Download ppt slides

Document related concepts

Medical ethics wikipedia , lookup

Clinical trial wikipedia , lookup

Epidemiology wikipedia , lookup

Rhetoric of health and medicine wikipedia , lookup

Adherence (medicine) wikipedia , lookup

Patient safety wikipedia , lookup

Randomized controlled trial wikipedia , lookup

Fetal origins hypothesis wikipedia , lookup

Electronic prescribing wikipedia , lookup

Management of multiple sclerosis wikipedia , lookup

Multiple sclerosis research wikipedia , lookup

Transcript
University of Nebraska Medical Center
Methods Used in
Comparative Effectiveness Research
September 18, 2012
by
Susan D. Horn, Ph.D
Institute for Clinical Outcomes Research
699 East South Temple, Suite 300
Salt Lake City, Utah 84102
801-466-5595 (V) 801-466-6685 (F)
[email protected]
www.isisicor.com
1
The Problem
• How best to treat the patient in your office right now?
• “Scientific studies” provide imperfect guidance
• Clinical medicine is untidy; innumerable variables
describe patients, providers, and practices
• Must consider clinical variability of patient
populations, intervention combinations, and
outcomes
2
Objectives of Presentation
1. Describe, compare, and contrast models to conduct comparative
effectiveness research, including randomized controlled trials,
analysis of claims databases, electronic medical record databases,
condition or treatment registries, and practice-based evidence (PBE)
study methodology.
2. Demonstrate how evidence-based practice and comparative
effectiveness of treatments and treatment strategies can be derived
using PBE study designs and health information systems, including
clinical examples.
3. Present examples of PBE clinical findings related to quality, policy,
and safety improvements.
3
Comparative Effectiveness Research –
What is it?
• Comparative effectiveness research is designed to
inform health-care decisions by providing evidence
on the effectiveness, benefits, and harms of
different treatment options.
• Evidence is generated from research studies that
compare drugs, medical devices, tests, surgeries, or
ways to deliver health care.
4
Comparative Effectiveness Research –
What types of evidence are used?
There are two ways that evidence is found:
1. Systematic reviews of existing evidence. Researchers
look at all available evidence about benefits and
harms of each treatment choice for different groups of
people from existing clinical trials, clinical studies,
and other research.
2. New studies. Researchers conduct studies that
generate new evidence of effectiveness or comparative
effectiveness of a test, treatment, procedure, or healthcare service.
5
What We Have and What We Need
We have Efficacy trials that determine whether an
intervention produces a specified result(s) under well
controlled conditions in a selected population – includes
randomized controlled trials (RCTs)
We need Effectiveness trials that measure outcomes of an
intervention under “real world” conditions in an
unselected clinical population. Hypotheses and study
designs of an effectiveness trial are formulated based on
conditions of routine clinical practice and on outcomes
essential for clinical decisions.
6
More Definitions
•
Efficacy – benefit under the best possible
circumstances (Can it work?)
»
»
•
Generation - RCT, explanatory
Synthesis – systematic review of RCTs (e.g. Cochrane)
Effectiveness – benefit under ordinary practice (Does
it work?)
»
»
Generation – effectiveness trial, pragmatic trial,
observational studies
Synthesis – systematic review of RCTs and observational
studies
Effectiveness Trials and RCTs
RCT
Practice effects
of RCT results
Progenitor of
RCTs
Effectiveness
8
Databases for Effectiveness Trials
•
RCT databases
•
Large claims databases, e.g., Medicare, Medicaid, CDC
•
HMO or VA databases from claims and electronic medical
records
•
Specific condition registries such as arthritis registry
•
Practice-based evidence study registries
 PBE studies overcome limitations of RCTs (that limit patient types and
treatments)
 More detailed patient, process, and outcome evaluation than is possible with
9
traditional registries or large claims datasets
RCT Databases
for Comparative Effectiveness Research
•
RCT databases – gold standard for efficacy and
value of interventions
•
Advantages include:

