10.1192/bjp.171.4.355Access the most recent version at doi:
1997 171: 355-359 The British Journal of Psychiatry
CD Rossau and PB Mortensen
Risk factors for suicide in patients with schizophrenia: nested case-
Article cited in:
To obtain reprints or permission to reproduce material from this paper, please write
to this article at
You can respond
top right corner of the article or
Receive free email alerts when new articles cite this article - sign up in the box at the
The Royal College of PsychiatristsPublished by
on July 14, 2011 bjp.rcpsych.org
go to: The British Journal of PsychiatryTo subscribe to
1987, and diagnosed at least once within this
period as having schizophrenia. Diagnoses
were according to ICDâ€”8(World Health
Organization,1967) and the cohort
identical to the one used in the mortality
study by Mortensen & Juel (1993).
A nested caseâ€”controlstudy (Clayton &
Hills, 1993) was conducted
The applicationof this design of study to
psychiatric research has been described
Mortensen(1988). The cases were all the
508 patientsfrom the cohort
138 females) who had committed suicide in
the period between their first schizophrenia
of death was taken
Death Certificates at the Danish Institute of
Mortensen& Juel (1993). Prom the same
were identifiedfor each suicide case. The
sampling strategy was that of time-matched
Thomas, 1982). The matching criteria were:
(a) diagnosedas having
before the date of death of the suicide case;
and (b) alive at the time of death of the case.
package Epilog (Epicenter Software, 1993).
The odds ratio estimates resulting from this
sampling and analysis can be interpreted
estimates of the incidence rate ratio (IRR)
between the exposed category or categories
and the reference level of exposure (Flanders
&Louv, 1986). In other words, in all
variables the reference category
tion has an IRR of 1.
Data on admissions to general hospitals
were obtained from the Danish National In
patient Register (Sundhedsstyrelsen,
nationwidebasis in 1977;
data on suicide attempts and admissions to
generalhospitalswere only available
cases and the corresponding
suicide occurred on 1 January 1977 or later.
Data were analysed initially including
one explanatory variable at a time, adjusting
either relevant on the basis of expectations
only variables(such as age and gender,
which were consistently found to be impor
tant in other studies),
1 April 1970 to 31 December
on this cohort.
from the Register of
controls if the
and age; and after
In the final model,
Denmark between 970and 987and
diagnosedfor the firsttime ashaving
The purpose ofthe study was to identify
schizophrenia,particularly factors relating
for suicide among patientswith
patients,the 508 who hadcommitted
suicidewere individuallymatchedto I0
controlsfrom the samecohort, anddata
highduringthe first 5daysafterdischarge,
multipleadmissions duringthe previous
previousadmissions to generalhospitalsfor
physicaldisorders.After adjustingfor these
factors, no effect wasfoundfor age.There
was some evidence ofan excessof suicides
duringtemporary leavefrom the psychia
Conclusions Thefindings suggest that
the first period after discharge, when
It is well known that patients with schizo
phreniaare at greater
(Simpson, 1988). A lifetime risk of up to
10% has been suggested by Miles (1977), but
even this figure may be under-estimated
(Mortensen,1995). It has been a recurring
concern that the suicide risk among patients
studies have aimed at identifying risk factors
for suicide in patients with schizophrenia, as
reviewed by Allebeck (1989) and Caldwell &
Gottesman (1990). However, up till now, the
risk factors identified for suicide in schizo
phrenia do not have sufficient sensitivity and
specificity to enable us to identify high-risk
groups of individuals so that they could be
targeted with preventive intervention.
course of schizophrenia,
vention strategies could be aimed at most
priority for suicide research is to study in
detail the relationship
services, but also found that this necessitates
the use of a substantial sample of psychiatric
suicides and a caseâ€”controldesign.
