The British Journal of Psychiatry (2010)
196, 372–376. doi: 10.1192/bjp.bp.109.066050
Effect of early intervention on 5-year outcome
in non-affective psychosis{
Rafael Gafoor, Dorothea Nitsch, Paul McCrone, Tom K. J. Craig, Philippa A. Garety, Paddy Power
and Philip McGuire
Background
Early specialised care may improve short-term outcome in
first-episode non-affective psychosis, but it is unclear if these
benefits endure.
Aims
To assess the long-term effect of early intervention in
psychosis.
Method
Individuals with first-episode psychosis were randomised to
specialised care or care as usual (trial number:
ISRCTN73679874). Outcome after 5 years was assessed by
case-note review.
Specialist services aim to reduce hospitalisation in patients with
psychosis by initiating treatment as soon as possible after the first
episode.1 In the OPUS study,2 participants with first-episode
psychosis randomised to specialist services had shorter hospital
admissions at 1 year after presentation than those assigned to
care as usual; however, these benefits did not endure at 2-year
follow-up. The Lambeth Early Onset (LEO) trial3 found that
individuals treated by a specialist service had lower hospital bed
use at 18 months than those receiving care as usual. Whether
the beneficial effects of specialist services endure in the longer
term is unclear. We addressed this issue by following up
participants in the LEO trial 3.5 to 5 years after inception. The
trial was registered with the International Standard Randomised
Controlled Trial Number Register and was assigned the trial
number ISRCTN73679874.
Method
Initial study – randomised controlled trial
In 2002, 144 people presenting to psychiatric services in Lambeth,
South London, with a first or second episode of non-affective
psychosis were randomly allocated to either ‘specialist services’
or ‘care as usual by community mental health teams (CMHTs)’
at inception. Inception was defined as the point of first treatment
for psychosis. Inclusion required an age of 16–40 years, and a
presentation of a non-affective psychosis (schizophrenia, schizotypal, and delusional disorders, F20–29 in the ICD–10).4
Individuals with an ‘organic’ psychosis or a primary alcohol
or drug addiction were excluded. Participants were randomised
by permuted random blocks of between two and six. The
randomisation and concealment group allocation was carried out
by sealed envelope method by the trial statistician. Follow-up
interviews were conducted at 18 months (while the participant
was still receiving care from the original clinical teams). Data on
{
See pp. 377–382, this issue.
372
Results
There were no significant differences in the admission rate
(coefficient 0.096, 95% CI 70.550 to 0.742, P = 0.770) or the
mean number of bed days (coefficient 6.344, 95% CI 746 to
58.7, P = 0.810).
Conclusions
These findings that specialist intervention did not markedly
improved outcome at 5 years accord with those from a
larger OPUS study. The sample size of this study was small
and these results should be generalised with caution. More
research is needed.
Declaration of interest
None.
readmissions were obtained from centralised computer case
records and from interview with psychiatric care coordinators.
Follow-up information on clinical status at 18 months was
collected from 131 (91%) people.
The specialist team (LEO) was a new service comprising 10
mental health professionals who delivered specialised interventions including low dose atypical antipsychotic regimens,
cognitive–behavioural therapy based on manualised protocols,5–7
family counselling and vocational strategies based on established
protocols.8,9 Participants in the control group received standard
care from established adult general CMHTs in Lambeth. These
teams had received no formal training in specialist interventions
but had access to guidelines on the approach.8,10
Although participants and clinicians were not masked to
treatment arm, it is unlikely that participants had contact with
other participants in the other arm. Allocation remained
undisclosed until completion of the ratings. However, assessors
were able to guess the arm of treatment to which 60% of
participants had been allocated (95% CI 52 to 63, k = 0.20).
Participants were analysed by intention-to-treat analysis (ITT)
in the group to which they were initially randomised. Regression
analyses were adjusted for variables that were unbalanced between
the two groups at baseline (ethnicity, past episode and gender).
Results from the initial study
A previous study3 had found that individuals receiving specialist
treatment had fewer admissions in the intervening 18-month
review period than those assigned to care as usual: (0.4 admissions
v. 0.8 (b = 0.36, 95% CI 0.04 to 0.66, P = 0.030)), but were not less
likely to have ever been admitted (OR = 0.53, 95% CI 0.26 to 1.12,
P = 0.095) or to have had shorter admissions (b = 19.4, 95% CI
710.6 to 48.6, P = 0.197). Participants treated by the specialist
team had better social and vocational functioning (mean 6.9
months in employment for the specialist group v. mean of 4.2
months, P = 0.008), quality of life (mean score of 59.2 for specialist
services v. mean score of 53.3, P = 0.010) and medication
Early intervention and outcomes in non-affective psychosis
adherence (mean adherence score of 5.4 for specialist services v.
mean score of 4.5, P = 0.036) at 18 months).
Initial LEO cohort (2000–2001)
In 2005, the above participants were retrospectively assessed
during an 18-month period (3.5 to 5 years after inception). The
primary (null) hypothesis was that there would be no difference
in the odds of having ever been admitted in the 18 months prior
to this second assessment. The secondary (null) hypotheses were
that the relative admission rates and the length of bed occupancy
during this period would be the same for both groups.
