Methodological Aspects in the
Evaluation of Predictors of
Recovery: The SOHO Study
Johns Hopkins University
November 15th, 2007
Josep Maria Haro
St. John of God Research Foundation, Barcelona
Outline
The SOHO Study and the challenge of observational research
for the treatment of schizophrenia
Summary of 36-month results
Analysis of remission
Methodological considerations in the analysis of the SOHO
findings
What is the rationale of the SOHO study.
current knowledge on medication outcomes ?
Findings mostly come from randomized clinical trials, which are
the gold standard for evaluation treatment efficacy.
Approximately 20% of patients are elegible for enrolment in
clinical trials of schizophrenia (Hofer, 2000).
Very low retention rates (less than 50%)
Clinical trials are of short duration.
Focus on clinical outcomes.
Few observational studies.
Internal vs external validity
high
Important Trade off’s
Prospective
Observational
Studies
External
Validity
Randomised
Open Trials
low
RCTs
low
Internal Validity
high
RCT vs OS
RCT
Protocol driven
Homogeneous ‘ideal’
patients
‘Experienced’ investigators
‘Motivated’ patients
‘Pharmacological’
differences
Randomization
OS
Usual Care
Heterogeneous ‘real’
patients
‘Everyday’
practitioners
’Usual’ patients
’Overall care’ differences
Control of selection bias
Characteristics of an OS on the
treatment of schizophrenia (SOHO study)
Large:
Number of patients
Allow for comparisons of key measures.
Simple:
Shorter Study Description
Briefer Data collection forms (DCF)
Investigator recruitment / training
Patient enrolment process
Time needed to complete DCF (data load and complexity)
Follow-up required with investigator
Pragmatic: reflect actual practice, patient heterogeneity and
treatment practices
OS vs RCT Summary
Results from Observational studies and Clinical trials complement
themselves
Observational studies collect data and do not direct course of care
Observational studies need to be large and simple
Observational studies differ from RCT in:
type of information you get
analysis of the data
implementation
Bias can be reduced by careful consideration of design and data
collection.
The SOHO Study
SOHO Study Objectives
1.
2.
To understand the comparative costs and outcomes of
therapy for schizophrenia with olanzapine versus other
antipsychotic medications in a real world setting
To understand the pharmacological treatment patterns for
schizophrenia, how these patterns are associated with
olanzapine versus other antipsychotics, and how these
patterns are associated with outcomes
1. Haro JM, et al. Acta Psychiatry Scand 2003;107:222-32.
Study Design
Observational study
All patient care is at the discretion of the participating
psychiatrists
Description
Prospective study
Enrolment
10,972 patients enrolled across 10 countries:
•
•
•
•
•
Denmark
France
Germany
Greece
Ireland
•
•
•
•
•
Italy
Netherlands
Portugal
Spain
United Kingdom
Patient selection
Patient selection criteria:
Treatment in the outpatient setting
Treated for schizophrenia
Initiating or changing antipsychotic treatment
(treatment decision made prior to enrolment)
18 years of age or older
Two groups; patients who:
Initiate with or change antipsychotic treatment to olanzapine
Initiate with or change treatment to non-olanzapine
antipsychotic
Approximately, 50% of sample in each group
Initiation of treatment
versus prevalent sample
Objective of the study was to assess medication outcomes
Incident vs prevalent cohort
Need to assess patients at a common point in time
(episodes of care)
Reasons for medication change
Speed of recruitment
Outcomes in SOHO
Functioning
Patient reported
Clinical severity
Social relations
Marital relation
Hostility
Positive
Negative
Quality of life
Depressive
Housing
Work
Cognitiv
e
Overall
Depressive
symptoms
Measurements
Patients followed for 3 years:
Assessments at 3, 6, 12, 18, 24, 30, and 36 months post-baseline
Measures
Demographics
Functional status
Medical resource Use
Antipsychotic and other medications
Clinical status
Tolerability
Compliance
Criminality, violence, victimisation
Quality of life
The SOHO study
No change in clinical practice (visits or assessment) - Simple
evaluation not at predetermined time points.
Patients are included regardless of comorbidities, use of
concomitant medication, severity of the disorder.
