Title: Randomized Trials: Design, Subjects, and Randomization
1Randomized Trials Design, Subjects, and
Randomization
- Clay Johnston, MD, PhD
- Neurology and Epidemiology
2Randomized Trials the Evidence in
Evidence-Based
- Today
- Randomized trials why bother?
- Randomization
- Adaptive designs
- Selection of participants (Inclusion/exclusion)
- Design options for trials
- Factorial designs
- Cross-over designs
- Matched pairs
- Cluster or group randomization
3Randomized Controlled Trial (RCT)
- An experiment in which subjects are randomly
allocated into groups, usually called study and
control groups, to receive or not to receive an
experimental preventive or therapeutic procedure,
maneuver, or intervention. The results are
assessed by rigorous comparison of rates of
disease, death, recovery, or other appropriate
outcome in the study and control groups,
respectively. -
4Number of randomized trials published
Based on Medline search restricted to
Randomized clinical trials
5Disadvantages of RCTs
- Expensive typically in millions
- Time Consuming typically years
- Can only answer a single question
- May not apply to most patients in practice
- May not be practical
- Generally very difficult to get funded
- Time consuming, organizationally complex
- Class dismissed.
6Alternatives to RCTs(30 second Epi. Course)
- Case-control studies
- Compare those with and without disease
- Cross-sectional studies
- Compare rates of risk factor among those with
and without disease at a single time point. - Cohort studies (prospective)
- Identify those with and without risk factor
- Follow forward in time to see who gets
disease - Case-control, cross-sectional, and cohort studies
are observational (not experimental)
7Reasons for doing RCTs
- Only study design that can prove causation
- Rodney Dangerfield syndrome for observational
researchers - Required by FDA (and others) for new drugs and
some devices - Most influential to clinical practice
8Example Estrogen Replacement Therapyin
post-menopausal women
- Important therapeutic question
- Applies to 30 million women in US
- Prempro (estrogen/progestin combo)may have been
most prescribed drug in US - Potentially huge impact on public health
- Complex ERT effects multiple diseases
9Estrogen Replacement Therapy (ERT)
- Disease Effect on Risk
- Coronary heart disease Decrease by 40 -
80Osteoporosis (hip fx) Decrease by 30 -
60Breast cancer Increase by 10 -
20Endometrial cancer Increase by 700 - Alzheimers Decrease by ?
- Pulmonary embolism Increase by 200 - 300deep
vein thrombosis
From observational (case-control and cohort)
studies
10Nurses Health Study (NEJM, 9/12/91)
- Prospective cohort study, n 48,470
- 337,000 person years of follow-up
- Risk of Major Estrogen Use Coronary
Disease Relative Risk - Never Used 1.4 1.0
- Current user 0.6 0.56 (0.40-0.80)
- Former user 1.3 0.83 (0.65-1.05)
Events per 1000 women-years of follow-up
Relative Risk (95 CI) compared to never users
11Meta-analysis of ERT, Published 4/10/97
- Benefits (for CHD, osteoporosis) outweigh risks
(breast cancer) and side effectsAll
post-menopausal women should be taking ERT
CNN, 4/10/97
12Virtually all estrogen results werebased on
observational data
- Women chose to take ERT
- Are ERT users different from non-users?
- Age
- Health status
- More exercise
- Health behaviors (see Dr.)
- SES
- Try to adjust in analysis, but may not be
possible - Randomized trials alleviate these problems
13Heart and Estrogen-Progestin Replacement Study
(HERS)
- Secondary prevention of heart disease
- HRT (Prempro) vs. placebo (4-5 years)
- 2763 women with established heart disease
- Postmenopausal, lt 80 years, mean age 67
- 20 clinical centers in U.S./UCSF Coordinating
center - Funding by Wyeth-Ayerst (post-NIH refusal)
- Results JAMA 8/98
14HERS Summary of results
- Endpoint Placebo HRT RR P
- New CHD 176 172 0.99 0.91
- Any fracture 138 130 0.95 0.70
- Conclusion Randomized trials can lead to big
surprises!