Ethical permission for this study was given by the Institute of
Psychiatry at the Maudsley (Kings College London, University of
London). Ethical permission for the analysis of the data was given
by the ethical committee of the London School of Hygiene and
Tropical Medicine, University of London. Clinical ethical
permission for participants to be traced, contacted and interviewed was given by the South London and Maudsley NHS
Mental Health Trust (SlaM).
We assumed that the relative reduction in the rate of ever
having been admitted would decrease by 10% in the next 5 years
as the effects of specialist services waned; there are no comparative
admission data currently available at 5 years for individuals who
were exposed to specialist services as this is the first follow-up
study of such a randomised cohort. Using data from clinical audit,
the rate of ever having been admitted for individuals in ‘care as
usual’ CMHTs is approximately 65% over 18 months at 5-year
follow-up. We assumed a relative reduction in the rate of ever
having been admitted of 23% between the two groups and a rate
of 46% of ever having been admitted (for people who had been
randomised to the specialist group) over the 18-month period
leading up to 5 years post first admission. At a power of 80%
and an alpha of 0.05, we determined that 234 people would be
required in total to find a difference between the two groups if
one existed. If all 144 individuals had been traced, the study would
have a power of 56% of discovering a difference between the
groups if such a difference existed at an alpha of 0.05. This study
was underpowered but it is no longer possible to increase the
sample size given that it would be unethical to randomise
individuals again to care as usual. This follow-up study (although
imperfect) provides the best evidence currently available in the
clinical circumstances.
An extensive sequential tracing algorithm was used to find
participants: individuals were traced using electronic patient
records held by the local National Health Service (NHS) trust,
the NHS Strategic Tracing Service (NSTS), land registry records,
general practitioner (GP) records, the UK register of deaths or
via the last known relative. We were not able to access prison
records or Home Office immigration records for deported
individuals. Migrants and prisoners may have had worse
prognoses and more admissions than other people; differential
loss to follow-up for this subgroup may have resulted in an
‘emigration bias’.11–15 Follow-up was not masked to intervention
arm as individuals often disclosed their treatment on interview
and treatment group was apparent on review of the notes. Poor
concealment was also an issue in the original LEO trial (as
discussed above). To reduce the chance of information bias, dates
of recorded admissions and the number of documented
admissions were used as the primary sources for data collection.
Information was verified by reference to written material or by
reference to the team treating the individual.
All traced participants were successfully contacted (including
those living overseas), and all agreed to provide follow-up
information. Individuals were given information about the study
5-year follow-up study
Follow-up study
Potentially eligible
participants (n = 319)
Fig. 1
Participants randomised
(n = 144)
Excluded
(n = 175)
Allocated to specialist
services (n = 71)
Allocated to standard
care (n = 73)
No reported deaths
1 reported death
Participants
uncontactable
(n = 26)
Participants
uncontactable
(n = 18)
Participants
followed up
(n = 45)
63% of original cohort
Participants
followed up
(n = 54)
74% of original cohort
Participant flow.
Reasons for exclusion: not resident in Lambeth, too old or too young (n = 38); did not
meet diagnostic criteria (n = 90); already engaged with services (n = 35); lost before
confirmed (n = 12).
on the telephone and this was followed up by written materials
where requested. Participants were interviewed on the telephone
or in person (whichever method they preferred). Figure 1 shows
that of the original 144 participants who were randomised, 99
(69%) were traced and contacted. In comparison, Nordentoft et
al in their 5-year follow-up study of service interventions in
Denmark were able to contact 301 participants (55%) out of an
incipient cohort of 547.16
To remove the potential influence of differential access to
home treatment teams on hospital admission rates (as it is often
provided as an alternative to admission), all home treatment team
events were counted as equivalent to an admission. Where the
home treatment team care was provided in isolation, the case
was recorded as an admission with the length of home treatment
counting as the length of hospital admission. In some cases,
individuals initially receiving home treatment teams were then
admitted: in this case, home treatment and hospital admission
were counted as a single admission event. The total time of
admission was the time in hospital plus the time with home
treatment teams. The same algorithm was used for individuals
who were discharged from the ward directly to home treatment
teams.
Diagnoses were assigned at a median time of 5.3 years and
were made using a notes-based diagnostic tool, the operational
criteria OPCRIT checklist17 for psychotic and affective illness,
which provided diagnoses according to the ICD–10 system.4
Analyses were performed using STATA release 9 on Windows
Vista. Linear regression models were employed to assess the
relationship between initial randomisation group and outcome
variables. There were two potential sources of confounding: failure
of randomisation to produce balanced groups in the original LEO
trial and differential contactability between treatment arms.
373
Gafoor et al
(Details of the strategy for analysis for confounding and the results
of these analyses are contained in the online supplement.) A
sensitivity analysis was performed to assess the robustness of the
findings when participants who could not be traced were assigned
extreme values.
Table 1
cohort
Comparison of followed-up participants and original
Results
Figure 1 shows the participant flow for both the original trial in
2000 and during the 18-month period of follow-up (i.e. 3.5 to 5
years after inception). One person died due to accidental causes.