High internal validity
 Causal inferences can be made since patients with
known confounders are excluded and randomization
eliminates unknown confounders
10
RCT Databases
for Comparative Effectiveness Research
•
RCT databases – gold standard for efficacy and value of
interventions
•
Limitations include:
 Small sample sizes – too small to detect uncommon risks
 Follow-up periods too short to assess long-term benefits/risks
 Higher-risk patients typically excluded; limited external validity
 Level of monitoring more rigorous than done in routine practice
 High rates of treatment discontinuation
11
Electronic Databases
for Comparative Effectiveness Research
•
Three primary sources of electronic databases:
 Large health systems with electronic medical record data,
e.g., HMO, HMO Research Network (HMORN), and VA
databases
 Insurance claims databases, e.g., Medicare, Medicaid,
HCUP data
 Disease or procedure-specific registries
12
Electronic Databases
for Comparative Effectiveness Research
•
Advantages include:
 Identify and track patient populations over time
 Measure treatment exposures over time
 Assess some health status, health behaviors, and other
potential confounders
 Assess both positive and negative outcomes over time
 Better suited to evaluate safety as opposed to effectiveness
13
Insurance Claims Electronic Databases
for Comparative Effectiveness Research
•
Many States and health care and pharmacy insurance
providers make administrative claims data available,
including Medicare, Medicaid, Healthcare Cost and
Utilization Project (HCUP), Blue Cross Blue Shield,
and United Health.
•
Databases contain information on all health care
encounters in which billable services were delivered
14
Insurance Claims Electronic Databases
for Comparative Effectiveness Research
•
Advantages include:
 Cover millions of people
 Ability to provide treatment exposures and adverse
events, including hospitalizations and mortality, over
extended periods of time
 Provide population and subpopulation-based estimates
for various outcomes
15
Insurance Claims Electronic Databases
for Comparative Effectiveness Research
•
Limitations include:
 Functional and cognitive status, severity of illness,
and health behaviors cannot be obtained and can be
important unmeasured confounders
 Possibility of exposure misclassification if patient
does not apply for insurance coverage for some
treatment
16
Health System Electronic Databases
for Comparative Effectiveness Research
• Advantages
include:
 Data are of high quality usually
 Cover millions of persons including minority and
elderly populations
 Exist in electronic form so some data elements may be
exported for use in comparative effectiveness research
17
Health System Electronic Databases
for Comparative Effectiveness Research
•
Limitations include:
 Restricted ability to capture patients’ severity of
illness, functional and cognitive status, health behaviors
other than smoking or pain.
 Restricted access for CER unless researcher is part of
organization that owns the data
 Often required variables are in text so are not
exportable
18
Clinical Registry Electronic Databases
for Comparative Effectiveness Research
•
•
Systematically collected and stored health-related information
on specific patient populations, most often defined by a
particular illness or procedure.
Advantages include:
 Designed to collect detailed information related to a particular illness
or procedure, e.g., arthritis registry, CABG registry, cancer registries
•
Limitations include:


