You are on page 1of 10

Psychological Medicine (2017), 47, 16681677.

Cambridge University Press 2017 OR I G I N A L A R T I C L E


Association between alcohol, cannabis, and other

illicit substance abuse and risk of developing
schizophrenia: a nationwide population based
register study

S. M. Nielsen1,2, N. G. Toftdahl1,2, M. Nordentoft1,2 and C. Hjorthj1,2*

Copenhagen University Hospital, Mental Health Center Copenhagen, Hellerup, Denmark
The Lundbeck Foundation Initiative for Integrative Psychiatric Research, iPSYCH, Aarhus and Copenhagen, Denmark

Background. Several studies have examined whether use of substances can cause schizophrenia. However, due to meth-
odological limitations in the existing literature (e.g. selection bias and lack of adjustment of co-abuse) uncertainties still
remain. We aimed to investigate whether substance abuse increases the risk of developing schizophrenia, addressing
some of these limitations.

Method. The longitudinal, nationwide Danish registers were linked to establish a cohort of 3 133 968 individuals
(105 178 673 person-years at risk), identifying 204 505 individuals diagnosed with substance abuse and 21 305 diagnosed
with schizophrenia. Information regarding substance abuse was extracted from several registers and did not include psych-
otic symptoms caused by substance abuse in the denition. This resulted in a large, generalizable sample of exposed
individuals. The data was analysed using Cox regression analyses, and adjusted for calendar year, gender, urbanicity,
co-abuse, other psychiatric diagnosis, parental substance abuse, psychiatric history, immigration and socioeconomic status.

Results. A diagnosis of substance abuse increased the overall risk of developing schizophrenia [hazard ratio (HR) 6.04,
95% condence interval (CI) 5.846.26]. Cannabis (HR 5.20, 95% CI 4.865.57) and alcohol (HR 3.38, 95% CI 3.243.53) pre-
sented the strongest associations. Abuse of hallucinogens (HR 1.86, 95% CI 1.432.41), sedatives (HR 1.68, 95% CI 1.49
1.90), and other substances (HR 2.85, 95% CI 2.583.15) also increased the risk signicantly. The risk was found to be sign-
icant even 1015 years subsequent to a diagnosis of substance abuse.

Conclusion. Our results illustrate robust associations between almost any type of substance abuse and an increased risk of
developing schizophrenia later in life.

Received 31 August 2016; Revised 7 January 2017; Accepted 10 January 2017; First published online 7 February 2017

Key words: Alcohol, cannabis, schizophrenia, substance use disorders.

Introduction Edell, 1991; Abi-Dargham et al. 1998; Curran et al.

2004; Bhattacharyya et al. 2009; Jordaan et al. 2012).
The aetiology of schizophrenia is complex. The role in
Similar neurobiological response is the leading theory
the development of the disorder of an extensive num-
of the psychotic symptoms in schizophrenia (Howes
ber of genetic and environmental factors has been
et al. 2012; Salavati et al. 2015). Combined with a
examined, but a clarication of the pathogenic
high prevalence of substance use among patients
mechanisms is still needed (van Os & Kapur, 2009;
with schizophrenia, this has raised the hypothesis of
van Os et al. 2010). Substance use has been suggested
whether this could be a cause of the disorder
as a potential risk factor (Moore et al. 2007). Research
(Koskinen et al. 2010; Toftdahl et al. 2015).
suggests that cannabis, amphetamine, cocaine, and
The hypothesis has been tested in several studies,
alcohol can cause transient psychosis and that a patho-
with cannabis as the dominant focus. The majority of
physiological explanation of the psychotic outcome of
the papers suggests a causal correlation (Moore et al.
some of the substances could be an increase in the stri-
2007; Kuepper et al. 2011; Callaghan et al. 2012;
atal dopamine level (Janowsky & Risch, 1979; Satel &
Jordaan & Emsley, 2014). In the most comprehensive
meta-analysis on the subject (Moore et al. 2007), the
authors found a 40% higher risk of developing psych-
* Address for correspondence: C. Hjorthj, Ph.D., Copenhagen
University Hospital, Mental Health Center Copenhagen,
osis comparing people who at some time in their lives
Kildegrdsvej 28, 2900 Hellerup, Denmark. had used cannabis with never users. The authors esti-
(Email: mated that if the association was truly causal, 14% of

Downloaded from https:/ IP address:, on 11 Jul 2017 at 01:55:46, subject to the Cambridge Core terms of use, available at
Association between substance abuse and risk of developing schizophrenia 1669