Of those who had been randomised to care as usual, 15 (28%)
v. 15 (33%) of those who had been allocated to the specialist team
had been discharged to their GP (Pearson w2 = 0.359, P = 0.549,
d.f. = 1). The length of time for which individuals were with their
GP after discharge from services was highly skewed. Participants in
specialist services who had been discharged to their GP spent a
median of 0 days (IQR = 0–826) with their GP v. a median of
0 days (IQR = 0–264) for those randomised to care as usual. There
was no significant difference in the length of time spent with their
GP for either group (Mann–Whitney U = 1147.5, Z = 70.571,
P = 0.568). Participants treated by the specialist service were
followed up by psychiatric services for a median of 1778 days
(IQR = 1152–2030) after first randomisation v. a median of 1887
days (IQR = 1640–2034) for those randomised to care as usual.
There was no statistically significant difference in the length of
time spent in psychiatric services for either group (U = 1077.5,
Z = 70.973, P = 0.330). Individuals randomised to specialist
services spent a median of 861 days in the specialist service
(IQR = 720–1027). In total, 99 participants (70%) were
successfully traced from the original cohort. By comparison,
Nordentoft et al16 in their 5-year follow-up were able to contact
only 301 participants (55%) of an incipient cohort of 547,
underscoring the difficulties of follow-up in this cohort.
The followed-up participants were representative of the
original LEO sample (Table 1). Contactable participants were
not significantly different in baseline clinical variables from
those who were traced (see online supplement). Analysis for
confounding (see online supplement for details) identified that
‘ever having been discharged from psychiatric services to a GP’
was a confounder in the relationship between whether of not a
participant could be contacted and the main outcome of ever
having been admitted. We also adjusted for imbalances in baseline
variables: gender, previous psychotic episodes and ethnic group.
Table 2 shows the cross-sectional differences between the two
groups during the 18-month follow-up (i.e. from 3.5 to 5 years
post-inception) at baseline. There was no difference in the chances
of any admission, the number of admissions or the number of bed
days used during the follow-up period. We noted that there was an
increase in the number of admissions during the follow-up period
relative to the initial review period. This increase was attributable
to an increased admission rate for each individual relative to that
during the initial follow-up period. Thus, during the second
follow-up period, 21% of people were not admitted, 49% were
admitted once, 20% were admitted twice and 8% were admitted
three or more times. The reason for the increased number of
admissions during this period is unclear. It may reflect changes
in service structures over the period of the study, and/or a change
in the pattern of service use or need as psychotic illnesses become
more chronic.
Regression modelling was employed to estimate effect size
of interventions and to allow for adjustment with potential
confounding factors.
Although the data were not normally distributed, because the
data from the original LEO trial used a parametric analysis, we
374
Age at baseline, years:
median (IQR)
Original cohort
baseline
characteristics
(n = 144)
Baseline
characteristics
of participants
traced 5 years later
(n = 99)a
25 (21.0–25.0)
25.0 (20–25)
Male, n (%)
93 (65)
60 (60)
Ethnicity, n (%)
White
Black British
Black Caribbean
Black African
Mixed
Other
45
16
22
41
12
8
28
15
15
25
9
7
(31)
(11)
(15)
(28)
(8)
(5)
(28)
(15)
(15)
(25)
(9)
(7)
Single, n (%)
100 (69)
69 (69)
First episode,b n (%)
113 (78)
73 (73)
Living situation, n (%)
Family
Alone
Otherc
77 (54)
41 (28)
24 (17)
65 (65)
26 (26)
8 (8)
Employment, n (%)
Full time
Part time
Unemployed
17 (12)
9 (6)
90 (63)
8 (8)
7 (7)
55 (55)
Migrant, n (%)
56 (39)
38 (38)
a. Percentages based on traceable population of 99.
b. Inclusion criteria for the incipient cohort allowed participants to have had a
previous psychotic episode, providing it had been untreated and the individual had
subsequently disengaged from clinical services.
c. Living with friends or living in a hostel.
replicated this analysis so that the results during the initial and
subsequent18-month periods could be more directly compared.
Data on number and length of admission during the 18month period leading to 5 years were highly skewed. The median
number of admissions was identical in both groups (median 0,
IQR = 0–1). The mean number of bed days was 42.25 days
(s.d. = 112.8, median 0, IQR = 0–31.0) and 51.41 days (s.d. = 125,
median 0, IQR = 0–38.0) in the ‘specialist services’ and ‘care as
usual’ groups respectively.
Table 3 shows the relationship between outcome and randomisation group for crude and adjusted regression modelling. There
were no significant differences between the two groups in terms of
any of the outcome measures before and after adjusting for
confounders.
Sensitivity analysis
In total, 26 participants in the specialist services arm and 18
people in the care as usual arm could not be traced at 5-year
follow-up. We performed sensitivity analyses on the data from
these missing individuals to examine the robustness of the finding
from the initial logistic regression model that there was no group
difference in ever having been admitted (see online supplement).
