
= 4,007)
Outcome+
*
(
n
= 38)
Outcome­
(
n
= 4,035)
Outcome+
(
n
= 10)
Outcome­
(
n
= 4,038)
Outcome+
(
n
= 7)
No.
%
No.
%
No.
%
No.
%
No.
%
No.
%
Baseline any use
304
7.59
8
21.05
305
7.56
7
70
308
7.63
4
57.14
Cumulative
frequency
No use
3,622
91.7
30
79.0
3,649
91.7
3
30.0
3,649
91.6
3
42.9
Lowest level
264
6.7
3
7.9
265
6.7
2
20.0
265
6.7
2
28.6
Middle level
34
0.9
2
5.3
34
0.9
2
20.0
36
0.9
0
Highest level
32
0.8
3
7.9
32
0.8
3
30.0
33
0.8
2
28.6

Page 5
Cannabis Use and Psychosis
323
Am J Epidemiol
Vol. 156, No. 4, 2002
disorder. These fractions indicated that, assuming that the
relation between cannabis use and psychosis is causal, the
incidence of the three psychosis outcomes would be reduced
by 13, 67, and 50 percent, respectively, if the cannabis expo-
sure were eliminated from the population.
Interaction with vulnerability to psychosis
Of the 59 subjects with a DSM-III-R diagnosis of any
psychosis at baseline for whom follow-up data at T2 were
available, nine (15.3 percent) reported cannabis any use at
baseline. In the risk set, this proportion was 7.7 percent (
n
=
312). On the additive scale, the increase in the risk of having
the psychosis outcome at T2 was much higher in the group
with a baseline diagnosis of psychotic disorder than in that
without such a diagnosis, although the increase was still
significant in the latter group. The difference in risk between
those with and without psychotic disorder at baseline was
statistically significant for two psychosis outcomes (table 5).
By following the procedure developed by Darroch (31)
(refer to the Appendix), we calculated what proportion of the
BPRS pathology-level psychosis outcome in subjects both
TABLE 2. Associations between cannabis use and time 2 psychosis outcome, the Netherlands Mental Health Survey and Incidence
Study, 1996­1999
*
T2, time 2 (between 1997 and 1999); BPRS, Brief Psychiatric Rating Scale (Psychol Rep 1962;10:799­812); OR, odds ratio; CI, confidence interval.
Adjusted for age, sex, ethnic group, single marital status, level of education, urbanicity, and level of discrimination.
Reference category.
§ No exposed persons had the outcome.
¶ The increase in risk with one unit change in cannabis frequency.
Cannabis exposure
T2
*
psychosis outcome
BPRS
*
any psychosis
(
n
= 38)
BPRS pathology-level psychosis
(
n
= 10)
Needs-based diagnosis of psychotic disorder
(
n
= 7)
OR
*
95% CI
*
Adjusted
OR
95% CI
OR
95% CI
Adjusted
OR
95% CI
OR
95% CI
Adjusted
OR
95% CI
Baseline any
use
3.25 1.48, 7.15
2.76
1.18, 6.47
28.54
7.34, 110.91
24.17
5.44, 107.46
16.15 3.60, 72.47
12.01
2.24, 64.34
Cumulative
frequency
No use
1
1
1
1
1
1
Lowest level
1.37 0.42, 4.52
1.23
0.36, 4.23
9.18
1.53, 55.18
7.90
1.15, 54.09
9.18 1.53, 55.18
5.10
0.66, 39.21
Middle level
7.10 1.63, 30.91
4.90
10.4, 23.14
71.55 11.58, 441.94
54.46
7.17, 413.73
Highest level 11.32 3.29, 38.99
6.81
1.79, 25.92
114.03 22.17, 586.50
74.67
11.76, 474.32
73.72 11.92, 455.76
47.77
5.91, 385.94
Linear trend¶
2.23 1.53, 3.24
1.89
1.25, 2.85
4.96
3.08, 8.00
4.27
2.44, 7.45
3.97 2.22, 7.10
3.53
1.76, 7.09
TABLE 3. Effects of cannabis use distal and proximal to psychosis outcome, the Netherlands Mental Health Survey and Incidence
Study, 1996­1999
*
T2, time 2 (between 1997 (T1) and 1999); BPRS, Brief Psychiatric Rating Scale (Psychol Rep 1962;10:799­812); OR, odds ratio; CI,
confidence interval; T1, time 1 (between baseline and 1997).