incident cases of psychosis could be prevented if can- Abuse Register (Andersen et al. 1999; Pedersen et al.
nabis was not used. This is a cause of concern because 2006; Kildemoes et al. 2011; Mors et al. 2011; Uggerby
of the increase in the incidence rate for cannabis use et al. 2013; Danish Health Data Authority, 2015a, b).
(EMCDDA, 2015). Substance abuse was dened as reported in Table 1.
However, there are several limitations in the existing The date of onset of the abuse was dened as the
literature. A large number of studies are based on date of rst contact leading to the diagnosis of the sub-
designs with high risk of selection bias, loss to stance abuse. Psychotic symptoms caused by substance
follow-up, and misclassication because of the difcul- use (ICD-10: 1x.5, 1x.7, and 1x.8) was not included in
ties in reducing the risk of a transient intoxication of a the denition of substance abuse, as patients with
substance being an important confounder. Lack of these diagnoses represent a group with a higher
information on poly-substance abuse could also have vulnerability of developing psychotic disorder, and
confounded the results (Koskinen et al. 2010). Only a including this patient group could overestimate the
few papers have examined the association between association.
users of other substances than cannabis and amphet-
amine and schizophrenia. This has resulted in a
Assessment of schizophrenia
retained uncertainty of the direction of causation
(Macleod et al. 2004; Moore et al. 2007). We aimed to The outcome of diagnosis of schizophrenia (ICD-8:
investigate whether several types of substance abuse 295.x, except 295.7 and ICD-10: F20.x) was extracted
could increase the risk of developing schizophrenia. from the Danish Psychiatric Central Research Register
By using the Danish registers we were able to address (including main and secondary diagnoses) (Mors
some of the limitations of prior studies. The registers et al. 2011). The register contains information on all
contain information on every citizen in Denmark, inpatients since 1969 and additionally on contacts
and with this information it is possible to examine with emergency rooms and outpatient clinics since
the hypothesis with the largest number of participants 1995. Because there are no private psychiatric hospitals
and types of substances to date with a minimum of in Denmark, the information regarding diagnosis
selection bias and low loss to follow-up. We were of schizophrenia contains almost complete data
able to compile data on a large number of registered (Uggerby et al. 2013). The date of schizophrenia onset
substance abusers, as we extracted information from was dened as the date of rst psychiatric contact
several of the registers. that led to the diagnosis of schizophrenia. To reduce
the effect of a misclassication of cohort members on
the basis of an intoxication effect, all individuals diag-
nosed with schizophrenia less than a year after they
Study population and the registers were registered with a diagnosis of abuse were clas-
sied as non-abuser.
Since 1968 the Danish Civil Registration System (CRS)
has collected information on the whole Danish popula-
tion in a database using the personal identication Statistical analyses
(CPR) number (Pedersen et al. 2006). The CPR number
The population was followed from birth until diagno-
is a unique number given to every individual born in
sis of schizophrenia, rst emigration from Denmark,
Denmark or with permanent residence in the country.
death, or 1 July 2013 (last extraction date from the reg-
The number is used in all the national registers, which
isters), whichever came rst.
enables an accurate linkage between the registers
The data were analysed in Stata/MP v. 13
(Tsuang et al. 2011). The study population was
(StataCorp., USA) using Cox regression with substance
extracted from the CRS and consisted of all people
use disorders as time-varying covariates (i.e. each indi-
born in Denmark between 1 January 1955 and 31
vidual entering the analysis as not having substance
December 1999.
use disorders, and then changing status to having the
substance use disorder at the rst date it occurred in
Assessment of substance abuse
the registers). The p value and 95% condence inter-
Information regarding substance abuse was extracted vals (CI) were based on the likelihood ratio test. The
from the following registers, containing unselected analyses were primarily adjusted for any other type
information on all treatment provided in Denmark: of substance abuse, as we aimed to determine the
Danish Psychiatric Central Research Register, the specic effect of a concurrent substance abuse on the
Danish National Patient Register, the Danish risk of developing schizophrenia. Secondary analyses
National Prescription Registry, the National Alcohol were additionally adjusted for calendar year (continu-
Treatment Register, and the National Substance ous, time-varying variable), gender, urbanicity (born in

Downloaded from https:/ IP address:, on 11 Jul 2017 at 01:55:46, subject to the Cambridge Core terms of use, available at
1670 S. M. Nielsen et al.

Table 1. Diagnostic classication of substance abuse

Type of substance ICD-8 Denition in registers that uses other

abuse codea ICD-10 codea ATC codeb classications beside ICD or ATC codesc

Alcohol 291, 303, F10 (except 10.5, 0.7, 0.8), E52, N07BB01, We dened substance abuse as using a
571.0 G31.2, G62.1, G72.1, K29.2, N07BB02, substance 426 times per week or when
K86.0, O35.4, Y57.3, Z50.2, Z71.4, N07BB03 registered as abuser of the substance
Amphetamined 304.6 F15 (except 15.5, 0.7, 0.8)
Cannabis 304.5 F12 (except 12.5, 0.7, 0.8)
Cocaine 304.4 F14 (except 14.5, 0.7, 0.8)
Hallucinogens 304.7 F16 (except 16.5, 0.7, 0.8)
Opioids 304.0, 304.1 F11 (except 11.5, 0.7, 0.8) N07BC01,
Sedativese 304.2. 304.3 F13 (except 13.5, 0.7, 0.8)
Otherf 304.8, 304.9 F18 (except 18.5, 0.7, 0.8) and
F19 (except 19.5, 0.7, 0.8)

We included both main and secondary diagnoses.