We calculated that the finding of no observed differences in the
odds of being admitted between the two groups would only be
reversed if at least 80% of those who were missing from the
specialised care group and had been admitted as well as if 20%
of missing participants from the care as usual group had also ever
been admitted (see online supplement). This is an unlikely
scenario and uncontactable individuals in the specialised care
arm would have to have been twice as likely to have ever been
Early intervention and outcomes in non-affective psychosis
Table 2
Primary outcome measures for participants receiving specialised care or standard care for early psychosis
18-month follow-up starting from time of entry
Specialised care group Standard care group
Any readmissions, n (%)
Admissions, mean (s.d.)
23 (33)
b
0.4 (0.7)
Bed-days in follow-up, mean (s.d.)
35.5 (78.9)
33 (51)
18-month follow-up starting 3.5 years after first entry
P
Specialised care group Standard care group
50.001
0.8 (1.0)
0.010
54.9 (93.6)
0.42
a
18 (33)
17 (39)
0.160a
1.65 (0.86)
1.83 (0.92)
0.56c
45.25 (112.8)
51.41 (125)
0.88c
c
c
P
a. Pearson w2-test.
b. Some participants were initially admitted at time of presentation mainly because of risk considerations. To remove bias when comparing admissions with individuals who were
initially incepted into the study in the community, the initial admission at baseline was not counted.
c. Student’s t-test.
Table 3
Regression models for main outcomes
Predictor variable
Ever admitted
Crude regression
coefficient or ORa
c
Number of admissionsd
Length of stay,e days
95% CI
Adjusted regression
coefficient or ORb
P
95% CI
P
1.26
0.549 to 2.89
0.586
1.24
0.460 to 3.34
0.670
0.040
70.463 to 0.544
0.875
0.006
70.542 to 0.555
0.981
0.800
2.15
76.16
754.3 to 42.0
750.4 to 54.7
0.935
a. Odds ratio of outcome for participants in specialist services v. those in care as usual.
b. Adjusted for any previous psychotic episode, ethnicity and gender.
c. Modelled using logistic regression.
d. Modelled using Poisson regression.
e. Modelled using linear regression with robust standard errors.
admitted as those from the specialised care arm who were traced.
We discounted the scenarios where all or none of those people
who could not have been traced had been admitted in either
arm as being unlikely.
We examined the stability of the initial diagnosis of ‘nonaffective psychosis’ that had been made 5 years previously when
the participants took part in the original LEO study. Data were
extracted from case notes and used in the OPCRIT algorithm to
produce ICD–10 diagnoses. Table 4 shows the ICD–10 diagnostic
category assigned by OPCRIT at the time of follow-up. The
median time from entry into the study to diagnosis was 64.2
months (5.3 years) (IQR = 58.3–64.2 months). Of those who could
be accorded a diagnosis at follow-up, more than 85% of them
continued to be diagnosed with a psychotic illness.
Discussion
Limitations
The power of the study to detect statistically significant differences
between the two groups was low. However, this is not surprising
given that the benefits of specialised services did not persist during
the period of 3.5 to 5 years post-inception and relative differences
between the two groups became smaller. Only a restricted range of
clinical outcomes were assessed and potential benefits in other
outcome domains may persist.
The study was conducted in an urban population with a
relatively high incidence of psychosis, and high levels of poverty,
unemployment and illicit drug use. The risk of developing
psychosis and of relapsing after a first episode in this population
may not be representative of the UK overall. Given these
limitations, the results of this study should be generalised to other
populations with caution.
Clinical implications
Although there is good evidence that specialised intervention in
first-episode psychosis improves outcomes in the first 1–2 years,3
the extent to which these benefits persist in the longer term
remains unclear, particularly after the specialised intervention is
withdrawn.
Table 4
Diagnoses at follow-up
Diagnosis (ICD–10)
Schizophrenia, schizotypal, and delusional disorders
n (%)
57 (58)
Other
5 (5)
Bipolar disorder
3 (3)
Depressive illness
2 (2)
Psychosis (not otherwise specified)
1 (1)
No diagnosis accorded
Sufficient information not available to give diagnosis
1 (1)
30 (30)
Prodromal therapies
Prodromal interventions (before the onset of the first psychotic
episode) may improve outcomes but at least two systematic
reviews18,19 conclude that the combined results of prodromal
interventions are equivocal. More research on the efficacy of
prodromal services is needed.
Future research
Aside from limited statistical power, the absence of a difference in
outcome between the two groups at 5-year follow-up may reflect
the withdrawal of the specialised intervention after 18 months
(when there was a significant group difference);3 further investigation of this issue will require trials involving longer durations of
specialised treatment. At present, specialised care for psychosis is
usually provided for the initial 1–2 years of illness.
Rafael Gafoor, BA, MSc, PhD, MRCPsych, Institute of Psychiatry, Kings Health
Partners, London; Dorothea Nitsch, MD, MSc, London School of Hygiene and
Tropical Medicine, London; Paul McCrone, PhD, Tom K. J. Craig, PhD, FRCPsych,
Philippa A. Garety, MA, MPhil, MA(Ed), PhD, FBPsS, Paddy Power, MD, MRCPsych,
Philip McGuire, PhD, FRCPsych, Institute of Psychiatry, King’s Health Partners,
London, UK
Correspondence: Philip McGuire, Institute of Psychiatry, King’s College London,
De Crespigny Park, London SE5 8AF, UK. Email: [email protected]
First received 11 Mar 2009, final revision 26 Aug 2009, accepted 16 Sep 2009
375
Gafoor et al
9 Edwards J, McGorry PD. Implementing Early Intervention in Psychosis:
A Guide to Establishing Early Psychosis Services. Martin Dunitz, 2002.