Adjusted for age, sex, ethnic group, single marital status, level of education, urbanicity, and level of discrimination; reference category,
those subjects who did not use cannabis at the three specified time points.
Drug exposure
T2
*
psychosis outcome
BPRS
*
any psychosis
(
n
= 38)
BPRS pathology-level psychosis
(
n
= 10)
Needs-based diagnosis of psychotic
disorder (
n
= 7)
Adjusted OR
*
95% CI
*
Adjusted OR
95% CI
Adjusted OR
95% CI
Separate effects
Any use at baseline
2.76
1.18, 6.47
24.17
5.44, 107.46
12.01
2.24, 64.34
Any use at T1
*
2.96
0.96, 9.22
14.91
3.39, 65.57
11.07
1.62, 75.76
Any use at T2
3.17
1.02, 9.92
15.08
3.35, 68.00
9.36
1.39, 63.13
Jointly entered in the same
model
Any use at baseline
2.22
0.78, 6.31
15.78
2.83, 88.05
7.73
1.09, 54.74
Any use at T1
1.13
0.18, 7.15
1.35
0.14, 13.11
1.75
0.08, 40.35
Any use at T2
1.68
0.28, 10.19
2.32
0.25, 21.90
1.86
0.09, 38.52

Page 6
324
van Os et al.
Am J Epidemiol
Vol. 156, No. 4, 2002
exposed to cannabis and having an established vulnerability
to psychosis was attributable to the synergistic action of
these two factors. This calculation revealed that between 79
and 82 percent was due to the synergistic action of these two
factors.
Sensitivity analyses
The results regarding the BPRS pathology-level psychosis
outcome and missing clinical reinterview data were as
follows: 1) assuming that all subjects for whom values were
missing were noncases-cannabis any use: OR = 24.36, 95
percent CI: 5.48, 108.37; cannabis cumulative frequency: OR
= 4.27, 95 percent CI: 2.45, 7.47; and 2) assuming that all
subjects for whom values were missing were cases-cannabis
any use: OR = 3.30, 95 percent CI: 1.30, 8.33; cannabis cumu-
lative frequency: OR = 2.17, 95 percent CI: 1.41, 3.34. The
results regarding the BPRS pathology-level psychosis
outcome and missing data due to general sample attrition were
as follows: 1) assuming that all subjects for whom values
TABLE 4. Effects of cannabis use on psychosis in relation to use of other substances, the Netherlands Mental Health Survey and
Incidence Study, 1996­1999
*
T2, time 2 (between 1997 and 1999); BPRS, Brief Psychiatric Rating Scale (Psychol Rep 1962;10:799­812); OR, odds ratio; CI, confidence
interval.
Adjusted for age, sex, ethnic group, single marital status, level of education, urbanicity, and level of discrimination.
Reference category, those subjects who did not use cannabis or other drugs at baseline (baseline any use) or at all three time points
(cumulative frequency).
§ Psychostimulants, cocaine, phencyclidine (PCP), and psychedelics were combined into one group of "other drugs."
¶ Linear trend for the adjusted odds ratio: an increase in risk with one unit change in cannabis frequency.
Drug exposure
T2
*
psychosis outcome
BPRS
*
any psychosis
(
n
= 38)
BPRS pathology-level psychosis
(
n
= 10)
Needs-based diagnosis of psychotic
disorder (
n
= 7)
Adjusted OR
*
95% CI
*
Adjusted OR
95% CI
Adjusted OR
95% CI
Separate effects
Baseline any use
Cannabis
2.76
1.18, 6.47
24.17
5.44, 107.46
12.01
2.24, 64.34
Other drugs ,§
5.38
1.48, 19.56
23.47
4.99, 110.35
9.55
0.90, 101.52
Cumulative frequency¶
Cannabis
1.89
1.25, 2.85
4.27
2.44, 7.45
3.53
1.76, 7.09
Other drugs
2.57
1.31, 5.04
4.59
2.17, 9.74
2.84
0.95, 8.49
Jointly entered in the same model
Baseline any use
Cannabis
2.11
0.78, 5.71
16.93
3.33, 86.13
10.51
1.75, 63.21
Other drugs
2.99
0.68, 13.25
3.95
0.77, 20.27
1.89
0.15, 23.57
Cumulative frequency¶
Cannabis
1.65
1.00, 2.73
3.73
1.99, 7.01
3.47
1.64, 7.37
Other drugs
1.64
0.71, 3.80
1.83
0.75, 4.50
1.09
0.26, 4.52
TABLE 5. Interactions between cannabis any use and existing vulnerability to psychosis on the additive scale (risk difference), the
Netherlands Mental Health Survey and Incidence Study, 1996­1999
*
BPRS, Brief Psychiatric Rating Scale (Psychol Rep 1962;10:799­812); CI, confidence interval.