Data from the Danish Psychiatric Central Research Register and the Danish National Patient Register.
The Anatomical Therapeutic Chemical Classication System (ATC code). Data from the Danish National Prescription
The National Substance Abuse Register and the National Alcohol Treatment Register do not use ICD or ATC codes.
Including other central stimulating substances like MDMA.
Including barbiturates and other sedative, hypnotic and tranquillizing substances.
Other substances: Dened as the diagnosis of abuse of other, multiple or unknown psycho-stimulant. Abuses of multiple
substances are only diagnosed, when the different abuses are equally serious.

cities with >100 000 or <100 000 residents. The 1980 abuse group once 2 years had passed since the last regis-
information was used to dene the urbanicity variable tered treatment for that particular type of substance
for all individuals born before 1980, as place of birth abuse. People in the former abuse group could then
were only available from 1980. Individuals censored switch back to the current abuse group at the time the
before 1980 and individuals with no information on next, if any, treatment for substance abuse was initiated.
place of birth were classied as unknown status of The assumption of proportional hazards was tested
urbanicity), other non-schizophrenia, non-substance- graphically by plotting Schoenfeld residuals v. time,
abuse psychiatric diagnosis (ICD-8: 290315.9 except and we did not identify any meaningful deviations
291, 295, 303305, and ICD-10: F00-09, F21-99 except from this assumption.
F25, X60-84, Z55-69, Z7 except Z714, Z715, Z717,
Z721, Z722, Z77-Z79, Z91 except Z910), parents psy-
Ethical approval
chiatric history (ICD-8: 290315.9 except 303-305 and
ICD-10: F00-09, F20-99), parents history of substance The authors assert that all procedures contributing to
abuse (Table 1), parents country of birth (World this work comply with the ethical standards of the rele-
Bank classication: Denmark, High income, Other), vant national and institutional committees on human
and parents socioeconomic status (highest educational experimentation. The identity of the cohort members
level, ISCED classication system, 1997). was blinded to the investigators. The study was
We performed additional analysis testing for inter- approved by the Danish Data Protection Agency.
action of gender in the association as well as sensitivity Register-based studies do not require ethical approval
analyses on the population born after 1980 because infor- by Danish law.
mation on outpatients and emergency-room contacts
were added in 1995 and by that completing the register.
Finally, we tested whether current abuse differed from
former abuse in its association with schizophrenia. In Among the cohort (3 133 968 individuals) followed for
these analyses, patients were classied in the former 105 178 673 person-years at risk, a total of 21 305

Downloaded from https:/ IP address:, on 11 Jul 2017 at 01:55:46, subject to the Cambridge Core terms of use, available at
Association between substance abuse and risk of developing schizophrenia 1671