15Womens Health Initiative HRT study (7/10/02)
- Randomized trial (2)
- 16,608 women with uterus (ERT progestin vs.
placebo) - 11,000 women without uterus (ERT alone vs.
placebo) - Ages 50-79, mean age 64
- Represent broad range of U.S. women
- 40 clinical centers
16WHI EP study 7/10/02
- Combination therapy arm stopped early (3 years)
- Mean 5.2 years of follow-up
- Overall, health risks outweigh benefits
- Significant increased risk for invasive breast
cancer HRT users
17WHI E P Coronary Heart Disease
years 1 2 3
4 5 6
18HERS/WHI Trials Take Home
- Observational studies can be wrong.
- Cohort studies can be wrong
- Meta-analysis of observational studies can be
wrong. - What went wrong with observational studies of HRT?
19Major Observational Study Limitation
- CHARM
- Candesartan in Heart failure Assessment of
Reduction in Mortality and morbidity - Compared those taking vs. not taking either drug
or placebo - 35 RRR in all-cause mortality
- For drug and placebo.
- Take home
- Confounders are impossible to fully identify
Lancet 2005, 3662005
20CLINICAL TRIALS IN THE NEWS
- Lots of trials of things other than drugs
- Surgical techniques
- Weight loss
21Clinical Trials in the NewsJAMA 1/5/05
Vs.
Vs.
22RCT of 4 Popular Weight Loss Programs
Vs.
- Compare
- Atkins (low carbohydrate)
- Weight Watchers (low calorie/portion size)
- Zone (high protein/low-glycemic load)
- Ornish (very low fat)
JAMA 1/5/05
23Diet study Design
- N 160
- Randomize to 1 of 4 diets
- Follow for 12 months
- Endpoints
- Weight loss
- Heart disease risk factors (cholesterol, BP,
triglycerides)
JAMA 1/5/05
24Diet study Results at 12 months
- Year Atkins Zone Weight Ornish
- watchers .
- Weight (kg) -3.9 -4.9 -4.6 -6.6
- LDL (mg/dL) -13.5 -18.1 -14.2 -25.2
- SBP (mm/Hg) 0.3 2.1 -4.1 0.9
JAMA 1/5/05
25Diet study Summary
- All diets lead to modest reductions in weight
and cardiac risk factors - Poor compliance for all diets, especially Atkins
and Ornish - Those who adhered well had better results
JAMA 1/5/05
26Examples of major breakthroughs from RCTs
- Protease inhibitors and AIDS
- Aspirin and heart disease
- Lipid lowering (statins) and heart disease
27NINDS Trials
28So you want to do a randomized trial
29Steps in a Classical Randomized, Controlled
Trail (RCT)
- 1. Select participants
- 2. Measure baseline variables
- 3. Randomize (to 1 or more treatments)
- 4. Apply intervention
- 5/6. Follow-up--measure outcomes
- Most commonly one treatment vs. control
- Can be used for various types of outcomes
(binary, continuous)
30Randomization
- Key element of RCTs
- Assures equal distribution of both...
- Measured/known confounders
- Unmeasured/unknown confounders
- Important to do well
- True random allocation
- Tamper-proof (no peeking, altering order of
participants, etc) - Simple randomization
- Low tech
- High tech
31Other types of randomization
- Blocking equal after each n assignments
- e.g., block size of 4, treatments a and b
- abab aabb abba baba bbaa baab
- Randomly choose blocks
- Assure relatively equal number of ppts. to each
treatment - Disadvantages of blocking (in unblinded trials)
- Size of block 2 treatments--4 or 6
- Very commonly used
- Formally random, permuted blocks
32Randomized blocks to balance prognostic variables
- Stratified permuted blocks
- Blocks within strata of prognostic variable
- e.g., Stroke prevention after TIA. Time from
event a key predictor - Stratum
- lt12 hour aabb baba
- gt12 hour baab abab .