No restriction regarding treatment changes.
Specialized research and non-research centers.
Baseline description
Patients Eligible for Analysis by Country at
baseline
Portugal
2%
UK/Ireland
3%
Denmark
<1%
N = 10,205
France
9%
Spain
20%
Netherlands
2%
Germany
29%
Italy
29%
UK/Ireland N=323
Denmark N=33
France N=933
Germany N=2,957
Greece N=701
Italy N=2,911
Netherlands N=166
Spain N=2,020
Portugal N=160
Greece
7%
2.Haro JM, Edgell ET, Frewer P, et al. The European Schizophrenia Outpatient Health Outcomes
Study: baseline findings across country and treatment. Acta Psychiatrica Scandinavica
2003:107(suppl 416):7-15
Sample by Cohort at Baseline
Medication Prescribed at Baseline Visit
Note: Study designed to enrol 50% of patients into olanzapine and non-olanzapine arms
Amisulpride
3%
Risperidone
19%
Quetiapine
8%
Clozapine
3%
Other Atypical
<1%
PO Typical
7%
Depot
5%
Olanzapine
53%
2+ AP's
3%
N = 10,205
Olanzapine N=5378
Risperidone N=1,918
Quetiapine N=790
Amisulpride N=328
Clozapine N=327
Other Atypical N=24
PO Typical N=688
Depot N=485
2+ AP's N=268
2+ Ap’s = Patients who initiate 2 or more antipsychotics at baseline
Baseline patient characteristics
Age (years)
40.1 ± 13.1
Age at first treatment contact for
schizophrenia (years)
28.9 ± 10.7
Percent female
Overall symptom score on CGI
42.2
4.4 ± 1.0
Relationship: spouse or partner (%)
30.0
Living independently (%)
48.3
Paid employment (%)
19.5
One or more social activities (%)
67.3
70% of the patients evaluated at 3 years
3-year results
Time to treatment discontinuation
Objectives
Replicate the design of CATIE study
Treatment discontinuation is an integrative variable, reflecting efficacy as
well as tolerability (overall usefulness of the medication)
To compare the relative effectiveness, in terms of treatment discontinuation,
of olanzapine, risperidone, quetiapine, amisulpride, clozapine, oral and
depot typical antipsychotic medications in outpatients with schizophrenia
during 3 years follow-up
Treatment discontinuation was defined as i) stopping the medication
initiated baseline or ii) adding a new antipsychotic to the medication
initiated at baseline
3. HaroJM, Suarez D, Novick D, Brown J, Usall J, Naber D, and the SOHO Study
Group. Three year antipsychotic effectiveness in the outpatient care of schizophrenia:
observational versus randomized studies results. European
Neuropsychopharmacology. 2007 Mar;17(4):235-44.
Methods
To analyse predictors of time to relapse a Cox proportional hazards model
was used. Model reduction (stepwise) was performed to determine the
effect baseline covariates have on treatment discontinuation
First episode
BMI
Age of first contact
Age
Sex
Country
Spouse
Housing
Employment
Social activities
Concomitant medication
CGI scales
EPS
Compliance
Substance abuse
Alcohol dependence
Hostility
EQ-VAS
Cohort
Maintenance of AP treatment
HaroJM, Suarez D, Novick D, Brown J, Usall J, Naber D, and the SOHO Study Group.Three
year antipsychotic effectiveness in the outpatient care of schizophrenia: observational versus
randomized studies results. European Neuropsychopharmacology. 2007 Mar;17(4):235-44.
Kaplan –Meier of time to discontinuation for
any cause by cohort
Olanzapine
Risperidone
Quetiapine
Amisulpride
Clozapine
Oral typical
Depot typical
Patients continuing with
treatment (%)
100
80
60
40
20
0
0
3
6
9 12 15 18 21 24 27
Time to discontinuation (months)
30
33
36
Haro JM, Suarez D, Novick D, Brown J, Usall J, Naber D, and the SOHO Study Group.Three year antipsychotic
effectiveness in the outpatient care of schizophrenia: observational versus randomized studies results. European
Neuropsychopharmacology. 2007 Mar;17(4):235-44.