Funding
This study was funded by the Medical Research Council (UK) via a Clinical Research
Training Fellowship to the lead author.
10 Aitchison K, Meehan K, Murray RM. First Episode Psychosis. Martin Dunitz,
1999.
11 Brinded PM, Simpson AI, Laidlaw TM, Fairley N, Malcolm F. Prevalence
of psychiatric disorders in New Zealand prisons: a national study.
Aust N Z J Psychiatry 2001; 35: 166–73.
References
1
Birchwood M, Todd P, Jackson C. Early intervention in psychosis. The critical
period hypothesis. Br J Psychiatry 1998; 172 (suppl 33): s53–9.
12 Brugha T, Singleton N, Meltzer H, Bebbington P, Farrell M, Jenkins R, et al.
Psychosis in the community and in prisons: a report from the British National
Survey of psychiatric morbidity. Am J Psychiatry 2005; 162: 774–80.
2
Petersen L, Jeppesen P, Thorup A, Abel MB, Ohlenschlaeger J, Christensen
TO, et al. A randomised multicentre trial of integrated versus standard
treatment for patients with a first episode of psychotic illness. BMJ 2005;
331: 602.
13 Butler T, Allnutt S, Cain D, Owens D, Muller C. Mental disorder in the New
South Wales prisoner population. Aust N Z J Psychiatry 2005; 39: 407–13.
3
Craig TK, Garety P, Power P, Rahaman N, Colbert S, Fornells-Ambrojo M, et
al. The Lambeth Early Onset (LEO) team: randomised controlled trial of the
effectiveness of specialised care for early psychosis. BMJ 2004; 329: 1067.
4
World Health Organization. The ICD–10 Classification of Mental and
Behavioural Disorders: Clinical Descriptions and Diagnostic Guidelines. WHO,
1992.
5
Jackson H, McGorry P, Edwards J, Hulbert C, Henry L, Francey S, et al.
Cognitively-oriented psychotherapy for early psychosis (COPE). Preliminary
results. Br J Psychiatry 1998; 172 (suppl 33): 93–100.
6
Jackson H, McGorry P, Henry L, Edwards J, Hulbert C, Harrigan S, et al.
Cognitively oriented psychotherapy for early psychosis (COPE): a 1-year
follow-up. Br J Clin Psychol 2001; 40: 57–70.
7
Jolley S, Garety P, Craig T, Dunn G, White J, Aitken M. Cognitive therapy in
early psychosis: a pilot randomised controlled trial. Behav Cog Psychother
2003; 31: 473–8.
18 de Koning MB, Bloemen OJ, van Amelsvoort TA, Becker HE, Nieman DH, van
der GM, et al. Early intervention in patients at ultra high risk of psychosis:
benefits and risks. Acta Psychiatr Scand 2009; 119: 426–42.
8
Early Psychosis Prevention and Intervention Centre. EPPIC Early Psychosis
Training Pack. EPPIC, 1997.
19 Marshall M, Rathbone J. Early intervention for psychosis. Cochrane Database
Syst Rev 2006; 4: CD004718.
psychiatry
in pictures
14 Butler T, Andrews G, Allnutt S, Sakashita C, Smith NE, Basson J. Mental
disorders in Australian prisoners: a comparison with a community sample.
Aust N Z J Psychiatry 2006; 40: 272–6.
15 Coid J. How many psychiatric patients in prison? Br J Psychiatry 1984; 145:
78–86.
16 Bertelsen M, Jeppesen P, Petersen L, Thorup A, Ohlenschlaeger J, Le Quach
P, et al. Five-year follow-up of a randomized multicenter trial of intensive
early intervention vs standard treatment for patients with a first episode of
psychotic illness: the OPUS trial. Arch Gen Psychiatry 2008; 65: 762–71.
17 McGuffin P, Farmer A, Harvey I. A polydiagnostic application of operational
criteria in studies of psychotic illness. Development and reliability of the
OPCRIT system. Arch Gen Psychiatry 1991; 48: 764–70.
Ian Curtis
Geoff Dickens and Marco Picchioni
18 May marks the 30th anniversary of the suicide of Ian Curtis, lead
singer of Joy Division, who was just 23 when he hanged himself.
Curtis’s work comprises little more than two dozen songs recorded
over 3 years but it remains disproportionately influential. A lineage
can be traced from Joy Division through Siouxsie and the
Banshees, The Cure, U2, through to contemporary artists such
as Interpol and Editors.
Prescribed barbiturates for his poorly controlled epilepsy, Curtis
was hospitalised following an overdose 6 weeks before his death.
He self-discharged the next day to front the band at a chaotic gig.
Curtis’s last show at Birmingham University, 2 weeks before his
suicide, was featured posthumously on the Joy Division album Still;
the song Isolation is notable for the following lyrics: ‘I’m ashamed
of the things I’ve been put through, I’m ashamed of the person I
am’. Curtis’s band mates, however, felt his lyrics did not reflect
his state of mind. Curtis had already told his wife that his life’s
ambition to release one album and one single had been fulfilled,
and that he wanted to leave the band and ‘join a circus’.
