An Introduction to Clinical Trials: Type of Studies

Design Issues
• Non-experimental (Observational)
– Case report
Edgar R Miller III PhD, MD
– Case series
Welch Center for Prevention, Epidemiology and
Clinical Research – Cross-sectional (survey)
Johns Hopkins University – Case-control
School of Medicine and Bloomberg School of – Prospective, observational (cohort)
Public Health • Experimental
– Randomized, clinical trial (RCT)

Study designs Advantages of Clinical Trials

• Observational studies: • Often provides the strongest evidence in
– Observe both exposures and outcomes support of cause-effect relationships
• Experimental studies (clinical trials)
– Assign exposures • Basis for clinical and public health policy
– Observe outcomes

• Minimize/eliminate bias and confounding

3 4
 Comparison of Study Designs
Randomized Clinical Trial
Target Population Type of Study Design

Cross- Case-
Study Population Dimension Sectional Control Cohort RCT
Estimate A - B -
Prevalence
RANDOMIZED Estimate - - A B
Incidence
Prove C B- B+ A
Standard Treatment New Treatment Causality
Generalizability A B+ B+ B
Disease Disease Feasability A A B C
5 6

Core Elements of a Clinical Trial The Research Question

• Research Question • Data
• Critical in the design of a trial
• Hypotheses • Analytical Issues • Types of questions:
• Core Design • Interpretation of – Assessing efficacy of an intervention
• Study Participants Results – Assessing the effectiveness of an
intervention
• Recruitment
• Allocation
• Masking (Blinding)
• Treatment Groups
7 8
 Types of Hypotheses Types of Hypotheses
• Comparative Trial (a.k.a. Superiority Trial) • Equivalence (non-inferiority trial)
– Objective: to demonstrate that a new therapy – Objective: to demonstrate that a new therapy
(n) is superior to standard therapy (s) in terms (n) is no worse than standard therapy (s) in
of incident outcome (I) terms of incident outcome (I)
HO: In = Is HO: In > Is
HA: In < Is (one tailed) or HA: In ≠ Is (two tailed) at HA: In = Is at some ∆, the maximum tolerable
some minimally detectable ∆ judged to have clinical difference considered to be clinically acceptable
significance

9 10

Basic Types of Design Parallel Study Design (PREMIER)

A
Parallel Randomization

B ADVICE ONLY
EST
A A EST + DASH
Cross-Over

B B
Primary End of
Outcomes Intervention
= Data Visit (18 months)
11
(6 months)
 Cross-Over Study Design
4
Control Diet Fruits-and-vegetables Diet DASH (OmniHeart)
2

0 Randomization
Washout Washout
to 1 of 6
-2 Period Period
sequences
2–4 wk 2-4 wk
-4

-6
* Screening/ Run-In Period 1 Period 2 Period 3
-8
Baseline 6 days 6 weeks 6 weeks 6 weeks
-10
**
Data:
-12
Baseline 1 2 3 4 5 6 7 and 8
Participants Ate Their Own Food
Intervention Week
Conlin et al., Am J Hypertens, 2002 Participants Ate Study Food

Blood Pressure Results Mixed Study Design (DASH-Sodium)

(mmHg) Randomized Sequence
Randomization
Mean Change from Baseline in to Diet
Each Diet Lower
Sodium
Intermediate
Sodium
Higher
Sodium

Systolic BP Baseline CARB PROT UNSAT Usual Diet Usual Diet

DASH Diet
All 131.2 -8.2 -9.5 -9.3 Lower Intermediate Higher
Sodium Sodium Sodium
HTN Only 146.5 -12.9 -16.1 -15.8
Run-in (11-14 days) Intervention (three 30-d periods in random order)
PreHTN Only 127.5 -7.0 -8.0 -7.7

Diastolic BP 77.0 -4.1 -5.2 -4.8

Appel et al. 2005
16
 Effect of Increased Sodium Intake on
Systolic Blood Pressure in Two Diets: Results of the DASH- Factorial Design
Sodium Trial*
• Type of trial in which individuals are randomized to
two or more therapies (example: Physician’s Health
135 American Diet Study: tested aspirin (ASA) and β-carotene
+2.1