Risk of having the psychosis outcome at T2 (time 2 (between 1997 and 1999)).
Tests whether an increase in risk in the "any psychosis" group is significantly greater than an increase in risk in the "no psychosis" group.
BPRS
*
any psychosis (
n
= 54)
BPRS pathology-level psychosis
(
n
= 22)
Needs-based diagnosis of psychotic
disorder (
n
= 15)
Increase (%)
95% CI
*
Increase (%)
95% CI
Increase (%)
95% CI
Increase in risk associated with
baseline cannabis any use
No psychosis (
n
= 4,045)
1.8
0.0, 3.5
2.2
0.5, 3.8
1.3
­0.05, 2.5
Any psychosis (
n
= 59)
46.7
13.9, 79.4
54.7
22.6, 86.8
23.3
­8.6, 55.2
Additive interaction
2
= 7.21, df = 1,
p
= 0.0073
2
= 10.26, df = 1,
p
= 0.0014
2
= 1.85, df = 1,
p
= 0.17

Page 7
Cannabis Use and Psychosis
325
Am J Epidemiol
Vol. 156, No. 4, 2002
were missing were noncases-cannabis any use: OR = 25.17,
95 percent CI: 5.64, 112.24; cannabis cumulative frequency:
OR = 4.27, 95 percent CI: 2.44, 7.45; and 2) assuming that all
subjects for whom values were missing were cases-cannabis
any use: OR = 1.11, 95 percent CI: 0.89, 1.38; cannabis cumu-
lative frequency: OR = 4.27, 95 percent CI: 2.44, 7.45.
DISCUSSION
This population-based prospective study showed that a
baseline history of cannabis use increased the risk of a
follow-up psychosis outcome for subjects with a lifetime
absence of psychosis, with a dose-response relation between
exposure load and psychosis outcome. A baseline lifetime
history of cannabis use was a stronger predictor of psychosis
outcome than was use over the follow-up period and use of
other drugs. A strong additive interaction was found between
cannabis use and established vulnerability to psychotic
disorder: the difference in risk of psychosis at follow-up
between those who did and did not use cannabis was much
stronger for those with an established vulnerability to
psychosis at baseline than for those without one. Sensitivity
analyses showed that differential attrition was unlikely to
have contributed to the results. In addition, previous analyses
of this sample have shown that psychopathology had only
weak-to-moderate effects on attrition at T1 (32).
The present findings have to be interpreted in light of poten-
tial methodological limitations. We cannot exclude underre-
porting of drug use, because it was assessed by using self-
reported data and was not confirmed with toxicologic
screening. However, urine testing does not provide informa-
tion on lifetime use, and there is little reason to suspect that
cannabis use was underreported; personal use of cannabis is
legal in the Netherlands, and cannabis is widely accepted as a
recreational drug. As a consequence of the limited number of
subjects who had a needs-based psychotic disorder, the effect
size of some associations could not be calculated, and some
interval estimations were imprecise. Some cases of psychosis
arising between baseline and T1, when no clinical reinter-
views were conducted, may have been missed at the T2 inter-
view if subjects' symptoms did not persist beyond the T1
interview. This possibility could, in theory, have biased our
findings if these had been the cases in which cannabis had no
effect, or a protective effect, on the development of psychotic
symptoms. However, when we examined the association
between baseline any cannabis use and any self-report of
psychotic symptoms at T1 in subjects who had been psychosis
free at baseline, the odds ratio was large and in the same direc-
tion (OR = 2.64, 95 percent CI: 1.54, 4.52). Although this
association was based on self-reported psychotic symptoms at
T1 instead of clinical assessment, it is unlikely that clinical
reinterview would have substantially changed the strength or
direction of this association.