Discontinuation:
1
SOHO/CATIE
Approximately 42% (74%) of the patients changed medication:
Quetiapine (66%/82%)
Oral typicals (53%/72%)
Depot typicals (50%)
Amisulpride (50%)
Risperidone (42%/74%).
Clozapine (33%) and olanzapine (36%/64%) were associated
to the lowest discontinuation rate.
1. Lieberman et al., N Engl J Med 2005: 353:1209–1223.
Discontinuation in Register Studies
Tiihonen, BMJ, 2006
Number of
patients
Number
discontinued
Percentaje
discontinued
Clozapine
150
85
57%
Olanzapine
197
118
60%
Risperidone
240
176
73%
Covell et al., 2002
42% Change in two years years
Hazard ratios for discontinuation of treatment by 36
months compared with the olanzapine cohort
Risperidone
HR 1.28 (95% CI 1.16, 1.42)
Quetiapine
HR 2.21 (95% CI 1.95, 2.51)
Amisulpride
HR 1.63 (95% CI 1.33, 1.99)
Clozapine
HR 0.82 (95% CI 0.65, 1.02)
Oral typical
HR 1.70 (95% CI 1.46, 1.96)
Depot typical
HR 1.42 (95% CI 1.19, 1.70)
0.5
1
1.5
Hazard ratio
2
2.5
Haro JM, Suarez D, Novick D, Brown J, Usall J, Naber D, and the SOHO Study Group.Three year
antipsychotic effectiveness in the outpatient care of schizophrenia: observational versus randomized
studies results. European Neuropsychopharmacology. 2007 Mar;17(4):235-44
3
Baseline factors associated with a
higher risk of discontinuation
Effectiveness as reason for change at baseline
HR 1.15 (95% CI 1.06, 1.24)
Higher CGI overall severity
HR 1.10 (95% CI 1.05, 1.14)
Mood stabilisers vs no usage
HR 1.18 (95% CI 1.04, 1.33)
Substance vs no substance abuse
HR 1.26 (95% CI 1.01, 1.57)
Hostile vs no hostile behaviour
HR 1.11 (95% CI 1.01, 1.21)
0.5
1
Hazard ratio
1.5
2
Haro JM, Suarez D, Novick D, Brown J, Usall J, Naber D, and the SOHO Study Group.Three year
antipsychotic effectiveness in the outpatient care of schizophrenia: observational versus randomized
studies results. European Neuropsychopharmacology. In press
Baseline factors associated with a lower risk of
discontinuation*3
Never treated vs Previously treated
HR 0.81 (95% CI 0.69, 0.94)
One social activity vs no social activity
HR 0.87 (95% CI 0.80, 0.95)
Time since first treatment
per year
HR 0.99 (95% CI 0.99, 0.99)
Hazard ratio with 95% confidence interval
0.5
*Cox regression analysis
A hazard ratio of >1 indicates an increased risk of discontinuing treatment
1
Hazard ratio
Haro JM, Suarez D, Novick D, Brown J, Usall J, Naber D, and the SOHO Study Group.Three
year antipsychotic effectiveness in the outpatient care of schizophrenia: observational versus
randomized studies results. European Neuropsychopharmacology. 2007 Mar;17(4):235-44
1.5
Relapse
Definition of relapse
Relapse was defined as an increase in CGI overall severity
score or as an inpatient admission.
To qualify for relapse, the increase in the CGI overall severity score had
to be at least 2 points when the minimum score for that patient during
follow-up was 1 (normal), 2 (borderline ill) or 3 (mildly ill) or at least 1
point when the minimum score was 4 (moderately ill), 5 (markedly ill)
or 6 (severely ill).
The increment in CGI score had to lead to the patient being at least
moderately ill (CGI 4).
Novick D, Haro JM, Suarez D, Ratcliffe M. Factors associated with risk of relapse in Schizophrenia. 36month results from the Schizophrenia Outpatient Health Outcomes (SOHO) study.