+4.6
+6.7
No β-carotene β-carotene
Systolic 130 p<.0001

Blood No Neither β-carotene

10,000
ASA only
Pressure +3.0
125
+1.3
P<.0001
ASA only Both 10,000
+1.7 DASH Diet ASA
120
65 100 140
10,000 10,000 20,00018
*Sacks et al, 2001 Approximate Daily Sodium Intake (mmol/day)

The African American Study of Kidney Disease AASK Research Questions

and Hypertension (AASK)
Among African-Americans with early evidence
of hypertension-related kidney disease:

• Does aggressive blood pressure control to a

target blood pressure below current
recommendations retard the progression of
kidney disease?

• Do specific classes of anti-hypertensive

medications retard the progression of kidney
disease?
 Treatment Assignments
Design of AASK (2:2:1 ratio of drug assignment)
3 X 2 Factorial Design

• Randomized, active controlled trial with a

Metoprolol* Ramipril Amlodipine
2 x 3 factorial design
• Participants: 1,094 African-Americans MAP <92 20% 20% 10%
with hypertension-related renal
insufficiency
MAP 102-107 20% 20% 10%
• Planned follow-up of 2.5 to 5 years
N 441 436 217

MAP = Mean Arterial Pressure; * = referent group

Mean Arterial Pressure During Follow-up Composite Clinical Outcome

130 Declining GFR Event, ESRD or Death
Lower BP Goal (Achieved: 128/78) 40
Lower BP (Achieved: 128/78)
120 Usual BP Goal (Achieved: 141/85) 35 Usual BP (Achieved: 141/85)
MAP (mm Hg)

% with Events

30 Low vs. Usual:

RR=2%, (p=0.85)
110 25
20
100 15
10
90 5
0
80 0 6 12 18 24 30 36 42 48 54 60
0 4 12 20 28 36 44 52 60 Follow-Up Time (Months)
Follow-up Month 23 24
RR=Risk Reduction, adjusted for baseline covariates
 Main Clinical Composite Outcome
Declining GFR Event, ESRD, or Death
40 Amlodipine
35
Ramipril
Metoprolol
30
% wi t h Even t s

Ramipril vs. Metoprolol

25
RR = 22%, p = 0.042
20
15
Metoprolol vs. Amlodipine:
10 RR= 20%, p=0.17
5 Ramipril vs. Amlodipine:
RR= 38%, p=0.004
0
0 6 12 18 24 30 36 42 48 54 60
Follow-up Month
25
RR = Risk Reduction
RR = Risk Reduction, Adjusting for Baseline Covariates

Study Participants Study Participants: Example

Target
Accessible • Target Population -> Healthy Elderly
Population
Population

• Accessible Population -> Retired Teachers

Study
Samples • Study Sample -> Volunteer Teachers who
respond to mass mailing

27 28
 Study Participants Enrollment Criteria
• Inclusion Criteria
• Ideal ‘Accessible’ Population
– characteristics of accessible population
– high risk for disease
• Exclusion Criteria
– candidates for treatment
– considerations related to:
– representative of target population
• adherence to therapy
– feasibility considerations • follow-up
• recruitment • safety
• follow-up • ethics
• high quality data

29 30

Common Recruitment Strategies Comments on Recruitment

• General mailings • Screenings
– Licensed drivers – Worksite
• Recruitment begins with design
– Voters
– Employee paychecks
– Community
• Response rate is always lower than
• Physician Referral
• Targeted mailings • Medical Record Review expected
– HMO enrollees • Internet / WWW
– AARP members – Clinical trial registries
• Required resources are more than
• Mass media – Banner ads expected
– Radio – Social networks
– TV ads • Dedicated personnel are necessary
– Newspapers
– Posters/flyers

31 32
 Recruitment “Funnel”
More Comments on Recruitment
(Example: VITAL Pilot Study)
• Recruitment period is often longer than expected 4,774 Mailed Invitations
43%
• Implement several strategies to identify best
2,034 Questionnaires Returned
source
38%
• Prepare back-up strategies 765 Interested After Initial Mailing
• Monitor recruitment 41%
– Early 323 Randomizable after Second Mailing (7% cumulative)
– Often
297 Randomized
– Locally

33 34

Allocation Why randomize?