In accordance with the findings reported by Andreasson et
al. (8), the present study shows that psychosis-free subjects
who have a lifetime history of cannabis use are at increased
risk of a psychosis outcome. As in the Swedish study (8),
further evidence supporting the hypothesis of a causal rela-
tion is demonstrated by the existence of a dose-response
relation (33) between cumulative exposure to cannabis use
and the psychosis outcome.
The strengths of the present study are that drug use was
documented in the context of a structured diagnostic inter-
view at baseline and over the follow-up period and that defi-
nition of psychosis outcome was not restricted to cases of
psychosis requiring hospitalization, which are but the
extreme of a continuum of psychotic experiences (29). Thus,
the present findings provide answers to questions raised by
the Swedish cohort study (8). First, the association between
cannabis use and psychosis outcome is not restricted to the
most severe outcome; that is, there is a continuum of risk
ranging from increased occurrence of psychotic symptoms
to incidence of cases in need of treatment. Second, the asso-
ciation is independent of use of other drugs at baseline and
over the follow-up period. Third, the finding that psychosis
outcome is more strongly predicted by a baseline lifetime
history of use than by recent use contributes to clarifying the
temporal relation between cannabis exposure and increased
risk of psychosis. This finding suggests that this association
is not fully explained by the short-term effects of cannabis
leading to acute occurrence of psychotic experiences (5).
Any claim of a causal relation between cannabis use and
psychosis would be further supported by a plausible biologic
mechanism (33, 34). Long-term effects of cannabis on the risk
of psychosis outcome may be due to long-lasting dysregula-
tion of endogenous anadamide/cannabinoid systems medi-
ating the effects of tetrahydrocannabinol on the brain. Other
neurotransmission systems modulated by cannabinoid recep-
tors may also be involved. Some recent findings indirectly
support this cannabinoid-receptor hypothesis. An increased
density of cannabinoid-1 receptors has been found in the
caudate-putamen of cannabis users (35), and there are close
interactions between cannabinoid and dopaminergic systems
(36) that are thought to underlie psychotic symptoms. Experi-
mental evidence in rodents has demonstrated that chronic
exposure to delta9-tetrahydrocannabinol induces sensitization
of monoaminergic neurotransmitter systems thought to be
involved in psychosis (37). A limited number of studies have
reported changes in the endogenous cannabinoid concentra-
tion (38) or in the number of cannabinoid receptors (35) in
subjects who have schizophrenia. These findings, which have
to be interpreted with caution since they have been obtained in
samples of patients with chronic disease, nevertheless suggest
that dysregulation of the cannabinoid system may be impli-
cated in the pathophysiology of psychosis. It can be
hypothesized that the neurobiologic changes induced by
tetrahydrocannabinol may interact with a preexisting vulnera-
bility to dysregulation of the cannabinoid system or to other
neurotransmission systems interacting with the cannabinoid
system. In accordance with this hypothesis, the present find-
ings demonstrate that the impact of cannabis use on psychosis
outcome is especially marked in subjects with an established
vulnerability to psychosis. The difference in the risk of
psychosis at follow-up between those who did and did not use
cannabis was much stronger for those with a baseline vulner-
ability to psychosis (54.7 percent) than for those without a
baseline experience of psychosis (2.2 percent).
About 80 percent of the psychosis outcome associated
with exposure to both cannabis and an established vulnera-

Page 8
326
van Os et al.
Am J Epidemiol
Vol. 156, No. 4, 2002
bility to psychosis was attributable to the synergistic action
of these two factors. This finding indicates that, of the
subjects exposed to both a vulnerability to psychosis and
cannabis use, approximately 80 percent had the psychosis
outcome because of the combined action of the two risk
factors and only about 20 percent because of the action of
either factor alone.
In conclusion, this prospective study confirms previous
suggestions that cannabis use is an independent risk factor for
the emergence of psychosis in psychosis-free persons and
that those with an established vulnerability to psychotic
disorder are particularly sensitive to its effects, resulting in a
poor outcome. These findings have public health implica-
tions. If a causal relation between cannabis use and psychosis
outcome is assumed, the population attributable fraction-
that is, the maximum proportion of psychosis outcomes
attributable to cannabis use in psychosis-free subjects-is
higher than 50 percent. Although this high percentage can be
misleading because it also includes the effects of all other
causal risk factors that interact with cannabis, there is never-
theless cause for concern given the widespread use of
cannabis by adolescents and young adults (39­41). The
percentage of cases that may be prevented by suppressing
exposure to this risk factor may not be negligible.