- Limited number of risk factors
- Very commonly used in multicenter studies to
balance within clinical center - Fancier techniques for assuring balance
- Adaptive randomization
33Adaptive Randomization
- Various designs that reduce the total number of
subjects necessary when comparing multiple groups - Determine assignment on the fly based on prior
data - Often used for dose finding
- Simple designs based on a priori decision rules
- Eg, go to next highest dose if no side effects in
first 4 treated - Complex designs based on recalculation of odds
- ASTIN trial of 15 doses of neutrophil inhibitory
factor (Stroke, 2003)
34(No Transcript)
35Implementation of randomization
- Less challenging for blinded studies
- Sealed envelopes in fixed order at clinical sites
- List of drug numbers
- a b a b b b a a
- 1 2 3 4 5 6 7 8
- Clinic receives bottles labeled only by
numbers--assign in order - IVRS Interactive Voice Response System
- Unblinded studies important to keep next
assignment secret - Problem with blocks within strata
36Randomization Summary
- Key element of clinical trials
- Not really very complicated (usually)
37Who to Study Principles for Inclusion/exclusion
- Widest possible generalizability
- Sufficiently high event rate (for power to be
adequate) - Population in whom intervention likely to be
effective and safe - Ease of recruitment
- Likelihood of compliance with treatment and F/U
38Who to Study Principles for Inclusion/exclusion
- Homogeneity --------------------Heterogeneity
39Explicit criteria for inclusion in a trial
- Typically written as inclusion/exclusion
criteria in protocol - Generally, the more explicit the better
- Want centers or investigators to be consistent
- Examples of exclusion criteria decisions
- 1. Women with heart disease vs. Women with CABG
surgery or documented MI by ECG (criteria) or
enzymes (criteria) - 2. Users of estrogen vs Use of oral ERT (.625 mg
prempro) for more than 3 months over last 24 mos.
40Valid reasons to exclude participants
- Treatment would be unsafe
- Adverse experience from active treatment
- Risk of placebo
- Active treatment cannot/unlikely to be effective
- No risk of outcome
- Disease type unlikely to respond
- Competing/interfering treatment (history of?)
- Unlikely to adhere or follow-up
- Practical problems
41Enrollment rates in stroke trials
42Example of Inclusions/Exclusions in Protocol
- Inclusion Criteria
- Male or female gt 40 years
- TIA (gt 10 minutes) or minor acute ischemic stroke
(NIHSS lt 2 at time of randomization) occurring
less than 24 hours before randomization - Informed consent signed
- Able to have MRI scan within 24 hours of symptoms
onset - Exclusion Criteria
- Related to absolute contraindications to the use
of clopidogrel and/or ASA - History of drug allergy to thienopyridine
derivatives or ASA - Severe uncontrolled hypertension (SBP gt 160 mm Hg
or DBP gt 110 mm Hg on two or more measures over
the last 6 months) - History of clinically significant or persistent
thrompocytopenia - History of clinically significant or persistent
neutropenia - Women of child-bearing potential who are not
following an effective method of contraception - Women who are breast-feeding
43Example of Inclusions/Exclusions in Protocol
- Inclusion Criteria
- Male or female gt 40 years
- TIA (gt 10 minutes) or minor acute ischemic stroke
(NIHSS lt 2 at time of randomization) occurring
less than 24 hours before randomization - Informed consent signed
- Able to have MRI scan within 24 hours of symptoms
onset - Exclusion Criteria
- Related to absolute contraindications to the use
of clopidogrel and/or ASA - Related to concomitant or planned medication(s) /
treatment(s) - Related to TIA/Stroke characteristics
- Related to the presence of other medical problems
that would either interfere with participation in
the trial or lead to inability to complete the
trial
44Design-a-trial Inclusion criteria options for
HRT
- Study HRT and prevention of heart disease, 4
years (HERS-like) - Women over age 50 years
- Women over 60 years
- Women over 75 years
- Women with existing heart disease
- Generalizability?