individuals developed schizophrenia, resulting in an (95% CI 2.533.00). For cannabis, current abuse (HR
overall incidence rate (IR) of 20.3/100000 person-years 13.17, 95% CI 12.3314.07) and former abuse (HR
at risk (Table 2). Substance abuse was diagnosed in 5.46, 95% CI 4.995.96) both predicted schizophrenia.
204 505 persons prior to a diagnosis of schizophrenia. The same was true for alcohol (current: HR 7.49, 95%
In this risk group 4627 of the schizophrenia cases CI 7.147.86; former: HR 2.81, 95% CI 2.652.97);
occurred (IR: 218.5/100000 person-years at risk). amphetamines (current: HR 2.24, 95% CI 1.952.57; for-
During follow-up, 416 569 individuals were censored mer: HR 1.28, 95% CI 1.091.49); sedatives (current:
owing to emigration from Denmark. A further 90 156 HR 2.73, 95% CI 2.403.11; former: HR 1.92, 1.66
individuals were censored owing to death. Men were 2.22); hallucinogens (current: HR 3.09, 95% CI 2.30
overrepresented in both the group diagnosed with sub- 4.15; former: HR 1.95, 95% CI 1.452.63); and other
stance abuse (66%) and the group diagnosed with substances (current: HR 9.95, 95% CI 9.1910.76; for-
schizophrenia (62%). However, we found no signi- mer: HR 3.98, 95% CI 3.604.40). For opioids, only cur-
cant difference between the genders in the risk of rent (HR 1.73, 95% CI 1.561.92) but not former (HR
developing schizophrenia subsequent to a substance 1.12, 95% CI 0.951.33) abuse predicted later schizo-
abuse [men: hazard ratio (HR) 6.21, 95% CI 5.95 phrenia. For cocaine, current abuse was associated
6.47; women: HR 5.74, 95% CI 5.396.11] (p value for with an increased risk of schizophrenia (HR 1.26,
the interaction of gender: 0.979). 95% CI 1.061.50), but former abuse was associated
The results are presented in Table 3. Substance abuse with a decreased risk of schizophrenia (HR 0.76, 95%
increased the risk of developing schizophrenia relative CI 0.610.95).
to not abusing (HR 8.83, 95% CI 8.549.13). Diagnosis KaplanMeier curves were performed to illustrate
of abuse of hallucinogens (HR 22.91, 95% CI 17.82 the cumulative proportion of incident diagnosis of
29.46) and cannabis (HR 22.67, 95% CI 21.4823.92) schizophrenia (Fig. 1). Fig. 2 illustrates the risk of
presented the highest association with developing developing schizophrenia over time since rst diagno-
schizophrenia in unadjusted analyses. After adjusting sis of abuse for the two substances presenting the high-
for other types of substance abuse, a signicant statis- est risk, alcohol, cannabis, and a pooled risk of all other
tical association remained between abuse of cannabis substances. The risk of being diagnosed with schizo-
(HR 7.52, 95% CI 7.008.07), alcohol (HR 4.51, 95% phrenia was strongest within a year after being diag-
CI 4.314.71), other substances (HR 2.62, 95% CI nosed with substance abuse. The risk decreased in
2.282.90), hallucinogens (HR 1.62, 95% CI 1.252.11), the subsequent years, but remained signicant even
and sedatives (HR 1.57, 95% CI 1.391.78) and devel- after 1015 years.
oping schizophrenia. Abuse of amphetamines or
opioids was only borderline signicant. Cocaine
abuse was no longer statistically signicant. A
co-abuse attenuated the results notably more than In this study, an association between almost any kind
adjusting for any other covariate (data not reported). of substance abuse and schizophrenia was found. As
Additional adjustment for calendar year, gender, urba- expected, the highest risk of developing schizophrenia
nicity, any psychiatric diagnosis prior to substance was found among cannabis abusers. Abuse of halluci-
abuse, parental psychiatric history, parental history of nogens, sedatives and other substances was also found
substance abuse, parental immigration to Denmark, to signicantly increase the risk of developing schizo-
and parental socioeconomic status did not greatly phrenia. Adjustments reduced the HR; however, the
affect the results further. Only the fully adjusted asso- association remained signicant for almost all types
ciation with cocaine abuse indicated a decreased risk of of substances. Interestingly, we found that alcohol
schizophrenia in the abuse group (HR 0.70, 95% CI abuse presented one of the highest risks, after all
0.580.83). This is in direct opposition to the adjustments were made (Jordaan & Emsley, 2014).
unadjusted result that suggested a highly increased We found in general stronger associations compared
risk of developing schizophrenia. with estimates reported in other studies (Curran et al.
Limiting the cohort to only including individuals 2004; Moore et al. 2007; Jordaan & Emsley, 2014).
born after 1980 resulted in an overall risk of HR 3.95 Cannabis and alcohol abuse increased the risk by ve
(95% CI 3.684.23). and three times, respectively, compared with 40% for
Generally, both current and former abuse was asso- cannabis and 0.412.36% for alcohol as previously
ciated with schizophrenia, with the association being reported (Curran et al. 2004; Moore et al. 2007;
stronger for current abuse, even in the fully adjusted Jordaan & Emsley, 2014). However, in our study we
model. Current abuse of any substance was strongly aimed to examine the risk of abusing substances in
associated with schizophrenia (HR 10.83, 95% CI contrast to previous papers that have examined the
10.5211.15), and former abuse had an HR of 2.76 outcomes of substance use on any level of severity.

Downloaded from https:/ IP address:, on 11 Jul 2017 at 01:55:46, subject to the Cambridge Core terms of use, available at
1672 S. M. Nielsen et al.

Table 2. Characteristics of the cohort

With substance abuse Without substance abuse

disorder (n = 204 505) disorder (n = 2 929 463)

Person years at risk . . . 8 612 933 96 565 740

N % N % p value

Male 133 925 65.5 1 484 915 50.7 <0.001
Female 70 580 34.5 1 444 548 49.3
Diagnosed with schizophrenia subsequent to a
substance abuse diagnosis
Yes 4627 2.3 16 678 0.6 <0.001
No 199 878 97.7 2 912 785 99.4
Age at onset of schizophrenia
416 6 0.1 595 3.6 <0.001
1725 1307 28.2 8167 49.0
2530 1019 22.0 3194 19.2
3035 817 17.7 2084 12.5
>35 1478 31.9 2638 15.8
Urbanicity at birth
Born in city with <100000 residents 134 952 66.0 1 949 654 66.6 <0.001
Born in city with >100000 residents 69 392 33.9 876 827 29.9
Unknown 161 0.1 102 982 3.5
Any psychiatric diagnosis other than diagnosis of
schizophrenia and substance abuse
Yes 17 338 8.5 237 265 8.1 <0.001
No 187 167 91.5 2 692 198 91.9
Mother diagnosed with psychiatric disorder other
than substance abuse
Yes 36 387 17.8 312 011 10.7 <0.001
No 168 118 82.2 2 617 452 89.3
Father diagnosed with psychiatric disorder other
than substance abuse
Yes 25 456 12.4 228 508 7.8 <0.001
No 179 049 87.6 2 700 955 92.2
Mother diagnosed with substance abuse
Yes 28 970 14.2 183 477 6.3 <0.001
No 175 535 85.8 2 745 986 93.7
Father diagnosed with substance abuse
Yes 41 708 20.4 308 995 10.5 <0.001
No 162 797 79.6 2 620 468 89.5
Mothers country of origin
Denmark 190 336 93.1 2 624 625 89.6 <0.001
High income country 5386 2.6 96 791 3.3
Other level income country 2681 1.3 85 753 2.9
Unknown 6102 3.0 122 294 4.2
Fathers country of origin
Denmark 185 348 90.6 2 589 931 88.4 <0.001
High income country 4263 2.1 81 031 2.8
Other level income country 3620 1.8 97 060 3.3
Unknown 11 274 5.5 161 441 5.5
Mothers educational level
Primary education 309 0.2 10 394 0.4 <0.001
Lower secondary education 105 772 51.7 1 051 868 35.9
Upper secondary education 58 087 28.4 992 613 33.9
Post-secondary education 7 0.003 294 0.01