The aim of this study was to identify risk
factors for suicide in patients with schizo
phrenia, with particular emphasis on factors
relatedto the hospitalisation
risk of committing
might be to identify
during which inter
between suicide and
The study was based on the nationwide
Danish Psychiatric Case Register which has
been computerisedsince 1969, and includes
The study population
persons (5658 males and 3498 females) who
were admittedfor the first time to any
consisted of all 9156
Riskfactors for suicidein patients with
schizophrenia: nested caseâ€”control study
C. D. ROSSAUand P.B. MORTENSEN
were statisticallysignificant at a 5% level,
year of first admission, which we wanted to
include in order to see whether the increase
in suiciderisk which
reported in the same population
& Juel, 1993) could be explainedby other
included. The onlyexceptionwasthe
we had previously
was included as a combination
sion) as reference category. These variables
relevant but theexact formofthe
exposure significantly reduced the fit of the
model, even if every single variable
cating a specific level of the exposure was
of all of the
in order to make theresults easier
Both the univariate
model are summarised in Table 1. The effect
of gender was as expected, with males being
at greaterrisk than females, whereas
effect of age (i.e. higher risk among
variables (in particular,
Suicide risk was highest during the first
six months after the first admission with
This remainedtrue after adjusting
time of diagnosisof schizophrenia,
those diagnosed as having schizophrenia
their first admission had a lower suicide risk
than those diagnosed later.
The number of psychiatric admissions
during the last year before the suicide was
found to have considerable effect. When the
hospitalduringthe last year
suicide were taken as a reference, the risk
was higher: a relative risk of approximately
3.5 in patients with 1â€”2 admissions during
the last year rose to a relative risk of almost
11 for those with more than eight admis
sions, implying that the so-called revolving
suicide risk considerably.
The current admission/discharge status
and the duration of this status also had a
strong effect. When the risk during the first S
resultsand the final
after this period.
Table IEffectsofvariousfactorsonsuicide riskinin-patients with schizophrenia
GenderFemale v. male
Present age> 45 years v. < 45 years
Later admissions V.first
Admittedto generalEver v. never
No. ofprevious suicide0
No. of psychiatric
Time of schizophrenic
Duration of illness
days after admission was taken as reference
(that is, the risk during what will often be an
acute psychotic episode), the risk declined
during the period of admission, even though
the risk was not significantly different from
that of the first S days. More surprising,
however, was that during the first S days
after discharge the risk of suicide was more
than twice (IRR=2.14) that in the first S days
of admission, and even during the first 28
days after discharge,the risk was still the
Time after admission
Time after discharge
DepressiondiagnosisEver v. never
I. All univariate resultsare adjustedfor gender andageusingconditional logistic regression.Univariate resultsforgender and
ageareadjusted for ageor gender.respectively.
@2=S@Â°83.d.f.=26, P< I0@
same as during the initial phase after admis
sion. After the first 28 days of discharge, the
risk declined to the same level as the risk
more than 28 days after being admitted.
should be noted that the only period in which
the suicide risk was significantly
from that in the period just after admission
whole set of categorical variables character
ising risk during the hospitalisation/discharge
course was highly significant.
were psychiatric diagnoses
to ICDâ€”8.Patients who had at
some stagebeen diagnosed
depressive disorder were at increased risk,
which would appear
findings that the risk of suicide in schizo
phrenia is increased
reflectedby a non-significant
suicide risk in the schizoaffective
(ICDâ€”8: 295.79). No other effect of schizo
able was not included in the final model.
A surprising finding was that patients
who had been admitted to a general hospital
for physical disorders were at increased risk
of later suicide. No particular
specific physical disorders could be identi
fied among those who committed
Our previousfinding (Mortensen
in the univaniate analysis, but disappeared
after adjusting for the other risk factors in
the final model.
This might suggest that the increase is
factors in the final model.
Marital status, and whether living in the
town or the country, yielded IRR estimates
close to 1 (i.e. no effect) and large P values
in the univaniateanalysis
considered in the multivaniate analyses.