Typically created as an add-on or separate database from those used
for clinical care or payment
Limited information outside of particular illness or procedure
19
The Problems
• Minimizing Bias in Existing Data
• Capturing variation in intervention
combinations
• Capturing variation in outcomes
20
The Balancing Act
Longer follow-up
Strong external validity
Strong internal validity
Balanced groups
Outcomes clearly defined
Different than routine care
Defined patient population
Confounded by life
Real world settings
Large sample size
21
Causal Inference Comparison
Randomized
Experiment
Quasi-Experiment
Cause precedes
effect
Yes
Yes
Cause covaries
with effect
Yes
Yes
Alternate
explanations
implausible
Yes
?
22
Basic Problems in
Non-Randomized Studies
• Non-independence of Observations
«
Many statistical analyses are based on the assumption
that the observations are independent.
«
Studying patients treated in a single hospital or unit
or provider breaks this assumption
«
Intra-correlation among observations
23
Modeling to Account for
Non-Independence
• Hierarchical designs that account for the
intra-correlation
• Generalized estimating equation (GEE)
regression models for clustered data
• Robust standard errors
24
Basic Problems in
Non-Randomized Studies
• Confounding by Indication
»
Therapies administered in non-random
fashion
»
Prognostic characteristics influence therapy
»
Recipients of therapy at high risk for outcomes
»
Users differ from comparators in key respects
25
Propensity Score Theory
•
Multivariable scoring method that collapses
multiple observed predictors of treatment into a
single value (a score)
» Probability that a subject with given
characteristics receives specified treatment
» Removes confounding by components of the
score
» Used to: match, stratify, or model
26
Instrumental Variable AnalysisTheory
•
•
•
•
Identify instrument variables that are randomly
associated with individual case, correlated with
treatment, but uncorrelated with outcomes
Therefore, the instrument can effectively
randomize subjects across treatment arms to
achieve equal distribution
Controls for underlying differences in groups that
are unobservable (endogeneity bias due to
unobserved heterogeneity)
Supposedly permits estimation of causal effects
even when important confounders are unmeasured27
Instrumental Variables (IV) to address
confounding by indication/selection bias
• Used to estimate effect of missing predictor variable or when
there is measurement error in predictor variable
• Instrument itself does not belong in prediction equation but is
correlated with missing predictor variable. For example,
• Severity of illness measure is missing; instead use distance
from patient home to treatment center as IV
• No measure of smoking in a community; instead use
amount of tobacco taxes collected as IV to assess smoking
effect on health outcome.
• Disadvantage: Assumptions about IV not testable
28
Overcoming Selection Bias/
Confounding by Indication
•
Statistical adjustments:
–Matching
–Propensity scoring/instrumental variables
–Covariate adjustments (Severity of Illness)
•
Ongoing debate about the adequacy of
adjustments
29
PBE Presentation Overview
• What is practice-based evidence (PBE) study methodology?
• How does it incorporate
 patient heterogeneity
 treatment heterogeneity
 outcome heterogeneity
• Discuss examples of PBE comparative effectiveness
findings
• Discuss implications for health information technology
30
PBE Methodology
What makes this approach
different?
31
31
Comparative Effectiveness Issues Addressed
Using PBE
• Both patients and providers report data
• Data come from existing EMR with standardized data
elements about patient characteristics, treatments,
processes, patient-reported data, and multiple outcomes
• Data are part of routine documentation, so not an ‘addon’
• Rapid patient accrual since documentation is standard of
care
• Longitudinal and ongoing
32
Comparative Effectiveness Issues Addressed
Using PBE (cont)
• Patient comparability is addressed with Comprehensive
Severity Index (CSI): disease-specific, physiologic-based,
>2,200 criteria, >5,500 disease-specific criteria sets
• CSI addresses confounding by indication/selection bias
• Database includes all treatments with date/dose/intensity/route
• Can assess drug and non-drug combination therapies
• Findings of PBE-CER more readily translated into practice
33
Practice Based Evidence (PBE) Design
PBE is observational research that:
»
captures differential outcomes
»
associated with naturally occurring variations in
treatment
»
while adjusting for severity of illness and injury,
co-morbid conditions, and co-occurring treatments
34
34
Practice-Based Evidence Study Design
Standardize documentation of :
Process Factors
• Management Strategies
• Interventions
• Medications
Control for:
Patient Factors
• Psychosocial/demographic Factors
• Disease(s)
• Severity of Disease(s)
Measure:
Outcomes of interest
• Clinical
• Health Status
• Functional
• Cost/LOS/Encounters
• Productivity
› physiologic signs and symptoms
• Genetic information
• Measured at Multiple Points in Time
35
7 Signature Features of PBE Studies
1. Hypotheses can be focused or broad
2. All interventions are considered to determine
relative contribution of each
3. Broad patient selection criteria maximize
generalizability and external validity
4. Detailed characterization of the patient by robust
measures of patient severity, genetic information,
and functional status
36
7 Signature Features of PBE Studies
5.
Patient differences controlled statistically rather than
through randomization
6.
Facility and clinical/patient buy-in through use of
trans-disciplinary Clinical Practice Team
7.
Strength of evidence built through the research process
PBE findings are more generalizable and transportable
37
than RCT findings
PBE Signature Feature: Interventions
2. Consider all interventions to determine relative
contribution of each.