Downloaded from https:/ IP address:, on 11 Jul 2017 at 01:55:46, subject to the Cambridge Core terms of use, available at
Association between substance abuse and risk of developing schizophrenia 1673

Table 2 (cont.)

With substance abuse Without substance abuse

disorder (n = 204 505) disorder (n = 2 929 463)

Person years at risk . . . 8 612 933 96 565 740

N % N % p value

First stage of tertiary education 26 282 12.9 651 261 22.2

Second stage of tertiary education advanced 113 0.1 5174 0.2
research qualication
Unknown 13 935 6.8 217 859 7.4
Fathers educational level
Primary education 181 0.1 5519 0.2 <0.001
Lower secondary education 76 993 37.6 812 687 27.7
Upper secondary education 74 813 36.6 1 179 755 40.3
Post-secondary education 38 0.02 819 0.03
First stage of tertiary education 23 544 11.5 581 286 19.8
Second stage of tertiary education advanced 195 0.1 10 000 0.3
research qualication
Unknown 28 741 14.1 339 397 11.6

The p value is the overall p value for the group comparison. It is assessed using the 2 test.

This could suggest that using substances on a more risk of schizophrenia, with the association being stron-
substantial scale is more strongly associated with ger for current abuse. This may reect a dose-response
schizophrenia than less severe use. Unfortunately, the relationship. However, caution is required when inter-
structure of the data used in the present study did preting these results, since patients with ongoing
not allow us to examine the hypothesis of a abuse, but who are no longer treatment-seeking,
dose-response relationship because of a lack of suf- would also be categorized in the former-abuse
cient information on the amount of intake. However, group, as this was dened according to type of
previous papers on cannabis use have found statistical treatment.
evidence in favour of the hypothesis (Moore et al. The tendency of a more intensive and regular use of
2007). One other study investigated substance abuse substances reported with men, as well as mens
as a risk factor, and found in general higher associa- increased risk of developing schizophrenia in general,
tions compared with the present study (Callaghan could have resulted in higher risk compared with
et al. 2012). However, a lack of adjustment for other women (Pedersen et al. 2014; EMCDDA, 2015).
types of substances could have overestimated these However, we observed no signicant difference in
results. The attenuation of the associations between the risk of developing schizophrenia subsequent to a
abuse and schizophrenia in adjusted analyses was substance abuse between the genders.
largely caused by the effect of a poly-drug abuse. The lower HR found in the sensitivity analysis of the
Conversely, the adjustment could also cause an over- 1980 population could be explained by the markedly
adjustment, illustrated by the nding of a negative lower number of cases (5857 cases) as well as a
association with cocaine, as the crude results highly young cohort causing a limitation of information for
indicate the reverse association. The group of other the Cox regression to predict future incidences of
substances unfortunately contains a large proportion schizophrenia. Conversely, the result could also be
of unknown types of substances, as the diagnosis of constituted as a more accurate measure of the associ-
substance dependence of other types of substances ation as it contains complete psychiatric information.
(ICD-8: 304.8, ICD-10: F19) as well as unspecied The strengths of the study are the prospective design
type of substance (ICD-8: 304.9) was included in this based on the population-based nationwide registers in
group, diminishing the possibility of analysing these Denmark. This ensures a large study population and a
results. high number of cases in which all exposures were
Finally, the protective effect of cocaine was only evi- recorded independently of the outcome which mini-
dent in those with former abuse. Generally, both cur- mized possible selection or recall bias effects. The
rent and former abuse of other substances increased study design entails a low loss to follow-up as the

Downloaded from https:/ IP address:, on 11 Jul 2017 at 01:55:46, subject to the Cambridge Core terms of use, available at
1674 S. M. Nielsen et al.

Table 3. Risk of Schizophrenia for all substance abuse disorders

No. of persons No. of new Unadjusted Adjustment 2a

Type of at risk during cases during
substance abuse follow-up follow-up HR 95% CI HR 95% CI

No abuse 2 929 463 16 678 1 Ref. 1 Ref.