Thirty years on, Curtis’s fame outshines that achieved in his brief
lifetime. Best known for the single Love will Tear Us Apart,
released a month after his death – the song features in Rolling
Stone’s top 500 songs of all time, with Curtis the subject of
two feature films, numerous books, documentaries and reissued
discs.
The British Journal of Psychiatry (2010)
196, 376. doi: 10.1192/bjp.196.5.376
Photo s Paul Slattery/Retna
376
The British Journal of Psychiatry (2010)
196, 372–376. doi: 10.1192/bjp.bp.109.066050
Online supplement
Details of analysis to identify confounders
Imbalances in initial Lambeth Early Onset (LEO) trial
Factors that were unbalanced at baseline (due to failure of
randomisation to produce balanced groups when using small
sample sizes) between the two randomised groups were entered
into the regression model. This adjustment was performed by
the first LEO study and we replicate this analysis to increase
comparability of the two studies. Both crude and adjusted ratios
are given as adjusting for variables that are known to be
unbalanced at baseline and could potentially introduce error into
the analyses. Randomisation potentially removes confounding for
both known and unknown confounders but may have been
ineffectual in the original LEO study because of insufficient size.
Adjusting for imbalances in known confounders could potentially
introduce confounding by creating imbalances in unknown
confounders.
Confounders in second trial
Potential confounders were chosen after review of the literature
identified variables that could potentially be related to both
outcome and chance of being traced for the LIFE study. As
contactability lies on the causal pathway between exposure and
outcome of interest, confounders of the relationship between
contactability and outcome are very likely to also act in the
relationship between randomisation group and outcomes. The
following confounders were identified: ethnicity;20 gender;21
age;22 duration of untreated psychosis;23 whether or not the
person was in a relationship at baseline;24 whether or not the
person was working at baseline;25 whether or not the person was
in education at baseline;26 migrant status.27
Although the list is not exhaustive, these factors had the
strongest relationship with outcomes of those considered.
Specialist psychiatrists were consulted about other factors they
considered important. The following factors are therefore also
considered: discharge to general practitioner; and total time with
psychiatric services.
Cannabis misuse at baseline was a potential confounder but
could not be assessed as this information had not been collected
in the original study. Duration of untreated psychosis is a
potential confounder and was recorded at baseline. However,
the recorded values were judged to have poor reliability and so
were not used in further analyses.
All potential confounders were entered sequentially into a
logistic regression model to ascertain strength of relationship with
contactability at follow-up. Any variables thus identified were
considered as potential confounders. There was insufficient
sample size to examine the potential role for interaction between
variables that were entered one at a time. Relationship with
outcome (where known for participants) was similarly assessed
by sequential entry of any variables identified in the first analysis
into a logistic regression model with ‘ever been admitted’ as the
primary outcome of interest.
Table DS1 shows the results of modelling for an association
between contactability and the potential confounding variables.
Details of statistical modelling for missing
data-sensitivity analysis
We were not able to contact 30% of participants. It is possible that
this cohort may have had a systematically better or worse outcome
than those who had been traced and thus the overall results
obtained may have been biased. We modelled for missing
data from uncontactable participants using sensitivity analyses
in which we considered the possibilities of this missing cohort
having different admission rates from those who had been
contactable. The underlying assumption is that the missing
individuals were ‘not missing at random’ (i.e. that missingness
was not as a result of completely random loss of data (which
would be termed ‘missing completely at random’). Examples of
mechanisms underlying missingness are demonstrated in
Appendix DS1.
For missing ‘not at random’ data we assume that participants
who were not followed up were less or more likely not to have ever
been admitted potentially because they had the best or worse outcomes after the end of the LEO study. Data were modelled using a
complete data-set that included both values of ‘missing’ as well as
predicted outcome values.
Sensitivity analysis was employed to assess the robustness of
the outcomes by randomly substituting possible outcome values
for missing data. A fixed proportion of data from uncontactable
participants was randomly chosen to be assigned the most severe
outcome (ever admitted to hospital), whereas the missing data
from the other participants was simultaneously assigned the best
outcome (i.e. ‘not admitted’). An alternative strategy would have
been to set a proportion of the missing values to ‘admitted’ while
simultaneously setting the other missing values to ‘missing’. This
analysis was not selected as the odds ratios thus produced for each
permutation are less directly comparable.
Initially, 20% of patients for whom outcome data were missing
in the care as usual arm were randomly selected and assigned the
worse outcome value (ever admitted in the follow-up period).