Uses a detailed characterization of the care
process through a well-designed point-of-care
(POC) documentation system
– User-defined and user friendly
– Time sensitive characterization of all
interventions
38
PBE Signature Feature: Patients
4. Detailed characterization of the patient by robust
measures of individual severity and functional status
®

Includes Comprehensive Severity Index (CSI )
– Over 2,200 condition-specific signs, symptoms,
and physical findings
– Continuous score: 0  ∞
– Admission, discharge, maximum during stay, visit

Genetic information

Includes Functional Independence Measure (FIM)
and/or other measures of functional status
39
PBE Signature Feature:
Clinical Practice Team
6. Facility and clinical buy-in through use of
transdisciplinary Clinical Practice Team that:







Develops and frames the questions
Defines variables
Gathers data
Interprets data
Implements findings
Fosters clinical and individual buy-in (bottom-up)
Facilitates knowledge translation
40
PBE Signature Feature:
Strength of Evidence
7.
Strength of evidence built through the research
process
 Added confounders preserve the significant
association
■ A change in outcomes follows a change in treatment
as predicted by the PBE model
■ Repeated studies on the same topic yield similar
findings
41
PBE Hallmarks
• Decisions are made by front-line clinicians vs.
researchers
• “Bottom-up” vs. “Top-down” approach
• Guidance from researchers (scientific
advisory board) and patient experience
42
42
PBE Hallmarks
–Non-experimental: Follows outcomes of
treatments actually prescribed
–Inclusive: Uses patient populations undergoing
routine clinical care
–Pragmatic: Uses actual clinical outcomes
–Lower Cost than RCTs
–Faster than RCTs
43
Practice-Based Evidence Study Design
• High external validity
-
-
Includes essentially all patients with specific
condition
Captures confounders that could affect relevant
treatment responses.
Reduces accidental associations between
treatments and outcomes.
44
Practice-Based Evidence Study Design
Standardize documentation of :
Process Factors
• Management Strategies
• Interventions
• Medications
Control for:
Patient Factors
• Psychosocial/demographic/history factors
• Disease(s)
• Severity of Disease(s)
Measure:
Outcomes of interest
• Clinical
• Health Status
• Functional
• Cost/LOS/Encounters
• Productivity
› physiologic signs and symptoms
• Genetic information
• Multiple Points in Time
45
Examples of Severity Systems to address
Selection Bias/Confounding by Indication
Diagnostic/Procedure Based Systems
• Clinical definition of severity
• Body Systems Count
•Charlson Comorbidity Index (1-yr death)
• 3 Disease Staging (hosp death)
Physiologic/Clinically Based
Systems
• 2 Apache II & III (ICU death)
• 2 Medisgroups (Atlas) (hosp death)
• Patient Management Categories (hosp
death)
• Resource-based definition of severity
•Acuity Index Method (LOS)
•APR DRGs (hosp $)
•Patient Management Categories (hosp $)
•Refined DRGs (hosp LOS, $)
CSI®
46
Comprehensive Severity Index (CSI®)
used to avoid selection bias or confounding by indication
• Severity defined as “physiologic complexity presented to medical
•
personnel due to the extent and interactions of a patient’s diseases”
Disease-specific: 5,500 disease-specific groups; over 2,200 distinct criteria.
ICD-9 codes trigger disease-specific patient signs, symptoms, and physical
findings used to score disease-specific and overall severity levels
•
No treatments used as criteria
• Comprehensive (all diseases)
• Clinically credible: computes disease-specific and overall severity levels
• Can measure severity at multiple time points
• Allows statistical comparison of interventions without confounding by
severity of illness
47
14
Characteristics of the CSI System