Abstract Published in Schizophrenia Research 2006; 81 (supp) 60
Baseline factors associated with relapse
Factors associated to higher risk
Having hostile behaviour vs.not having
hostile behaviour
RR (95% CI) => 1.23 (1.12, 1.35)
Higher CGI depressive score
RR (95% CI) => 1.08 (1.04, 1.12)
Higher CGI overall severity
RR (95% CI) => 1.09 (1.05, 1.14)
Taking anxiolytics vs. not taking anxiolytics
RR (95% CI) => 1.35 (1.24, 1.48)
Taking mood stabilizers vs not taking mood stabilizers
RR (95% CI) => 1.35 (1.19, 1.53)
-0.2
0.8
1.8
2.8
Relative risk
Novick D, Haro JM, Suarez D, Ratcliffe M. Factors associated with risk of relapse in Schizophrenia. 36month results from the Schizophrenia Outpatient Health Outcomes (SOHO) study.
Abstract Published in Schizophrenia Research 2006; 81 (supp) 60
Relative Risk of the Baseline Factors Found to
be Associated to Lower Risk of Relapse
Being employed vs. not being employed
RR 95% CI => 0.83 (0.73, 0.93)
Being the first time that a patient was receiving
treatment for schizophrenia
vs.
Not being the first time that a patient was receiving
treatment for schizophrenia
RR 95% CI => 0.78 (0.67, 0.92)
Being older (per year)
RR 95% CI => 0.99 (0.99, 0.99)
-0.2
0.3
0.8
1.3
Relative Risk
1.8
2.3
A Relative Risk >1 indicates an increased risk of having a relapse
* Cox proportional hazards model
Novick D, Haro JM, Suarez D, Ratcliffe M. Factors associated twith risk of relapse in Schizophrenia. 36-month results
from the Schizophrenia Outpatient Health Outcomes (SOHO) study. Abstract Published in Schizophrenia Research
2006; 81 (supp) 60
2.8
Relative Risk of Having a Relapse after 36 month of
Treatment compared with the Olanzapine Cohort*
Quetiapine
RR 95%CI => 1.67 (1.43, 1.94)
Risperidone
RR 95%CI => 1.17 (1.05, 1.31)
Oral Typicals
RR 95%CI => 1.25 (1.05, 1.48)
Depot Typicals
RR 95%CI => 1.60 (1.33, 1.92)
Amisulpride
RR 95%CI => 1.33 (1.06, 1.67)
Clozapine
RR 95%CI => 0.99 (0.79, 1.25)
-0.2
A Relative Risk >1 indicates an increased risk of having a relapse
* Cox proportional hazards model
0.8
1.8
2.8
Relative risk
Novick D, Haro JM, Suarez D, Ratcliffe M. Factors associated with risk of relapse in
Schizophrenia. 36-month results from the Schizophrenia Outpatient Health Outcomes (SOHO)
study. Abstract Published in Schizophrenia Research 2006; 81 (supp) 60
Remission
Remission: Andreasen et al.
Consensus-based operational criteria based on distinct thresholds for
reaching and maintaining improvement as opposed to change criteria
from baseline
Operational criteria based on:
1. Maintenance of low level of symptoms (a severity of mild or less as
assessed by a clinical severity scale) for
2. At least six months
3. Three main dimensions of schizophrenia syndrome
Psychoticism
Disorganization
Negative symptoms.
Social functioning not included. Not tested empirically.
6. Andreasen NC, Carpenter WT Jr, Kane JM, et al. Remission in schizophrenia: proposed criteria and rationale for
consensus. Am J Psychiatry 2005;162:441–9.
Andreasen et al. criteria (2005)
Validity of the CGI-SCH scale
4.Haro et al. Acta Psychiatrica Scandinavica, 2003
Remission: SOHO
Remission was defined as:
I)
Achieving a level of severity of mild or less (<3 in a scale from 1 to
7) in the CGI-S positive, negative, cognitive, overall severity score
that had been maintained for six months or longer, and
II)
Not having any inpatient admission during that period.