• Random
– stratified • Two critical reasons:
– blocked – to eliminate selection BIAS
– to reduce/avoid CONFOUNDING from known and,
• Non-Random
more importantly, unknown confounders
– haphazard
– systematic

35 36
 Masking (Blinding) Masking (Blinding)
• Single Blind Single Double Triple
– Observers (persons who collect outcome Masked Masked Masked
variable) do not know treatment assignment Outcome X X X
• Double Blind Assessor(s)
– Study participants AND observers do not know Participant X X
treatment assignments
• Triple Blind Data X
– Data interpreters, study participants, and Interpreter
observers do not know treatment assignments37 38

Problems with selecting

Selection of Groups
active treatment group
• Active Treatment Group
• Many Candidate treatments
– observation studies, animal models, or
• Comparison Group theoretically based
– Placebo (no active therapy)
– Usual care (referral back to personal MD) • Strong evidence rarely exists to guide
– Active control group (provision of standard selection of intervention
therapy)

39
• Dose/intensity are uncertain 40
 Problems with standard of
Comparison Group
care approach
• Placebo – used in setting of: • Efficacy of ‘Usual care’ often not tested
– No standard therapy OR
– Standard therapy but risk of not providing it • Variations in standard of care are common:
is minimal – across providers
– between experts and providers
• Usual care OR active control – common – secular trends occur

41 42

Data Outcome Variables

• Baseline data • Principal outcome
– Determine eligibility – most important variable after
– Describe study participants randomization code
– Define subgroups – specified in hypothesis
– Address confounding – determinant of sample size
• Measures of Adherence
• Secondary Outcomes
• Outcome Variables
– relevant to research question
43 44
 Desirable Features of Surrogate Outcomes
Outcome Variable
• Definition: a laboratory measurement
• clinically relevant or physical sign used as a substitute
• easy to measure for a clinically meaningful outcome
• little measurement error
–random error – leads to imprecision • Types: physiologic variable, clinical
–systematic error – leads to bias risk factor, or sub-clinical disease
• masked (blinded) ascertainment
45 46

Advantages of Surrogate Advantages of Surrogate

Outcomes Outcomes (continued)
• Surrogate outcomes typically increase • Enhanced power means
statistical power compared to clinical
outcomes – shorter duration of follow-up and/or
reduced sample size
– Surrogate outcomes
• often continuous – less cost
• measured repeatedly • Less contamination by competing
– Clinical outcomes comorbidities if the study duration is short
• often categorical • Useful in studies of mechanisms
• surveillance till outcome occurs 47 48
Surrogate and clinical outcomes: Surrogate and clinical outcomes:
a continuum an example

Antecedent Established Morbid Cause- Total Weight Blood Angina MI CVD Total
of the Risk Risk Factor Events Specific Mortality Pressure Mortality Mortality
Factor Mortality

Relationship between Change in

Relationship between Surrogate
Surrogate Outcome and Change
and Clinical Outcomes
in Clinical Outcomes

Surrogate Outcome Change in Surrogate Outcome

 Clinical and Surrogate
Outcomes: Cardiovascular
Clinical Surrogate
Stroke Ultrasound measurement of intimal
medial thickness of the carotid artery
Blood pressure
Myocardial Quantitative coronary angiography
infarction Electron beam computerzied tomography
Sudden Ventricular arrhythmia
death
Heart failure Ejection fraction

Disadvantages of Surrogate
Outcomes
Weaknesses • Measurement of surrogate outcomes can involve
complex, technical procedures
– procedures sometimes new (therefore,
longitudinal data is scant)
– procedures become obsolete
– many technical and analytic issues, often
unapparent
 Disadvantages of Surrogate
Outcomes (continued)
• Missing values are commonplace Models for success and
• Missing values result from loss to follow-up and
poor quality of data
failure of surrogate
• Potential for bias outcomes*
– missing values occur in the sickest people,
sometimes because of the clinical outcome of
interest *Fleming TR, DeMets DL. Surrogate End
– informative censoring, that is, loss of follow- Points in Clinical Trials: Are we being mislead?
up data potentially related to treatment Ann Int Med 1996;125:605-613.
assignment