A diagnosisof drug or alcohol
appearedto increase risk in the univaniate
reversed in the multivaniate analysis, where
matelyhalf the risk of patients
voluntarilyin the univaniate
this effect disappearedin the multivariate
Because of the relatively strong effects of
the various factors related to the course of
hospitalisation,we also tried to compare the
risk during admission
temporaryleave from hospital with the risk
after discharge from hospital. Unfortunately,
it wasnot possibleto getexactdatesfor the
beginning and end of temporary
only for admissions
temporary leave could not be calculated and
the risks could not be directly compared.
However, cases of suicide could be distrib
in the presence
by the otherrisk
but this was
and the risk during
timeat risk during
Methodofsuicide andadmission statusat
discharge, and 54 (10.6%) occurred during
admission to a psychiatric hospital, whereas
temporaryleave. Since we would
periods of temporary leave to be considerably
shorter than the admission periods, and since
we know that, after their first admission,
patients with schizophrenia
would indicate that the risk of suicide is
from psychiatric departments,
an exact IRR cannot be calculated.
The distributionof methods of suicide is
shown in Table 2. As in the general Danish
population, there were among the patients
relatively more self-poisonings and relatively
fewer deaths by hanging
than among men, but generally there was a
higherproportion of suicides
means (which includes jumping from high
shooting, or fromcutting
phrenia thanin the general
women in this respect is much greater.
The differencesin method,
admission status at the time of suicide, are
shown in Table 3. Hanging, drowning, and
other violent methods of suicide were more
â€˜¿?other' methods dominated
ary leave,and poisoning
Unfortunately,it is not possible to specify
the group â€˜¿?other'in any more detail from the
informationto hand, but a number of such
cases will have jumped in front of trains.
of the suicides occurred during
time as in-patients
The main finding of this study was the close
relationship between factors related to the
course of hospitalisation
while taking other relevant risk factors such
as age and genderinto account.
have other studies used the exact number of
personâ€”years at risk in the calculations,
comparing suicide risk at different
during follow-up. This means that in most
other studies â€”¿?for example, comparing risk
during admissionand discharge â€”¿?
that patients spend more time outside than
inside a psychiatrichospital
been taken into account. Also, the fact that
follow-up to death was almost complete in
our study would presumably
and more accurate risk estimates for suicide
during discharge than in other studies.
However, irrespective of these methodo
logical differences, our findings are in accor
dance with a number
Temocheet a! (1964) found an increased
risk of suicide in both current and former
found the risk to be highest during the first
six months after discharge. Copas & Robin
(1982),studyingonly suicides during
patient episodes, found that the suicide risk
was considerably increased during the first
week of admission,and declined
thereafter. In a study
schizophrenia, Roy (1982) found that 30%
discharge, and Appleby (1992), in his review
of suicideriskin psychiatric
concludes that maximum risk occurs at two
points:(a) at the beginning
phase; and (b) after discharge from in-patient
stay, especially during the first three months.
and the duration of
the risk of suicide.
this has not
specifically for schizophrenia
will not have
lead to higher
of the acute
of illness is in accord
ance with the results of a number of studies,
our own previous findings (Mortensen
Juel, 1993),even after
other risk factors in the final model. A
number of other studies, including Sletten Ct
a! (1972) and Yarden
found that the number of psychiatric admis
sions during the last year before the suicide
has a considerable effect, but it is interesting
that we found this effect too, since our study
was based on a population
sion patients and thus was not selected with
respect to chronicity, and also, the duration
of the current admission or discharge period
(1974), and replicates
change from longer, less frequent
hospital, to shorter periods in hospital with
increases the suicide risk in patients
schizophrenia. In Denmark
1970s and 1980s there has been a reduction
in the number of available psychiatric beds,
as well as a change towards
more frequent admissions (Munk-JÃ¸rgensen
& Mortensen, 1992; Mortensen & Eaton,
changes, at least in part, may explain our
previous finding of an increasing suicide risk
in patientswith schizophrenia
first year after discharge (Mortensen & Juel,
A relatively large proportion of suicides
occurred during temporary
exact durationof these periods could not
be determined,it was
calculate exact relative risks, but the finding
is in accordancewith other reports of a
temporary leave (e.g. Niskanen et a!, 1974).