Validity remission definition
CGI-SCH compared with the PANSS scale: Kappa value 0.80 8
7.Haro JM, Novick D, Suarez D, Alonso J, Lepine JP, Ratcliffe M and the SOHO advisory board. Remission and
relapse in the outpatient care of schizophrenia: 3-years results from the Schizophrenia Outpatient Health Outcomes
(SOHO) study. Journal of Clinical Psychopharmacology. 2006;26:571-578
8. Haro et al. Acta Psychiatrica Scandinavica 2007
Methods: Definition of remission
CGI overall < 4
CGI positive < 4
CGI negative < 4
CGI cognitive < 4
Baseline
No hospitalization
6 months
CGI overall < 4
CGI positive < 4
CGI negative < 4
CGI cognitive < 4
12 months
18 months
7.Haro JM, Novick D, Suarez D, Alonso J, Lepine JP, Ratcliffe M and the SOHO advisory board. Remission and relapse in the
outpatient care of schizophrenia: 3-years results from
the Schizophrenia Outpatient Health Outcomes (SOHO) study. Journal of Clinical Psychopharmacology. 2006;26:571-578
Methods
• The follow-up observations were divided into 6-month periods (6–12 months,
12–18 months, 18–24 months, 24–30 months and 30–36 months).
• Each 6-month period was an observation in the statistical model.
• A Generalized Estimating Equation (GEE) model with a logit link and an
unstructured covariance matrix was used to account for correlation among
observations.
• The covariates included in the model were the socio-demographic and clinical
variables assessed at baseline and the medication taken at the beginning of
each of the periods, defined as the medication the patient had most recently
started
Medication started at study entry for the patients
included in the analysis
Risperidone
19.43%
Quetiapine
7.46%
Amilsupride
2.95%
C lozapine
3.47%
Oral Typicals
6.81%
Depot Typicals
5.11%
Olanzapine
52.30%
Olanzapine
Quetiapine
Clozapine
Depot Typic als
2+Antipsychotics
2.47%
Risperidone
Amisulpride
Oral Typic als
2+Antipsychotics
N=6,516
7.Haro JM, Novick D, Suarez D, Alonso J, Lepine JP, Ratcliffe M and the SOHO advisory board.
Remission and relapse in the outpatient care of schizophrenia: 3-years results from the
Schizophrenia Outpatient Health Outcomes (SOHO) study. Journal of Clinical Psychopharmacology.
2006;26:571-578
Baseline Clinical and
Socio-Demographic Patient Characteristics
Remission
(N=4206)
No Remission
(N=2310)
Years since first treatment for schizophrenia a
10.6(10.4)
14.0(11.7)
Age at First contact a
28.7 (10.2)
28 (10.2)
CGI Overall severity score at baseline a
4.2 (1.0)
4.8 (0.8)
CGI Positive score at baseline a
3.6 (1.4)
4.1 (1.4)
Negative CGI at baseline a
3.8 (1.3)
4.6 (1.1)
Gender (Male)
55.2%
62.1%
Never treated for schizophrenia before study entry
11.7%
5.8%
Paid employment
24.4%
10.7%
2.1%
2.3%
Usen of anxiolytics after baseline
33.7%
41.9%
Extrapyramidal symptoms
36.1%
43.1%
Substance abuse at baseline
7.Haro JM, Novick D, Suarez D, Alonso J, Lepine JP, Ratcliffe M and the SOHO advisory board.
Remission and relapse in the outpatient care of schizophrenia: 3-years results from the
Schizophrenia Outpatient Health Outcomes (SOHO) study. Journal of Clinical
Psychopharmacology. 2006;26:571-578
Remission: Andreasen et al.1
Consensus-based operational criteria based on distinct thresholds for
reaching and maintaining improvement as opposed to change
criteria from baseline.
Operational criteria based on:
1. Maintenance of low level of symptoms (a severity of mild or less
as assessed by a clinical severity scale) for
2. At least six months
3.Three main dimensions of schizophrenia syndrome
Psychoticism
Disorganization
Negative symptoms.
1. Andreasen NC, Carpenter WT Jr, Kane JM, et al. Remission in schizophrenia: proposed criteria and rationale
for consensus. Am J Psychiatry 2005;162:441–9.
Odd Ratios of the Baseline Factors Found to be Associated with
Achieving Remission
Being employed vs. not being employed
OR (95% CI) => 1.49 (1.31, 1.69)
Being the first time that a patient was receiving treatment
for schizophrenia
vs.