Model for potential success: Surrogate Model for potential success: Surrogate
outcome in the casual pathway outcome in the casual pathway

Intervention Diuretics

Disease Surrogate Clinical Hypertension Blood Stroke

Outcome Outcome Pressure

Time Time
 Model for failure: the surrogate is Model for failure: the surrogate is
not in the causal pathway of the not in the causal pathway of the
disease process disease process
Fluoride
Intervention

Surrogate
Outcome  Bone Density
Clinical
Disease Osteoporosis fractures
Outcome

Model for failure: the intervention Model for failure: the intervention
affects only the pathway mediated affects only the pathway mediated
through the surrogate through the surrogate

Intervention Protein
Restriction

Surrogate Clinical Kidney

 Proteinuria Failure
Outcome Outcome
Disease Kidney
Damage
Model for failure: The intervention Example: Dihydropyridine calcium
has several mechanisms of action channel blockers
Intervention Calcium Channel
Blockers

Surrogate Clinical +
Disease Outcome Outcome Blood Myocardial
ASCVD
_ Pressure Infarction

The Cardiac Arrhythmia Suppression

CAST Research Question
Trial (CAST*): Background
• Ventricular arrhythmias are a risk factor Does suppression of
for sudden death after MI
• Four fold higher risk of cardiac mortality ventricular ectopy after a
among persons with frequent premature MI reduce the incidence of
ventricular contractions (PVCs)
• In the CAST pilot study, the sudden death?
antiarrhythmic drugs (encainide,
flecainide) suppressed PVCs
*Echt DS et al. Mortality and morbidity in patients receiving
encainide, flecainide, or placebo. NEJM 1991: 324(12): 781-8.
 CAST Design
• Design: randomized trials of
– encainide vs placebo
– flecainide vs placebo
• Participants (n=1498)
– recent MI (6 days to 2 years ago)
– ventricular ectopy (6 or more PVCs /hr)
– at least 80% suppression of PVCs by
active drug during open label titration
period prior to randomization
Source: Echt DS, Liebson PR, Mitchell B, et al. Mortality and morbidity in patients
receiving encainide, flecainide, or placebo. The Cardiac Arrhythmia Suppression
Trial. NEJM 1991: 324(12): 781-8.

CAST results: number of deaths

Lessons from CAST
and cardiac arrests by group
• Active treatment: 63 events / 755 • Active treatments can be harmful (one of
• Placebo: 26 events / 743 several recent trials in which placebo was
superior to active treatment)
p = 0.0001
• Reliance on surrogate outcomes can be
misleading
• same pattern of results for • The scientific community should encourage
– death from arrhythmia researchers and sponsors to conduct studies
– death from any cardiac cause with ‘hard’ clinical outcomes
– death from any cause
 Examples from the Field Model for potential success: Surrogate
• Surrogate that go in that go the right direction
outcome in the casual pathway
(easy to explain –fit your hypothesis)
Diet Change
• Surrogates that go in unexpected directions
(create a greater need for hand-waving and
but can still be made to fit your hypothesis) ASCVD
↑ oxidative ↓ oxidative
stress stress
• Surrogates that behave badly
Time

Dietary Patterns

Dietary Antioxidants LDL Cholesterol

Vitamin C Vitamin C
Vitamin E
beta-carotene
β-carotene
Oxidized LDL
Alpha-tocopherol
Oxidative stress Inflammatory
Free Radical Activity Markers Markers

Fatty Streak Formation

Atherosclerosis
Figure 2b
 Nurses Health Study
• Design: Prospective Cohort Study
• Participants: 121,700 female nurses free of
diagnosed cardiovascular disease
• Exposure Dietary questionnaire at baseline
Assessment Vitamin E and Multivitamin Use
• Follow-up: 8 years
• End Points: 1) Major Coronary Disease
2) Non-fatal MI
3) Deaths Due to Coronary Disease

N Engl J Med 1993;328:1444-1449

Prospective observational studies of vitamin E: Effects

on cardiovascular end points

N Engl J Med 1993;328:1444-1449

Adapted from: Jha, P. et. al. Ann Intern Med 1995;123:860-872
 Summary of Biological Evidence

• Antioxidants are necessary

• Oxidized lipids are associated with CVD

• Oxidation of lipids is reduced by antioxidant

supplementation

• Does supplementation lower risk of CVD?