Any abuse 204 505 4627 8.83 8.549.13 6.04 5.846.26
Total 3 133 968 21 305

Type of No. of persons No. of new Unadjusted Adjustment 1b Adjustment 2c

substance Age at onset at risk during cases during
abuse of abuse follow-up follow-up HR 95% CI HR 95% CI HR 95% CI

No abuse 2 973 632 18 080 1 Ref. 1 Ref. 1 Ref.
Abuse 160 336 3225 6.84 6.587.10 4.51 4.314.71 3.38 3.243.53
No abuse 3 127 535 20 989 1 Ref. 1 Ref. 1 Ref.
Abuse 6433 316 15.86 14.1917.73 1.22 1.081.39 1.24 1.091.41
No abuse 3 111 105 19 857 1 Ref. 1 Ref. 1 Ref.
Abuse 22 863 1448 22.73 21.5523.99 7.52 7.008.07 5.20 4.865.57
No abuse 3 129 422 21 158 1 Ref. 1 Ref. 1 Ref.
Abuse 4546 147 13.05 11.0915.35 0.84 0.711.01 0.70 0.580.83
No abuse 3 133 208 21 244 1 Ref. 1 Ref. 1 Ref.
Abuse 760 61 22.91 17.8229.46 1.62 1.252.11 1.86 1.432.41
No abuse 3 133 698 20 701 1 Ref. 1 Ref. 1 Ref.
Abuse 20 270 604 10.71 9.8711.63 1.15 1.041.28 1.20 1.081.32
No abuse 3 126 116 20 935 1 Ref. 1 Ref. 1 Ref.
Abuse 7852 370 18.70 16.8620.74 1.57 1.391.78 1.68 1.491.90
No abuse 3 126 197 20 761 1 Ref. 1 Ref. 1 Ref.
Abuse 7771 544 16.13 14.8117.58 2.62 2.282.90 2.85 2.583.15

HR, Hazard ratio; CI, condence interval.

Adjusted for calendar year, gender, urbanity, any psychiatric diagnosis prior to substance abuse, parents psychiatric
history, parents history of substance abuse, parents immigration to Denmark and parents socioeconomic status.
Adjusted for any other type of substance abuse.
Adjusted for any other type of substance abuse, calendar year, gender, urbanity, any psychiatric diagnosis prior to
substance abuse, parents psychiatric history, parents history of substance abuse, parents immigration to Denmark and
parents socioeconomic status.

Danish registers contain almost complete information disorders due to use of substances (F1x.5, 1x.7, 1x.8)
on the diagnosis of schizophrenia (Mors et al. 2011; in the denition of substance abuse. This was done
Uggerby et al. 2013). in the interpretation that this group consists of indivi-
Extracting information from several registers duals more prone to develop psychotic disorders com-
increased the number of registered abusers (data not pared with the population. Overall, this is likely to
reported), however, an underestimation of the diag- cause results to be underestimates of the true
nosis of substance abuse must be anticipated. association.
Furthermore, the non-abusing groups will also include The study is representative of the Danish popula-
people who use the substances below the threshold of tion, because all residents born in Denmark were
abuse. We did not include the diagnoses of psychotic included (Pedersen et al. 2006). All substances

Downloaded from https:/ IP address:, on 11 Jul 2017 at 01:55:46, subject to the Cambridge Core terms of use, available at
Association between substance abuse and risk of developing schizophrenia 1675

Fig. 1. Cumulative incidence of schizophrenia over time according to substance use disorders.

Fig. 2. Hazard ratios for schizophrenia according to time since incident diagnosis of alcohol, cannabis, or substance use

investigated in the present study, except for alcohol, The rule of classifying anyone diagnosed with
are illegal in Denmark, as they are in most other coun- schizophrenia less than a year after a substance abuse
tries. The prevalence of substance use in Denmark is as non-abuser, used to diminish the effect of a possible
comparable to most other Western countries, which misclassication due to an intoxication effect, is conser-
is why we expect that our results would also be gener- vative. However, the rule is likely to additionally have
alizable to other populations outside of Denmark. modied the effect of a misclassication caused by

Downloaded from https:/ IP address:, on 11 Jul 2017 at 01:55:46, subject to the Cambridge Core terms of use, available at
1676 S. M. Nielsen et al.