Not being the first time that a patient was receiving
treatment for schizophrenia
OR (95% CI) => 1.60 (1.33 –1.93)
Having a relationship with a partner vs. non having a
relationship
OR (95% CI) => 1.24 (1.11 ,1.38)
Having social activities during the four weeks prior to
assessment vs. non having social activities
OR (95% C)I => 1.26 (1.13, 1.40)
-0.2
0.3
An odds ratio of >1 indicates an increased likelihood of achieving remission
0.8
1.3
Odds ratio
1.8
2.3
7.Haro JM, Novick D, Suarez D, Alonso J, Lepine JP, Ratcliffe M and the SOHO advisory board. Remission and relapse
in the outpatient care of schizophrenia: 3-years results from the Schizophrenia Outpatient Health Outcomes (SOHO)
study. Journal of Clinical Psychopharmacology. 2006;26:571-578
2.8
Odd Ratios of the Baseline Factors Found to be Associated with
Lower Frequency of Achieving Remission
Time since first contact per year
OR (95% CI) => 0.99 (0.98, 0.99)
Age at first treatment
OR (95% CI) => 0.99 (0.98, 0.99)
Male Gender
OR (95% CI) => 0.79 (0.71, 0.87)
Higher CGI positive score at baseline
OR (95% CI) => 0.91 (0.87, 0.95)
Higher CGI negative score at baseline
OR (95% CI) => 0.77 (0.74, 0.81)
Higher CGI cognitive score at baseline
OR (95% CI) => 0.90 (0.85, 0.93)
0
Odds ratio
1
7.Haro JM, Novick D, Suarez D, Alonso J, Lepine JP, Ratcliffe M and the SOHO advisory board. Remission and relapse in the
outpatient cae of schizophrenia: 3-years results from the Schizophrenia Outpatient Health Outcomes (SOHO) study. Journal of
Clinical Psychopharmacology. 2006;26:571-578
Odd Ratios of the Baseline Factors Found to be Associated with
Lower Frequency of Achieving Remissio
Higher CGI overall severity at baseline
OR (95% CI) => 0.76 (0.70, 0.82)
Taking anxiolytics vs. not taking anxiolytics
OR (95% CI) => 0.78 (0.70, 0.86)
Taking mood stabilizers vs not taking mood stabilizers
OR (95% CI) => 0.72 (0.61, 0.86)
An odds ratio of <1 indicates a decreased likelihood of achieving remission
0
Odds ratio and its 95% confidence interval
0.4
0.8
Odds ratio
7. Haro JM, Novick D, Suarez D, Alonso J, Lepine JP, Ratcliffe M and the SOHO advisory board.
Remission and relapse in the outpatient care of schizophrenia: 3-years results from the Schizophrenia
Outpatient Health Outcomes (SOHO) study. Journal of Clinical Psychopharmacology. 2006;26:571-578
1.2
Odd Ratios of achieving remission during 36-month of follow-up
compared with the Olanzapine Cohort
Quetiapine
OR 95%CI => 0.66 (0.56, 0.76)
Risperidone
OR 95%CI => 0.74 (0.66, 0.83)
Oral Typicals
OR 95%CI => 0.63 (0.55, 0.74)
Depot Typicals
OR 95%CI => 0.59 (0.50, 0.69)
Amisulpride
OR 95%CI => 0.73 (0.56, 0.94)
Clozapine
OR 95%CI => 0.78 (0.65, 0.95)
2+Antipsychotics
OR 95%CI => 0.64 (0.58, 0.70)
0
Odds ratio and its 95% confidence interval
(Haro
0.4
0.8
1.2
Odds ratio
JM, Novick D, Suarez D, Alonso J, Lepine JP, Ratcliffe M and the SOHO advisory board. Remission and
relapse in the outpatient care of schizophrenia: 3-years results from the Schizophrenia
Outpatient Health Outcomes (SOHO) study. Journal of Clinical Psychopharmacology. 2006;26:571-578
Limitations
Relapses occurring between two visits and not resulting in
hospitalizations were not detected.