ACKNOWLEDGMENTS
Supported by a grant from the Dutch Ministry of Health.
REFERENCES
1. Degenhardt L, Hall W. The association between psychosis and
problematical drug use among Australian adults: findings from
the National Survey of Mental Health and Well-Being. Psychol
Med 2001;31:659­68.
2. Regier D, Farmer M, Rae D, et al. Comorbidity of mental disor-
ders with alcohol and other drug abuse. Results from the Epide-
miologic Catchment Area (ECA) Study. JAMA 1990;264:
2511­18.
3. Kessler R, Crum R, Warner L, et al. Lifetime co-occurrence of
DSM-III-R alcohol abuse and dependence with other psychiat-
ric disorders in the National Comorbidity Survey. Arch Gen
Psychiatry 1997;54:313­21.
4. Thornicroft G. Cannabis and psychosis. Is there epidemiologi-
cal evidence for an association? Br J Psychiatry 1990;157:25­
33.
5. Johns A. Psychiatric effects of cannabis. Br J Psychiatry 2001;
178:116­22.
6. Linszen DH, Dingemans P, Lenior M. Cannabis abuse and the
course of recent-onset schizophrenic disorders. Arch Gen Psy-
chiatry 1994;51:273­9.
7. Hambrecht M, Hafner H. Substance abuse and the onset of
schizophrenia. Biol Psychiatry 1996;40:1155­63.
8. Andreasson S, Allebeck P, Engstrom A, et al. Cannabis and
schizophrenia. A longitudinal study of Swedish conscripts.
Lancet 1987;2:1483­6.
9. McGuire PK, Jones P, Harvey I, et al. Morbid risk of schizo-
phrenia for relatives of patients with cannabis-associated psy-
chosis. Schizophr Res 1995;15:277­81.
10. van Os J, Marcelis M. The ecogenetics of schizophrenia: a
review. Schizophr Res 1998;32:127­35.
11. Martinez-Arevalo MJ, Calcedo-Ordonez A, Varo-Prieto JR.
Cannabis consumption as a prognostic factor in schizophrenia.
Br J Psychiatry 1994;164:679­81.
12. Kovasznay B, Fleischer J, Tanenberg-Karant M, et al. Sub-
stance use disorder and the early course of illness in schizophre-
nia and affective psychosis. Schizophr Bull 1997;23:195­201.
13. Verdoux H, Liraud F, Gonzales B, et al. Suicidality and sub-
stance misuse in first-admitted subjects with psychotic disor-
der. Acta Psychiatr Scand 1999;100:389­95.
14. Bijl RV, van Zessen G, Ravelli A, et al. The Netherlands Mental
Health Survey and Incidence Study (NEMESIS): objectives and
design. Soc Psychiatry Psychiatr Epidemiol 1998;33:581­6.
15. Bijl RV, Ravelli A, van Zessen G. Prevalence of psychiatric
disorder in the general population: results of the Netherlands
Mental Health Survey and Incidence Study (NEMESIS). Soc
Psychiatry Psychiatr Epidemiol 1998;33:587­95.
16. Smeets RMW, Dingemans PMAJ. Composite International
Diagnostic Interview (CIDI). Version 1.1. Geneva, Switzer-
land: World Health Organization, 1993.
17. Eaton WW, Neufeld K, Chen LS, et al. A comparison of self-
report and clinical diagnostic interviews for depression: diag-
nostic interview schedule and schedules for clinical assessment
in neuropsychiatry in the Baltimore epidemiologic catchment
area follow-up. Arch Gen Psychiatry 2000;57:217­22.
18. Helzer JE, Robins LN, McEvoy LT, et al. A comparison of clin-
ical and diagnostic interview schedule diagnoses. Physician
reexamination of lay-interviewed cases in the general popula-
tion. Arch Gen Psychiatry 1985;42:657­66.
19. Anthony JC, Folstein M, Romanoski AJ, et al. Comparison of
the lay diagnostic interview schedule and a standardized psy-
chiatric diagnosis. Experience in eastern Baltimore. Arch Gen
Psychiatry 1985;42:667­75.