abuse being registered in the databases from the date removal of those who emigrate. Consequently, while
of rst psychiatric contact and not on the date of substance use was associated with emigration (e.g.
onset of the abuse, as well as a detection bias caused those with cannabis use disorder were at signicantly
by a hospital contact for schizophrenia possibly entail- lower risk of emigration), use of Cox regression still
ing a subsequent risk of being diagnosed with a sub- yield the least biased results, especially since there is
stance abuse, or vice versa. This is illustrated in little reason to believe that those who were thus cen-
Fig. 2 by the markedly higher risk of being diagnosed sored would have markedy different risks of develop-
with schizophrenia the rst year after being diagnosed ing schizophrenia compared to those who remained in
with abuse compared with the later years. the study.
Both schizophrenia and substance abuse could be The consumption of substances is an extensive prob-
present in an individual long before entry into treat- lem throughout the world and a current debate on
ment and diagnosis, which suggest that timing of legalizing cannabis in many countries has made unco-
onsets as well as the temporal order of onset is un- vering the risk of abusing substances an important area
certain. We have made several attempts to minimize of investigation (EMCDDA, 2015). The seriousness of
an effect of a prodromal period and by that, self- the legalization debate was addressed by reports indi-
medication, for example by the conservative rule previ- cating an association between the perceived risk and
ous described. Additionally, we adjusted the results for the incidence of substance use in US and European
any psychiatric diagnosis prior to any substance abuse. populations (EMCDDA, 2011; Okaneku et al. 2015).
However, a risk of bias due to self-medication is likely In conclusion, the ndings of the study illustrate
to have resulted, and might explain a part of the robust associations between a wide variety of sub-
found associations. Completely eliminating the bias is stance abuse and an increased risk of developing
extremely difcult. To do so would require compre- schizophrenia later in life. Addressing the issue of a
hensive information on the psychological development possible impact of abuse of other types of substances
from birth on any participant of the study. The sparse on the association as well as the signicant associations
data from the registers on this area is a study found even after 1015 years brings new information
limitation. to the area. A causal relationship cannot yet be
We have adjusted for parents psychiatric history; declared. However, due to ethical difculties in inves-
however, it is not possible to rule out the possibility tigating this hypothesis using higher quality studies
that variation in genes as well as other factors asso- such as RCTs, using the Danish registers could be
ciated with substance abuse or schizophrenia (e.g. vari- one of the most suitable ways to study the association
ation in potency and tobacco) could have had an between substance abuse and schizophrenia.
important role in the found associations (Myles et al.
2012; Power et al. 2015). Unfortunately, the Danish reg-
isters do not provide information on these parameters. Acknowledgements
Use of nicotine has been linked to development of
This work was supported by a grant from the
schizophrenia, and could therefore be an important
Lundbeck Foundation to S.M.N. The sponsors had no
confounder both in respect to alcohol and cannabis
inuence on the protocol, analysis, interpretation,
(Gurillo et al. 2015). There are some indications, how-
drafting of the manuscript, decision to publish, or
ever, that the association between nicotine and psy-
any other areas other than funding of the authors
chosis can be explained by use of cannabis, rather
than the other way around (Menezes et al. 2016).
Furthermore, the ndings of the present study are sub-
stantiated by several factors. Fig. 2 illustrates that even
Declaration of Interest
after 1015 years subsequent to a diagnosis of abuse,
the risk of being diagnosed with schizophrenia is still None.
signicantly higher compared with non-abusers. In
addition, the biological evidence on the subject associ-
ating psychotic symptoms with the substances sup- References
ports the hypothesis (Janowsky & Risch, 1979; Satel
Abi-Dargham A, Gil R, Krystal J, Baldwin RM, Seibyl JP,
& Edell, 1991; Curran et al. 2004).
Bowers M, Van Dyck CH, Charney DS, Innis RB, Laruelle
We censored individuals upon emigration from M (1998). Increased striatal dopamine transmission in
Denmark, which allows us to keep them in the ana- schizophrenia: conrmation in a second cohort. American
lyses up until the time of censoring, which is a strength Journal of Psychiatry 155, 761767.
compared to approaches such as logistic regression at a Andersen TF, Madsen M, Jrgensen J, Mellemkjoer L,
given time, which would usually require complete Olsen JH (1999). The Danish National Hospital Register. A

Downloaded from https:/ IP address:, on 11 Jul 2017 at 01:55:46, subject to the Cambridge Core terms of use, available at
Association between substance abuse and risk of developing schizophrenia 1677