Developed as an addition to ICD-9-CM

Explicit severity criteria established for each ICD-9-CM
code

Severity criteria result in a continuous severity score for
each diagnosis

Failure to meet enough criteria results in a severity rating
of zero

Overall patient severity computed as continuous score,
taking all diagnoses into account
48
CSI Severity Indicators
 Physiological signs and symptoms of a disease
- Vital signs
- Laboratory values
- Radiology findings
- Other physical findings
 Severity indicators are specific to each disease
based on ICD-9-CM coding
49
Pneumonia Criteria Set
480.0-486; 506.3; 507.0-507.1; 516.8; 517.1; 518.3; 518.5; 668.00-668.04; 997.3; 112.4; 136.3; 055.1
CATEGORY
1
2
3
Cardiovascular
pulse rate 51-100; ST
segment changes-EKG;
systolic BP  90mmHg
pulse rate 100-129;
41-50; PACs, PAT,
PVCs-EKG;
systolic BP 80-89mmHg
pulse rate  130; 31-40;
systolic BP 61-79mmHg
pulse rate 30;
asystole, VT, VF,
V flutter;
systolic BP 60 mmHg
Fever
96.8-100.4 and/or chills
100.5-102.0 oral;
94.0-96.7
102.1-103.9; 90.1-93.9
and/or rigors
 104.0
90.0
Labs
ABGs
pH 7.35-7.45
pH >7.46 7.25-7.34
pH 7.10-7.24
pH 7.09;
pO2 51-60mmHg
pO2  50mmHg
WBC 11.1-20.0K/cu mm;
2.4-4.4K/cu mm;
bands 10-20%
WBC 20.1-30.0K/cu mm;
1.0-2.3K/cu mm;
bands 21-40%
WBC 30.1K/cu mm;
1.0K/cu mm;
bands 40%
chronic confusion
acute confusion
unresponsive
9-11
6-8
 5
Radiology Chest
X-Ray or CT
Scan
infiltrate and/or
consolidation in 1
lobe; pleural effusion
infiltrate and/or
consolidation in >1 but
3 lobes;
infiltrate and/or
consolidation in >3
lobes; cavitation or
lung necrosis
Respiratory
dyspnea on exertion;
stridor; rales 50%/3
lobes; decreased breath
sounds 50%/3 lobes;
positive for fremitus;
stridor
hemoptysis NOS;
blood tinged or purulent
or frothy sputum
cyanosis present
dyspnea at rest; rales
>50%/ 3 lobes;
decreased breath
sounds >50%/ 3 lobes
apnea
absent breath sounds
>50%/ 3 lobes
Hematology
pO2 61mmHg
WBC 4.5-11.0K/cu mm;
bands <10%;
Neuro Status
Lowest Glasgow
coma score
 12
white, thin, mucoid
sputum
4
 frank hemoptysis
50
Copyright
2006. Susan D. Horn. All rights reserved. Do not quote, copy or cite without permission.
Uses of PBE Findings
PBE findings can be used to

Create protocols depending on patient
characteristics that result in better outcomes