Participants were randomly selected by the computer programme
STATA on Windows Vista. The subsequent odds ratio and 95%
intervals of uncertainty of ever been admitted by randomised
group were then calculated. This procedure was carried out 100
times and the average odds ratio and the associated 95% intervals
of uncertainty were reported. Next, the percentage of missing data
for the care as usual arm was increased by 20% and the odds ratio
recalculated as before. This procedure was continued in intervals
of 20% until all of the care as usual participants for whom data
were missing had been assigned to the worst outcome. Then,
20% of participants for whom data were missing in the specialist
care arm were randomly assigned the worst outcome and resultant
mean odds ratios and associated 95% intervals of uncertainty were
calculated after 100 iterations of each of the possible six groups of
‘care as usual’ participants with missing data (i.e. those with 0%,
20%, 40%, 60%, 80% and 100% of missing data randomly
assigned to the worst outcome). The participants in the specialist
care arm were randomly reselected ten times and the resultant
odds ratios of ever being admitted calculated for each category
of participant in the care as usual arm with missing data for each
selection. The percentage of individuals in the specialist group
with missing data were increased by another 20% and the odds
ratios recalculated until 100% of the missing data in the specialist
group had been assigned the worst outcome.
In the second step, the selection process was reversed and 20%
of people in the specialist arm with missing data were randomly
selected to have the worst outcome. The procedure for the initial
sensitivity analysis was then repeated until all of the participants
with missing data (in both arms) were again eventually assigned
to the worst outcome.
In the third step, the results from the first and second
iterations were compared to see if they differed significantly. In
1
no case did the results from the two sets of analyses differ by more
than 1% and the results from the first set of analyses are presented.
The results are shown in Table DS2.
23 Bhugra D, Corridan B, Rudge S, Leff J, Mallett R. Social factors and first onset
schizophrenia among Asians and whites. Int J Soc Psychiatry 1999; 45:
162–70.
24 Harrow M, Westermeyer JF, Silverstein M, Strauss BS, Cohler BJ. Predictors
of outcome in schizophrenia: the process-reactive dimension. Schizophr Bull
1986; 12: 195–207.
Additional references
20 Shi L, Ascher-Svanum H, Zhu B, Faries D, Montgomery W, Marder SR.
Characteristics and use patterns of patients taking first-generation depot
antipsychotics or oral antipsychotics for schizophrenia. Psychiatr Serv 2007;
58: 482–8.
21 Hopper K, Wanderling J. Revisiting the developed versus developing country
distinction in course and outcome in schizophrenia: results from ISoS, the
WHO collaborative followup project. International Study of Schizophrenia.
Schizophr Bull 2000; 26: 835–46.
Table DS1
22 Hartmann E, Milofsky E, Vaillant G, Oldfield M, Falke R, Ducey C. Vulnerability
to schizophrenia. Prediction of adult schizophrenia using childhood
information. Arch Gen Psychiatry 1984; 41: 1050–6.
25 Ikebuchi E. Support of working life of persons with schizophrenia [in
Japanese]. Seishin Shinkeigaku Zasshi 2006; 108: 436–48.
26 Isohanni I, Jones PB, Järvelin MR, Nieminen P, Rantakallio P, Jokelainen J,
et al. Educational consequences of mental disorders treated in hospital.
A 31-year follow-up of the Northern Finland 1966 Birth Cohort. Psychol Med
2001; 31: 339–49.
27 Cooper B. Schizophrenia, social class and immigrant status: the
epidemiological evidence. Epidemiol Psichiatr Soc 2005; 14: 137–44.
Regression models to assess relationship of potential confounders with contactability
Variable
Contactable
Not contactable
OR (or regression coefficient)
of being contactable
1.196
a
95% CI
P
0.59–2.42
0.620
Ethnicity, n (%)
White
Black
Caribbean
Black African
Mixed/other
27
28
26
4
13
Male, n (%)
60 (61)
33 (72)
0.559
0.258–1.21
0.141
In a relationship, n (%)
19 (19)
22 (48)
3.343
1.16–9.66
0.026b
Working, n (%)
15 (15)
11 (23)
1.822
0.758–4.38
0.180
1.13c
0.549–2.32
0.743
0.003d
(27)
(29)
(27)
(4)
(13)
16
10
16
0
4
(32)
(22)
(35)
(0)
(9)
Employment, n (%)
Full time
Part time
Unemployed
8 (8)
7 (7)
81 (85)
9 (20)
2 (4)
35 (75)
Not migrant, n (%)
57 (58)
18 (39)
0.331
0.158–0.689
Discharge to GPe
0.048
0.005–0.404
0.005f
Log of total time with psychiatric servicese
4.05
1.20–13.7
0.024g
0.964
0.930–1.04
0.583h
Age at baseline, years: median (IRQ)
25.0 (20.0–31.3)
27.0 (22.0–30.0)
a. Recategorised as White or other. White is the reference category.
b. There was no significant association between ‘being in a relationship’ and the outcome of ‘ever been admitted’ (OR = 0.542, 95% CI 0.113–2.59, P = 0.443). Being in a relationship
was not further analysed as a potential confounder.
c. Odds of contactability for unemployed v. those who were employed.
d. There was no significant association between being a migrant and the outcome of ‘ever been admitted’ (OR = 1.12, 95% CI 0.449–2.82, P = 0.802). Being in a migrant was not
further analysed as a potential confounder.
e. Discharge to general practitioner (GP) and time in psychiatric services not measured at baseline.
f. Being discharged to a GP was also strongly related to the outcome ‘ever been admitted’ (OR = 0.124, 95% CI = 0.027–0.569, P = 0.007). This variable was included in future analyses
as a potential confounder.