valuable source of data for modern health sciences. Danish Macleod J, Oakes R, Copello A, Crome I, Egger M, Hickman M,
Medical Bulletin 46, 263268. Oppenkowski T, Stokes-Lampard H, Davey Smith G (2004).
Bhattacharyya S, Fusar-Poli P, Borgwardt S, Martin-Santos Psychological and social sequelae of cannabis and other illicit
R, Nosarti C, OCarroll C, Allen P, Seal ML, Fletcher PC, drug use by young people: a systematic review of longitudinal,
Crippa JA, Giampietro V, Mechelli A, Atakan Z, McGuire general population studies. Lancet 363, 15791588.
P (2009). Modulation of mediotemporal and ventrostriatal Menezes P, Busatto G, McGuire P, Murray R, Scazufca M
function in humans by 9-tetrahydrocannabinol. Archives of (2016). Tobacco smoking and risk of rst episode psychosis:
General Psychiatry 66, 442451. a case-control study in Sao Paulo, Brazil. Schizophrenia
Callaghan RC, Cunningham JK, Allebeck P, Arenovich T, (Supplement), T126. Abstracts from the 5th Biennial SIRS
Sajeev G, Remington G, Boileau I, Kish SJ (2012). Conference.
Methamphetamine use and schizophrenia: a population- Moore THM, Zammit S, Lingford-Hughes A, Barnes TRE,
based cohort study in California. American Journal of Jones PB, Burke M, Lewis G (2007). Cannabis use and risk
Psychiatry 169, 389396. of psychotic or affective mental health outcomes: a
Curran C, Byrappa N, McBride A (2004). Stimulant psychosis: systematic review. Lancet 319328.
systematic review. British Journal of Psychiatry 185, 196204. Mors O, Perto GP, Mortensen PB (2011). The Danish
Danish Health Data Authority (2015a). The National Substance Psychiatric Central Research Register. Scandinavian Journal
Abuse Register. ( of Public Health 39, 5457.
registre-og-services/om-de-nationale-sundhedsregistre/ Myles N, Newall HD, Curtis J, Nielssen O, Shiers D, Large
sygedomme-laegemidler-og-behandlinger/stofmisbrugere- M (2012). Tobacco use before, at, and after rst-episode
i-behandling). Accessed 25 January 2017. psychosis: a systematic meta-analysis. Journal of Clinical
Danish Health Data Authority (2015b). The National Alcohol Psychiatry 73, 468475.
Treatment Register. ( Okaneku J, Vearrier D, McKeever RG, LaSala GS,
registre-og-services/om-de-nationale-sundhedsregistre/ Greenberg MI (2015). Change in perceived risk associated
sygedomme-laegemidler-og-behandlinger/ with marijuana use in the United States from 2002 to 2012.
alkoholbehandlingsregisteret). Accessed 25 January 2017. Clinical Toxicology 53, 151155.
EMCDDA (2015). European Drug Report 2015. Trends and van Os J, Kapur S (2009). Schizophrenia. Lancet 374, 635645.
developments. European Monitoring Centre for Drugs and van Os J, Kenis G, Rutten BPF (2010). The environment and
Drug Addiction. schizophrenia. Nature 468, 203212.
EMCDDA (2011). Annual report 2011. The state of the drugs Pedersen CB, Gtzsche H, Mller JO, Mortensen PB (2006).
problem in Europe. European Monitoring Centre for Drugs The Danish Civil Registration System. A cohort of eight
and Drug Addiction, p. 112. million persons. Danish Medical Bulletin 53, 441449.
Gurillo P, Jauhar S, Murray RM, MacCabe JH (2015). Does Pedersen CB, Mors O, Bertelsen A, Waltoft BL, Agerbo E,
tobacco use cause psychosis? Systematic review and meta- McGrath JJ, Mortensen PB, Eaton WW (2014). A
analysis. Lancet Psychiatry 2, 718725. comprehensive nationwide study of the incidence rate and
Howes OD, Kambeitz J, Kim E, Stahl D, Slifstein M, lifetime risk for treated mental disorders. JAMA Psychiatry
Abi-Dargham A, Kapur S (2012). The nature of dopamine 71, 573581.
dysfunction in schizophrenia and what this means for Power Ra, Verweij KJH, Zuhair M, Grant W, Henders AK,
treatment. Archives of General Psychiatry 69, 776786. Heath AC, Madden PaF, Medland SE, Wray NR, Martin
Janowsky DS, Risch C (1979). Amphetamine psychosis and NG (2015). Genetic predisposition to schizophrenia
psychotic symptoms. Psychopharmacology 65, 7377. associated with increased use of cannabis. Molecular
Jordaan GP, Emsley R (2014). Alcohol-induced psychotic Psychiatry 19, 12011204.
disorder: a review. Metabolic Brain Disease 29, 231243. Salavati B, Rajji TK, Price R, Sun Y, Graff-Guerrero A,
Jordaan GP, Warwick JM, Nel DG, Hewlett R, Emsley R Daskalakis ZJ (2015). Imaging-based neurochemistry in
(2012). Alcohol-induced psychotic disorder: brain perfusion Schizophrenia: a systematic review and implications for
and psychopathology before and after antipsychotic Dysfunctional long-term potentiation. Schizophrenia Bulletin
treatment. Metabolic Brain Disease 27, 6777. 41, 4456.
Kildemoes HW, Srensen HT, Hallas J (2011). The Danish Satel SL, Edell WS (1991). Cocaine-induced paranoia and
National Prescription Registry. Center for Healthy Ageing, psychosis proneness. American Journal of Psychiatry 148,
Department of Public Health, University of Copenhagen, 17081711.
Denmark. Scandinavian Journal of Public Health 39, 3841. Toftdahl NG, Nordentoft M, Hjorthj C (2015). Prevalence
Koskinen J, Lhnen J, Koponen H, Isohanni M, Miettunen of substance use disorders in psychiatric patients: a
J, Lohonen J, Koponen H, Isohanni M, Miettunen J (2010). nationwide Danish population-based study. Social
Rate of cannabis use disorders in clinical samples of Psychiatry and Psychiatric Epidemiology 51, 129140.
patients with schizophrenia: a meta-analysis. Schizophrenia Tsuang MT, Tohen M, Jones PB (eds) (2011). Textbook of
Bulletin 36, 11151130. Psychiatric Epidemiology. John Wiley & Sons, Ltd:
Kuepper R, van Os J, Lieb R, Wittchen H-U, Her M, Chichester, UK.
Henquet C (2011). Continued cannabis use and risk of Uggerby P, stergaard SD, Rge R, Correll CU, Nielsen J
incidence and persistence of psychotic symptoms: 10 year (2013). The validity of the schizophrenia diagnosis in the
follow-up cohort study. British Medical Journal (Clinical Danish Psychiatric Central Research Register is good.
Research Edition) 342, d738. Danish Medical Journal 60, 14.

Downloaded from https:/ IP address:, on 11 Jul 2017 at 01:55:46, subject to the Cambridge Core terms of use, available at