Our measure of remission was based on the CGI scale, which is a valid
but less specific measure of clinical severity in schizophrenia than that
proposed by Andreasen et al1
The results should be interpreted conservatively due to the
observational study design
Observer and selection bias
in SOHO
Types of Bias
RCT
Selection Bias:
‘are the groups similar in all important
respects’
Information Bias:
‘has information been gathered in the
same way’
‘is the information bias random or in
one direction?’
Randomisation
Blinding
Try to avoid as
much as
possible
OS
Anticipate and
control for it
statistically
Subgroup
analyses
Study
different
outcomes
Structured
collection of
information
Comparison of several models of analyzing data:
CGI overall change from baseline to 6 months
Difference
Olanzapine –
Other AP †
Standard error
95% Confidence Intervals
(no controls)
–0.264
0.024
(–0.331, –0.216)
OLS
–0.203
0.022
(–0.246, –0.161)*
Propensity score
matching
–0.197
0.025
(–0.250, –0.124)
OLS with fixed
psychiatrist effects
–0.200
0.021
(–0.240, –0.159)
Bayesian MLMs
–0.187
0.021
(–0.227, –0.145)
Simple differences
regression
10.Haro et al. Appl Health Econ Health Policy. 2006;5(1):11-25
Marginal Structural Models
They are particularly useful when you have
Time dependent treatments
Time dependent confounders: observed covariates that are affected
by treatment and relevant to the outcome of interest Marginal
structural models
Subjects are assigned a weight that is proportional (inverse) to their
probability of having received their treatment exposure given their
covariates
Regression model can then be fit, using weights to control for
confounding
Increasingly used in HIV studies in which randomization is not feasible
Sterne Lancet 2005; Hogg et al Plos Med 2006; Peterson Am J
Epidem 2007
Comparison of "conventional" (logistic GEE model) and MSM
estimates for multiple treatment effects to achieve remission
(olanzapine is the reference treatment)
Treatment
"Conventional" estimates
MSM estimates
OR (95% CI)
OR (95% CI)
Olanzapine
--
--
Amisulpride
0.73 (0.56, 0.94)
0.84 (0.57, 1.22)
Clozapine
0.78 (0.65, 0.95)
1.04 (0.75, 1.43)
Depot typical
0.59 (0.51, 0.69)
0.77 (0.62, 0.96)
Oral typical
0.63 (0.55, 0.74)
0.72 (0.53, 0.99)
Quetiapine
0.65 (0.56, 0.77)
0.73 (0.58, 0.92)
Risperidone
0.74 (0.66, 0.83)
0.77 (0.65, 0.90)
Combination
0.63 (0.58, 0.70)
1.01 (0.85,1.20)
Box-plots of the logarithm of ‘stabilized’ inverse-probability of
treatment weights (SW) for by previous overall CGI severity
Olanzapine
L
o
g
Clozapine
4
4
2
2
L
o
g
0
0
S
W
S
W
-2
-2
-4
-4
1
2
3
4
O
ver al l
CG
I sever i t y
5
6
7
1
2
3
4
O
ver al l
CG
I sever i t y
5
6
7
Figure 2: Distribution of mean differences in CGI between
treatment groups, by psychiatrist. Results from the SOHO study
Percentage of psychiatrists
35
30
25
20
15
10
5
0
-5.1
-3.9
-2.7
-1.5
-0.3
0.9
2.1
3.3
4.5
Mean difference
10.Haro et al. Appl Health Econ Health Policy. 2006;5(1):11-25
Change in CGI score, assessed by the psychiatrist, compared
with the EQ-5D utility score, assessed by the patient10
Olanzapine
Not Olanzapine
EQ-5D score change
0.320
0.240
0.160
0.080
0.000
-0.080
-0.160
1
0
-1
-2
CGI score change
10.Haro et al. Appl Health Econ Health Policy. 2006;5(1):11-25
Other methodological aspects
Selection of patients: Include prevalent sample of patients /
patients who change medication
Size of estimates
The treatment model of schizophrenia facilitates the analysis of
medication outcomes and may not be extrapolated to other
disorders
Definition of cohort for:
Patients taking more than one medication
Patients changing medication
Prediction of outcomes – description of reality