20. Cooper SA, Collacott RA. Relapse of depression in people with
Down's syndrome. Br J Develop Disabil 1994;40:32­7.
21. Spitzer RL, Williams JB, Gibbon M, et al. The Structured Clin-
ical Interview for DSM-III-R (SCID). I: history, rationale, and
description. Arch Gen Psychiatry 1992;49:624­9.
22. Overall JE, Gorham DR. The Brief Psychiatric Rating Scale.
Psychol Rep 1962;10:799­812.
23. Lukoff D, Nuechterlein KH, Ventura J. Manual for the
Expanded Brief Psychiatric Rating Scale. Schizophr Bull 1986;
12:594­602.
24. American Psychiatric Association. Diagnostic and statistical
manual of mental disorders: DSM-III-R. 3rd ed, rev. Washing-
ton, DC: American Psychiatric Association, 1987.
25. Spitzer RL. Diagnosis and need for treatment are not the same.
(Comment). Arch Gen Psychiatry 1998;55:120.
26. Slade M, Phelan M, Thornicroft G, et al. The Camberwell
Assessment of Need (CAN): comparison of assessments by
staff and patients of the needs of the severely mentally ill. Soc
Psychiatry Psychiatr Epidemiol 1996;31:109­13.
27. Clayton D, Hills M. Poisson and logistic regression. In: Clayton
D, Hills M, eds. Statistical models in epidemiology. Oxford,
United Kingdom: Oxford Science Publications, 1993:227­36.
28. Stata Corporation. STATA statistical software, release 7.0. Col-
lege Station, Texas: Stata Corporation, 2001.
29. van Os J, Hanssen M, Bijl RV, et al. Prevalence of psychotic
disorder and community level of psychotic symptoms: an
urban-rural comparison. Arch Gen Psychiatry 2001;58:663­8.
30. van Os J, Hanssen M, Bijl RV, et al. Strauss (1969) revisited: a
psychosis continuum in the general population? Schizophr Res
2000;45:11­20.

Page 9
Cannabis Use and Psychosis
327
Am J Epidemiol
Vol. 156, No. 4, 2002
31. Darroch J. Biologic synergism and parallelism. Am J Epide-
miol 1997;145:661­8. (Comments published in Am J Epide-
miol 1998;147:89­90).
32. de Graaf R, Bijl RV, Smit F, et al. Psychiatric and sociodemo-
graphic predictors of attrition in a longitudinal study: the Neth-
erlands Mental Health Survey and Incidence Study
(NEMESIS). Am J Epidemiol 2000;152:1039­47.
33. Hill A. The environment and disease: association or causation?
J Royal Soc Med 1965;58:295­300.
34. Susser M. What is a cause and how do we know one? A gram-
mar for pragmatic epidemiology. Am J Epidemiol 1991;133:
635­48.
35. Dean B, Sundram S, Bradbury R, et al. Studies on [3H]CP­
55940 binding in the human central nervous system: regional
specific changes in density of cannabinoid-1 receptors associ-
ated with schizophrenia and cannabis use. Neuroscience 2001;
103:9­15.
36. Tanda G, Pontieri F, Di Chiara G. Cannabinoid and heroin acti-
vation of mesolimbic dopamine transmission by a common
µ
1
opoid receptor mechanism. Science 1997;276:2048­50.
37. Gorriti MA, Rodriguez de Fonseca F, Navarro M, et al. Chronic
(-)-delta9-tetrahydrocannabinol treatment induces sensitization
to the psychomotor effects of amphetamine in rats. Eur J Phar-
macol 1999;365:133­42.
38. Leweke FM, Giuffrida A, Wurster U, et al. Elevated endog-
enous cannabinoids in schizophrenia. Neuroreport 1999;10:
1665­9.
39. Webb E, Ashton C, Kelly P, et al. Alcohol and drug use in UK
university students. Lancet 1996;348:922­5.
40. Perkonigg A, Lieb R, Hofler M, et al. Patterns of cannabis use,
abuse and dependence over time: incidence, progression and
stability in a sample of 1228 adolescents. Addiction 1999;94:
1663­78.
41. Smart R, Ogborne A. Drug use and drinking among students in
36 countries. Addict Behav 2000;25:455­60.