g. The total length of time spent in psychiatric care was highly skewed. Time was therefore log adjusted. The log of time ranged from 4.40 to 7.70 (mean 7.33, s.d. = 0.485). The
relationship between contactability and the log of total time spent in psychiatric care is presented. Graphical analysis showed six outliers below the median. Statistical analysis of the
relationship with the outliers removed did not alter the odds ratio significantly. However, log of total time spent in psychiatry (entered as a continuous variable) was not significantly
associated with the outcome and so was not considered further as a potential confounder (OR = 1.74, 95% CI 0.534–6.01, P = 0.345).
h. There was a relatively small age span between the oldest and youngest participants. We therefore treated the potential relationship between odds of being contactable and age as
linear; it is unlikely that the odds of being contactable would change significantly over such a relatively small age range.
Appendix DS1
Mechanisms of missingness
Assumption of missingness
Mechanism of missingness
Missing completely at random (MCAR)
An example of a MCAR mechanism would be if a completely random process underlies the missingness of
data (i.e. data lost in transit).
If data are MCAR, then approximately equivalent results would be obtained by performing the analyses if
there had been no missing data although there would be some loss of precision. Thus, the analysis of only
those units with complete data gives valid inferences.
This is not a valid analysis for this data-set as the missing data are not due to MCAR mechanisms.
2
Missing at random (MAR)
This assumption would be valid if there were predictor of missingness. Statistical analysis of this data-set
identified no predictors of missingness and thus this assumption is not valid for this data-set.
No missing at random (NMAR)
This is the most suitable analysis if the assumptions for MCAR and MAR are not met and is used in the
analysis of this data-set.
Table DS2
Sensitivity analysis for ever been admitted (odds ratio (OR) of ever being admitted with 95% interval of uncertainty) a
Percentage of missing participants
in specialist care assigned
to ‘admitted’
Percentage of missing participants in care as usual assigned to ‘admitted’
0% (n = 0)
20% (n = 4)
40% (n = 7)
60% (n = 11)
80% (n = 14)
100% (n = 18)
0%
(n = 0)
1.26
0.549–2.89
1.63
0.733–3.62
2.07
0.956–4.50
2.44
1.14–5.23
2.89
1.36–6.10
3.26
1.56–6.84
20%
(n = 5)
1.03
0.460–2.31
1.33
0.615–2.89
1.70
0.802–3.59
2.00
0.959–4.17
2.36
1.14–4.87
2.67
1.30–5.45
40%
(n = 10)
0.872
0.396–1.92
1.12
0.530–2.40
1.44
0.691–2.98
1.69
0.827– 3.46
2.00
0.990–4.04
2.26
1.13–4.52
60%
(n = 16)
0.782
0.370–1.92
1.01
0.530–2.40
1.29
0.691–2.98
1.52
0.827–3.46
1.79
0.989–4.04
2.02
1.13–4.52
80%
(n = 22)
0.687
0.318–1.48
0.889
0.426–1.85
1.13
0.557–2.29
1.33
0.666–2.67
1.58
0.797–3.11
1.78
0.907–3.48
100%
(n = 26)
0.613
0.287–1.31
0.793
0.383–1.64
1.00
0.501–2.03
1.19
0.599–2.36
1.41
0.718–2.75
1.59
0.817–3.08
a. Grey shaded area identifies analyses that would have resulted in a change in the conclusion, i.e. that the efficacy of the intervention had been diluted. The white area identifies
sensitivity analyses that would have made no difference to the overall conclusions.
5
7
Contactable/
not
contactable
at follow-up
Sensitivity analysis
7
7
7
5
Outcomes
not
known
8
5
Outcomes
Outcomes
known
7
7
7
8
LEO
trial
cohort
End of initial
treatment.
Potential
discharge to
GP or to
community
team
5
Inadequate
randomisation leading
to differences in
groups at baseline
Fig. DS1
.
.
.
.
.
.
.
.
.
Discharge to GP
Time with psychiatric services
Ethnicity
Gender
Age
Duration of untreated psychosis
Relationship?
Working?
Education?
Schematic analysis for potential confounds.
3
Effect of early intervention on 5-year outcome in non-affective
psychosis
Rafael Gafoor, Dorothea Nitsch, Paul McCrone, Tom K. J. Craig, Philippa A. Garety, Paddy Power and
Philip McGuire
BJP 2010, 196:372-376.
Access the most recent version at DOI: 10.1192/bjp.bp.109.066050
Supplementary
Material
References
Reprints/
permissions
You can respond
to this article at
Downloaded
from
Supplementary material can be found at:
http://bjp.rcpsych.org/content/suppl/2010/05/04/196.5.372.DC1
This article cites 15 articles, 3 of which you can access for free at:
http://bjp.rcpsych.org/content/196/5/372#BIBL
To obtain reprints or permission to reproduce material from this paper, please
write to [email protected]
/letters/submit/bjprcpsych;196/5/372
http://bjp.rcpsych.org/ on July 28, 2017
Published by The Royal College of Psychiatrists
To subscribe to The British Journal of Psychiatry go to:
http://bjp.rcpsych.org/site/subscriptions/