– Observational studies
– trials

Do Vitamin E supplements Clinical Trials – Clinical

reduce risk? Outcomes
• Observational studies are confounded –vitamin E
takers exercise more, have a lower BMI, eat healthier • Cardiovascular Events
diets and smoke less often that non-vitamin users
– Fatal and Non-fatal MI
• Observational studies are hypothesis generating – Stroke
– Peripheral artery disease
• Surrogate markers are only indirectly related to
clinical events

• Benefits can only be assessed in randomized

• Mortality
controlled clinical trials
 ATBC Study
• Design: Randomized, double-blind, placebo-
controlled primary prevention
trial
• Participants: 29,133 male Finnish smokers,
age 50-69
• Intervention: 1) Vitamin E 50 IU/day
2) B-carotene 20 mg/day
3) Combination
4) Placebo
• Follow-up: 5-8 years
• End Points: Incident lung cancer & deaths
ATBC, 1993 NEJM

ATBC Trial Results CARET Study

• Design: Randomized, double blind, placebo-
• Beta-carotene group (20 mg/day) controlled primary prevention
– increase in total mortality (9%) trial
– increased incidence of angina (13%)* • Participants: 18,314 smokers, former smokers, and
workers exposed to asbestos
– increased CVD mortality (11%)*
• Intervention: 1) B-carotene (30 mg/day) and
– increased incidence of lung cancer (18%)
vitamin A (25,000 IU/day)
2) Placebo
• Vitamin E Group (50 mg/day) • Follow-up: 4 years
– reduction in total coronary events (3%) • End Points: Incident lung cancer
– reduction in incident angina (9%) Cardiovascular Disease
– reduction in non-fatal MI (11%)
Omenn, 1996 NEJM

ATBC, 1994 NEJM

Failed surrogate marker: example

β- carotene
supplements

+
↑β-carotene ↑Lung
Smoking
_ Cancer
↓β-carotene

Need for reliable surrogate markers

 Disadvantages of Surrogate
The Bottom Line
Outcomes (continued)
• The relationship between a surrogate “Trust but verify”
outcome and a clinical outcome has face
validity but is often uncertain
Ronald Reagan
• Relationship between change in surrogate
and risk of clinical outcomes is rarely known

93

Analytic Techniques:
Analytical Issues
Crude analyses
• Sample Size (Power Calculations)
• Analysis depends on the type of outcome data
• Basic tests
• Analytical Approach (a priori)
– Continuous outcome variable:t-test
• Examples: Blood pressure, serum cholesterol
• Intention-to-treat (vs ‘as treated’) – Dichotomous or categorical data: chi-squared,
logistic regression, cox modeling for time to
event
• Example: Incident HIV, MI, cancer, renal failure, death
95 96
 Analytic Techniques:
Epidemiology in a box: The 2x2 table
Adjusted (Regression) Analyses
D+ D- • Regression determines association between
• The EXPOSURE (E) exposure and outcome
– Example: obesity E+ a+b
a b • Procedures depends on outcome variable:
• The OUTCOME (D)
– Continuous outcome: linear regression
– Example: Hypertension E-
c d c+d – Dichotomous outcome: logistic regression
• Applicable to most
study designs – Time-to-event: Cox proportional hazards
a+c b+d Total

97 98

Interpretation of Results Why RCTs Can Be Difficult

• Hard to find and recruit the right people
• Internal Validity – Many don’t want to be “guinea pigs”
– conclusions correctly describe what • Greater responsibility, documentation
happened in the study
• May take years for outcomes to develop
• External Validity (‘generalizability’) • People are free to do as they please
– Some assigned to treatment don’t adhere
– the degree to which the conclusions apply
to the study population and other – Some assigned to control seek treatment
populations – Some drop out of the trial completely
99
 Adherence (compliance) Why be worried about adherence?