Together withthe fact that
suggests that the period of transition
the hospital environment
(most often the patient's
to preventive measures during this time.
account, as well
as the total
may have a
leave of absence
of suicides during
to another setting
own home) is a
should be paid
Age and gender
Age was not found to be of significant
importance,after the inclusion of other risk
factors. This should not, however, be inter
leavefrom the psychiatric department.
Patientsareat highersuicideriskimmediatelyafterdischarge andduringtemporary
U Suicide preventioncould be targetedat these vulnerable periods.
Iâ€˜¿?Revolving door'admissionpatternsare associated withincreasedsuicide risk.
U A limited
treatment, socialconditions andpsychopathology.
number of variables available,in particularlack of informationon
U The study is observational,
asto possibleeffects of intervention is,at best, circumstantial.
not a controlledinterventionstudy, so that the evidence
out-patients, and inclusion isbasedon clinical diagnoses,rather than research
diagnostic criteria for schizophrenia.
The study does not include patientswho have exclusively been treatedas
CHARLOTTE DAHL ROSSAU MD; PREBEN BO MORTENSEN, DrMedSc, Institute for Basic Psychiatric
Research, Aarhus, Denmark
Dr Rossau, Institute
for Basic Psychiatric
in Aarhus, University
of Aarhus, Denmark
(First received 21October 1996, final revision S March 1997,accepted 10 March 1997)
studies. When included alone in the present
study, it did show a significant effect, the
younger patients being at greater risk. The
including in the model the duration
number of previous admissions. This is not
surprising since young patients have a young
age at onset, and this is also closely related
as a failure to replicateprevious
(Mortensen & Eaton, 1994). The finding
that the risk is higher in male patients
consistent with the literature
on suicide in
Other risk factors
previous suicide attempts are in accordance
with those of most other studies (Caldwell
& Gottesman,1990); and the influenceof a
illness confirms the findings of Roy (1982)
(among others)that depressive
and hopelessness are important
for suicide in schizophrenia.
findingsregardingthe influence of
An unexpected finding was the influence
of previous hospital admissions for physical
Sundqvist-Stensman (1988) have previously
suggested that â€œ¿?suicide
a chronic physical disease might be the final
decidingfactorin a person
psychiatric illnessâ€•,and Bolin et a! (1968),
in their surveyof suicideamongpatientson
home leave from mental hospitals,
importantrisk factor for suicide. However,
furtherstudies are needed to confirm
importance of poor
assessing suicide risk in schizophrenia.
among persons with
be anhealth may
after admission and discharge, together with
the high proportionof suicide cases during
temporary leave of absence from the hospital,
suggest thatthese may be periods
preventive intervention might be particularly
appropriate.However, our study only gives
information as to where and when suicide
risk of suicide immediately
risk is highest: it does not give any informa
tion as to why the patients commit suicide, or
ment, relation to relatives, or social support
in general, of the suicidal patients.
research is needed to specify this in detail,
and even given this information,
tion of preventiveintervention
difficult, because of the low baseline rate of
suicides among patients with schizophrenia
(Mortensen,1995). However, we find that
since most risk factors identified in this study,
together with other
literature(such as depressive
or perceived loss or rejection, limited external
desirable irrespective of the possible effects
on suicide risk, the most promising avenue
for future research and intervention
probably be to focus on these general factors
confirmed in this study. Also, we suggest
focused on measures which can be applied to
most or all psychiatric patients, thus reducing
the prevalence of risk factors
among psychiatric patients in general, rather
than focusing on small high-risk groups of
suicidal psychiatric patients.
The study was financed by Psykiatrisk Forskningsfond,
Fondentil forskning af sindslidelser,
1967 and theTheodore and Vada Stanley Foundation.