Evaluate treatments or programs
51
Examples of PBE Study Findings
• Children Hospitalized with RSV. Birth at 33-35 weeks
•
GA is significantly associated with higher intubation
rates, longer ICU stays, and longer hospital LOS
(prompted guideline change for prophylaxis).
Low-Level Stroke. More time spent in high-level
rehabilitation activities, such as gait, upper extremity
control, and problem solving in the first three hours of
therapy is significantly associated with higher total,
motor, and cognitive FIM at discharge.
52
Pediatric Bronchiolitis Study
Length of Stay
8
7.1
7
6.2
6
Day
s
5
4
5
4.3
4
4.3
4.8
3.5
4.5
3.6
3
2
1
0
Site Site Site Site Site Site Site Site Site Site
6
5
7
9
4
1
3
8
2
10
53
Pediatric Bronchiolitis Study
Cost
$14,000
12,373
$12,000
$10,000 8,839
10,041
8,934
$8,000
$6,000
9,342
6,097
4,908
4,522
Site Site
6
5
Site Site
7
9
4,122
$4,000
$2,000
$0
Site Site
4
1
Site Site Site
3
8
2
Site
10
54
Pediatric Bronchiolitis Study
Outcome = Length of Stay
Assessment
n=804 R2=.62
Procedures
- Age in months (.0006)
+ O2 used (.01)
+ MCSIC (.0001)
+ Steroids (.0001)
+ Antibiotics (.0001)
+ Intubation (.0001)
+ Lasix (.0001)
+ Interaction of chest
physiotherapy and
atelectasis (.0001)
55
Pediatric Bronchiolitis Study
Outcome = Cost
n=722
Assessment
- Age in months (.0001)
+ MCSIC (.0001)
R2=.73
Procedures
+ Admitted to PICU (.0001)
+ Arterial line (.04)
+ Central line (.003)
+ Continuous nebulization (.0002)
+ Interaction: chest pt & atelectasis (.005)
+ Intubation (.0001)
+ Ipratropium bromide (.005)
+ Lasix (.0001)
+ Ribavirin (.0001)
+ Steroids (.0003)
Willson, et al. PEDIATRICS 2001;108(4):851-855.
56
Prematurity and RSV
Hospital Outcomes
Significant Differences by Gestational Age Groups
33-35 week GA infants had highest hospital resource use
< 32 wks
33-35 wks
36 wks
> 37 wks
p-value
Intubation
21.4%
38.7%
20%
12.1%
0.002
ICU LOS
5.8 days
7.7 days
4.2 days
3.8 days
0.021
Hospital LOS
6.8 days
8.4 days
4.9 days
4.1 days
<0.0001
Admitted to ICU
39.3%
48.4%
30.0%
27.9%
0.101
HX of Hosp. for
RSV /Bronchiolitis
14.3%
16.1%
6.7%
6.1%
0.137
57
RSV Hospital Outcomes and Policy Changes
Conclusions
•
33-35 week GA infants had highest hospital resource use
•
36 week infants have risk similar to full term infants
•
Changed guidelines for immunoprophylaxis for 33-35 week
infants
•
Changed guidelines for intubation – try ‘stimulating’ first
58
Post-Stroke Rehabilitation Study
2001 – 2003; 1,161 patients
Study Objectives
PBE study designed to discover what combinations of
medical devices, therapies, medications, feeding
approaches, and their interactions worked best for
specific types of stroke patients treated in realworld practices.
59
Post-Stroke Rehabilitation Study
Trans-Disciplinary Project Clinical Team
•
•
•
•
•
Physicians
Nurses
Social Workers
Psychologists
Physical Therapists
•
•
•
Occupational Therapists
Recreation Therapists
Speech/Language
Pathologists
60
Outcome: Discharge Motor FIM
Severe Stroke – Full Stay
General
Assessment
– Age
PT
Interventions
– Formal
assessment
– Bed mobility
+ Mild motor impairment
+ Gait
+ Admission Motor FIM
+ Advanced gait
+ Admission Cognitive FIM
– Black race
OT
Interventions
+ Home
management
SLP
Interventions
– Swallowing
– Orientation
+ Reading
comprehension
Medications
General
Interventions
– Days onset to rehab
+ Enteral feeding
– Anti-Parkinsons
– Modafinil
– Old SSRIs
+ Atypical antipsychotics
61
Outcome: Discharge Motor FIM
Severe Stroke–1st 3 hour Therapy block only
General
Assessment
PT
Interventions
OT
Interventions
SLP
Interventions
– Age
– Bed mobility
+ Home management
st
– Severe motor impairment time in 1 3 hrs
+ Gait time in 1st 3
+ Admission Motor FIM
hrs
+ Admission Cog. FIM
+ Advanced gait
+ No Dysphagia
time in 1st 3 hrs
+ Neurotropic Impairments
Medications
General
treated with meds
Interventions
– Other Antidepressant
– Days onset to rehab
– Old SSRIs
+ LOS