• Difficult to measure Active

Drop-In’s
• Difficult to promote
Drop-Out’s

• Must be promoted and measured, at

Control
least in efficacy or explanatory trials
Intention-to-Treat: analysis by randomized group,
101 not by final groupings 102

Adherence (compliance) How To Handle Participants Who

• Measurement Don’t Adhere to Trial Assignment
– self report • Intention-to-Treat Approach
– pill count – Least optimistic
– blood levels of drug – Maintains initial balance from randomization
– biological changes (urine or blood) – Highlights problems from adverse effects
• On-Treatment Approach
• Promotion
– Most optimistic
– exclude poor candidates before randomization – Upsets initial balance from randomization
– keep intervention simple – Downplays problems from adverse effects
– respond to evidence of inadequate adherence
103 Because of its conservatism, the Intention-to-Treat approach is strongly preferred.
 Cardiac Event-Free Survival in 192 Adults with Cardiac Event-Free Survival in 192 Adults with
Refractory Angina by Random Assignment and Refractory Angina by Random Assignment and
Cross-Over (from Medical Treatment to TMR) Status Cross-Over (from Medical Treatment to TMR) Status

Randomized to
TMR, no crossing
over to Medical Rx Were X-overs
reclassified
Randomized to as “TMR”, it
Medical Rx, did would tend to
poorly, needed TMR
make TMR
as last ditch effort
look worse
Randomized to
Medical Rx, did OK,
no need for TMR

TMR =transmyocardial laser revascularization

Cardiac Event-Free Survival in 192 Adults with

Refractory Angina by Random Assignment and
Cross-Over (from Medical Treatment to TMR) Status

Clinical Trials: Design and

interpretation Considerations
Were X-overs
classified as
“Medical Rx”,
it would tend
to make
Medical Rx
look better 108
 When Trials Are Impossible Statistical vs Clinical
(or Nearly Impossible) Significance
• Statistical significance pertains to
• Adverse Exposures (e.g. Cigarettes)
whether or not the observed results
• Rare Outcomes (e.g. Reye’s Syndrome) could occur from chance alone
• Intervention Already in Wide Use

In these circumstances, one must rely on observational studies—i.e.

• Clinical significance pertains to whether
prospective cohort studies and case-control studies. When interventions are
already in wide use, “outcomes research” is a good option. In outcomes
or not the observed results have
research, medical interventions (e.g. drugs, surgical procedures) are “important” clinical, research or public
considered as exposures. Data on these interventions, and on relevant clinical
outcomes, are available from medical records and often from large-scale health relevance.
electronic databases.
110

How To Interpret Negative Results Efficacy (Explanatory) Trial

vs Effectiveness (Pragmatic) Trial
• Treatment is worthless
• Treatment is worthwhile, BUT study had… • Theory
– Bias against the treatment (e.g. crossing in) – Efficacy: What is the effect of the therapy
under ideal conditions
– Inadequate contrast between groups
• Suboptimal treatment (e.g. unskilled surgeons)
– Effectiveness: What is the effect of therapy
under ‘real world’ conditions
• Low adherence (e.g. drug causes GI distress)
• Controls sought treatment despite assignment • Reality
– Insufficient statistical power – The dichotomy between efficacy and
effectiveness is artificial
• Very common cause of negative findings
– Broad continuum
• Meta-analysis a potential remedy
112
 Typical Implementation Units
• Clinical Centers
Oversight Units
– recruit participants
– collect data • Internal
– administer intervention/therapy – Sponsor
• Laboratory or Reading Centers – Data Coordinating Center or Contract
– perform assays or readings of procedures Research Organization
• Data Coordinating Center* • External
– receive/assemble data – Institutional Review Board
– coordinate activities – Data and Safety Monitoring Board
– perform data analyses
* similar to Contract Research Organization (CRO) 113 114

Organizational Structure of
a Multi-Center Trial
(Weight Loss Maintenance Trial)
Steering Committee NIH Project Office DSMB

Subcommittees Clinical Centers Coordinating Center

Center for Health Johns Hopkins

Design & Analysis Enrollment and Research University
Retention
Data Intervention
Management
Development
Publications
Intervention Pennington LSU Duke University

Measurement & Quality

Control Minority
Implementation

Clinic Coordinators

115