BntishJournal of Psychiatry, 161,
j. a Achte, K. (I!74) Suicidesina
Polcorny,A. D. (1960)Characteristicsof 44 patientswho
subsequently committedsuicide.Archives of GeneralPsychiatry,
Roy, A. (1982) Suicideinchronic schizophrenia. BritishJournal
Saugstad,L F.a Oie@d,
psychiatrichospitalsin Norway 1950â€”1974.
0. (1979)Mortality in
Simpson, j. C. (1988)Mortality studiesinschizophrenia.
InHandbookof Schizophrenia. Nosology, Epidemiology and
Genetics of Schizophrenia (edsM.T.Tsuang& j.C. Simpson),
pp. 245â€”274. Amsterdam:Elsevier.
Sletten, I., Brown, M. a E@nson,R. (1972)Suicidein
mentalhospitalpatients.Diseases ofthe NervousSystem, 33.
Sundhedsstyrelsen [The National Boardof Health)(1993)
Evaluering af Landspatientregisteret 1990(Evaluationofthe
National PatientRegister1990). Copenhagen: The National
Boardof Health,Sygehgusstatistik II. 57.
Stensman, R. a Sundqvlst.Stensman,(A B. (1988)Physical
Scandinavian Journalof Social Medicine,16,149â€”153.
Temoche, A., Pug,T. F.a MacNelson, B. (1964)Suicide
ratesamongcurrent andformer mentalinstitutionpatients.
Journalof NervousandMentalDisease, 138,124â€”l30.
â€”¿? aj@&,K. (1993) Mortalityandcausesofdeath in first
Vlrkkunen, N. (1974)Suicidesinschizophrenia and
paranoidpsychoses. ActaPsychiatrica Sandinavica
Supplementum, 250. 1â€”305.
@rld Health Organizadon (1967)Manualofthe
International Classification of Diseases (ICDâ€”8).
schizophrenics. Comprehensive Psychiatry, IS,325â€”333.
Allebeck, P.(1989)Schizophrenia: a life-shorteningdisease.
Appieby, L. (1992)Suicideinpsychiatricpatients:riskand
prevention.BritishJournalof Psychiatry, 161. 749â€”758.
Bolin, R. K.,Wright, R.E.,Wiikinson, M. N., et at (1968)
Survey of suicideamong patients on home leave from a
mentalhospital.Psychiatric Quarterly. 42. 81â€”89.
Caidweil, C. B. & Gottesman, I. I. (1990)Schizophrenics kill
themselvestoo: areviewof nskfactorsfor suicide.
Clayton, D. a Hills, H. (1993)Statistical Modelsin
Epidemiology. Oxford: Oxford University Press.
Copes.j. B. a Robin, A. (1982)Suicidein psychiatricin
patients.BritishJournalof Psychiatry. 141,503â€”SII.
Epicenter Software (1993)EpilogPlus. Version 3.Procedures.
Pasadena, CA: EpicenterSoftware.
Flanders,W. D. a Louv,W. C. (1986) The exposure odds
ratio in nestedcaseâ€”control studieswith competingrisks.
American Journalof Epidemiology. 124.584â€”692.
Greenland, S. a
rare diseaseassumptionincaseâ€”control studies.American
Journalof Epidemiology, 116,547â€”553.
@ D.C. (1982) On the need for the
Hosmer, D.W.a Lemeshow,S. (1989)AppliedLogistic
Regression. New York:Wiley.
Miles, C. P. (1977) Conditions
review.Journol of NervousandMentalDisease, 164,231â€”246.
predisposing to suicide: a
Mortensen. P.B. (1988)Ambidirectionalstudiesâ€”¿?
extensionof longitudinalstudiesin psychiatry.Psychiatric
Developments, 6. 73â€”181.
â€”¿?(1995) Suicide amongschizophrenic
schizophrenia. Psychological Medicine,24.223â€”232.
Eaton,W.W.(1994)Predictorsforreadmission risk in
andother aspectsofthe epidemiologyofschizophreniain
P. a Mortensen,R B. (1992) Incidence