+ Atypical antipsychotics
Horn et al., Arch Phys Med
Rehabil 2005;86(12 Supplement
+ Enteral feeding
62
2):S101-S114
Policy Changes from Stroke PBE Study
Early Rehabilitation Admission – get patient into
rehabilitation as soon as possible after stroke onset;
possibly start in Neuro ICU
Early gait in PT – start gait as soon as possible after rehab
admission; put patient in harness on treadmill for safety
Early Feeding – continue or start enteral nutrition at rehab
admission if patient is not able to eat full meals
Use Opioids for Pain – continue or start opioids at rehab
63
admission if patient misses therapy due to pain
RCTs, Observational Cohort Studies,
& PBE Compared
Feature
RCT
Observational Cohort
PBE
Researcher
control of
variables
Experimental
Observational
Observational
Time dimension
Prospective
Prospective or retrospective
Prospective
Intervention(s)
1 or 2 discrete
interventions;
standardized
protocols
All interventions deemed relevant;
1 or many; often a more global
documented separately and in great
intervention, e.g., a program of
detail so that one can examine
interventions
interactions among interventions
Hypotheses
Well-specified
Specified, general, or none
Focused or broad
Data source
Primary data
Primary data supplemented by
abstraction from clinical records
Protocol- vs.
data- intensity
Protocol-intensive
Primary data and/or large
administrative data sets
Not inherently protocolor data-intensive
Data intensive
Archives Physical Medicine & Rehabilitation 2012.
RCTs, Observational Cohort Studies,
& PBE Compared
Feature
RCT
Observational Cohort
PBE
Exclusion
criteria
Extensive exclusions to
minimize variation
Minimal exclusions (permits
studying specific subsets of
patients)
Minimal exclusions (permits
studying specific detailed subsets of
patients)
Sample size
Typically small (e.g.,
<200) because of narrow
hypothesis
Usually large
Large (e.g., >1,500)
Control for
participant
differences
Through exclusions and
randomization
Through propensity scoring,
instrumental variable (IV)
analysis, and statistical control
Thorough detailed characterization of comprehensive patient
severity of illness and other
relevant characteristics, and
statistical control
Blinding
Single, double, or triple
blinding
No
No
Outcomes
Few
Few or many
Effect size
Often small
Small or large
Many- appropriate to study
population
Usually small when samples are
large
Archives Physical Medicine & Rehabilitation 2012.
RCTs, Observational Cohort Studies,
& PBE Compared
Feature
Attitude to
Confounders
Validity
Causality
RCT
Observational
Cohort
Affect outcomes but may
Not interesting; exclude
not be measured so hard
them
to control
Internal—high
Internal—low
External—low
External—moderate
Assigned
Assumed
Ability to examine
treatment effects on
subgroups
Limited, unless predesigned and powered
for subgroup analyses
Research culture (1)
Top-down
More likely because of
large data sets, but
subgroup analyses may be
limited by patient control
strategies such as
propensity scoring
Varies
PBE
Affect outcomes, measure as many
as suggested by clinical team, and
are interesting
Internal—moderate to high
External—high
Assumed; with drill-down analyses,
implementation, and repeat studies
can move close to ‘assigned’
Very likely since measure details
about patients that are used to
create specific subsets
Highly collaborative; bottom-up
Archives Physical Medicine & Rehabilitation 2012.
RCTs, Observational Cohort Studies,
& PBE Compared
Feature
RCT
Observational
Cohort
Research culture (2)
Not depend on local
knowledge
Varies
Knowledge translation
Less buy-in
Varies
Science of ….
Confirmation
Confirmation
Science of ….
Efficacy
Associations
Cost
High
Varies
Archives Physical Medicine & Rehabilitation 2012.
PBE
Local knowledge highly
valued and sought
High level of buy-in;
findings more
“transporatable”
Confirmation, discovery,
and innovation
Associations and
effectiveness
Moderate