APPENDIX
Estimating the Amount of Biologic Synergism between
Two Causes
Let us assume that there are two risk factors for schizo-
phrenia,
E
and
G
.
Risk
is a measure of the proportion of
persons who develop schizophrenia. If there are two risk
factors,
G
and
E
, there are four possible exposure states
according to whether each factor is present (+) or absent (­),
and each of these four exposure states carries a specific risk.
Thus, the risk of schizophrenia in the population exposed to
E
only is
R
(
E
), and the risk in the population exposed to
G
only is
R
(
G
). The risk of schizophrenia in the population
exposed to neither
E
nor
G
is
R
, whereas the risk in the popu-
lation exposed to both
G
and
E
is
R
(
GE
). On the additive
scale, the
effect
of a risk factor is expressed as a
risk differ-
ence
. For example, if
R
(
G
) is 0.25 and
R
is 0.10, the effect of
G
is 0.25 ­ 0.10 = 0.15. We can thus express the effect of
G
as
R
(
G
) ­
R
, the effect associated with
E
as
R
(
E
) ­
R
, and the
effect associated with the
GE
exposure as
R
(
GE
) ­
R
. The
following table shows the effects associated with the four
different exposure states:
Because the combined effect of
G
and
E
is
R
(
GE
) ­
R
, the
excess of this effect over the sum of the solitary effects of
G
and
E
is as follows: [
R
(
GE
) ­
R
] ­ [
R
(
G
) ­
R
] ­ [
R
(
E
) ­
R
] =
[
R
(
GE
) ­
R
(
G
) ­
R
(
E
) +
R
].
If [
R
(
GE
) ­
R
(
G
) ­
R
(
E
) +
R
] > 0,
G
and
E
are said to interact
on the additive scale. We will hereafter refer to [
R
(
GE
) ­
R
(
G
) ­
R
(
E
) +
R
] as the statistical additive interaction.
How can we quantify the extent to which
G
and
E
act
synergistically, that is, in some way depend on each other, or
coparticipate, in disease causation? Let us consider the
proportion of persons in the population who developed schiz-
ophrenia after exposure to both
G
and
E
, or
R
(
GE
). It is
possible that some of these persons would also have
contracted the disorder after exposure to either
G
or
E
alone.
The degree to which some persons would also have
contracted the disorder after exposure to either
G
or
E
alone
is referred to as the degree of
parallelism
. If there is paral-
lelism,
G
and
E
"compete" to cause schizophrenia, and, the
more they compete, the smaller the proportion of persons
who contracted the disease because of the
coparticipation
of
G
and
E
. Thus, parallelism can be thought of as the opposite
of synergism. For example, in the extreme case of 100
percent parallelism, where all persons exposed to
G
and
E
had developed the disease because of the causal action of
either
G
or
E
alone, no person could have contracted schizo-
phrenia because of the coparticipation of
G
and
E
. In this
case, the amount of synergism would be zero. In practice, it is
impossible to assess the amount of parallelism and the
amount of synergy in persons exposed to both
G
and
E
.
However, it can be shown that the amount by which syner-
gism exceeds parallelism equals the excess of
R
(
GE
) over the
sum of the solitary effects of
G
and
E
(i.e., the statistical addi-
tive interaction as shown above) (33). In other words, |syner-
gism| ­ |parallelism| = [
R
(
GE
) ­
R
(
G
) ­
R
(
E
) +
R
].
The amount of synergy can then be approximated by using
the following table (33):
The variables
x
1 and
x
2 are two unknowns that sum with
synergism and parallelism to [
R
(
GE
) ­
R
(
E
)] and [
R
(
E
) ­ R],
respectively. In our study, the risks were
R
= 0.08 percent;
R
(
G
) = 12 percent,
R
(
E
) = 2.2 percent; and
R
(
GE
) = 67
percent. Filling in these risks in the table above reveals that
synergism is 79­82 percent.
G
+
R
R
(
G
) ­
R
E
+
R
(
E
) ­
R
R
(
GE
) ­
R
|synergism|
|
x
2|
R
(
GE
) ­
R
(
G
)
|
x
1|
|parallelism|
R
(
G
) ­
R
R
(
GE
) ­
R
(
E
)
R
(
E
) ­
R