- Feasible sample size?
- Population amenable to intervention?
- Logistic difficulties (recruitment? cost?
adherence)
45HERS inclusion options
- HERS trial options (event rate)
- Women over age 50 years (0.1/year)
- Women over 60 years (0.5/year)
- Women over 75 years (1/year)
- Women with existing heart disease (4/year)
46HERS inclusion options
- HERS trial options (event rate) n required
- Women over age 50 years (0.1/year) 55,000
- Women over 60 years (0.5/year) 45,000
- Women over 75 years (1/year) 34,000
- Women with existing heart disease (4/year)
3,000 - (Choose last option as most practical common to
generalize from secondary to primary prevention)
47Exclusions/inclusions examples
- Important impact on generalizability of both
efficacy and safety - Example Primary Stroke Prevention Trial
- Chlorthalidone vs. placebo
- Stroke rates dramatically increase with age
- Stroke more common than MI after age 65
- Who should we include?
- Age cut-off?
- Atrial fibrillation?
48Inclusion, Exclusion Conclusion
- Many factors to balance in deciding who to
include - Generally not a clear cut or single correct
decision - Many academics have simplistic understanding of
issues
49Alternative RCT designs Large-simple vs.
standard
- Large simple less information on more people
- Balance loss of precision with sample size
- Cost per patient with number of patients
- Example
- Primary Prevention of Stroke Trial
- COMMIT 45,852 patients in China, adding
clopidogrel to aspirin in acute MI - Standard lots of info on few
- Example CASTIA trial
50Alternative RCT designs Factorial design
- Test of more than one treatment (vs. placebo)
- Each drug alone and in combination
- Allows multiple hypotheses in single trial
- Efficient
- Example Physicians Health Study
- Test aspirin gt MI
- beta carotene gt cancer
51Factorial design Physicians Heath Study
Placebo
Beta-carotene
Aspirin vs. no aspirin (MI)
Aspirin plus Beta-carotene
Aspirin
Beta carotene vs. no beta carotene (cancer)
52Music, imagery, touch, and prayer as adjuncts to
interventional cardiac care the Monitoring and
Actualisation of Noetic Trainings (MANTRA) II
randomised study
53Factorial design Primary Stroke Prevention Trial
Placebo
Chlorthal.
Chlorthal. plus K Citr
K Citr
543-way factorial design of WHI
HRT vs. no HRT
Calcium vs. no calcium
Low fat vs. regular diet
55Factorial design issues
- Do treatments interact?
- Effect of chlorthalidone likely greater with
potassium citrate on board - HRT may increase calcium effect on bone
- Must test for interaction of treatments
- If present, power is lower large sample required
- May require more complicated stopping rules
- May require more complicated analysis plan (eg,
logistic regression)
56A Factorial Trial of Six Interventions for the
Prevention of Postoperative Nausea and Vomiting
- One third of patients have post-op N/V
- 5200 patients randomized to test 6 individual
medications - 26 64 combinations
- Ondansetron
- Dexamethosone
- Droperidol
- Propofol vs. other
- Nitrogen vs. nitrous oxide
- Remifentanil vs. fentanyl
Apfel, C. C. et al. N Engl J Med
20043502441-2451
57Study Design
Apfel, C. C. et al. N Engl J Med
20043502441-2451
58A Factorial Trial of Six Interventions for the
Prevention of Postoperative Nausea and Vomiting
- Results
- Ondensetron, dexamethasone, and droperidol each
reduced N/V by 26 - Propofol by 19
- Nitrogen by 12
- All agents acted independently
- RR associated with combined interventions could
be estimated by multiplying individual RRs.
Apfel, C. C. et al. N Engl J Med
20043502441-2451
59Factorial design conclusions
- Factorial designs are seductive but complicated
- Some attraction in combining a low-risk
hypothesis with a high-risk hypothesis - Must weigh benefits in efficiency against
compounded uncertainty and complexity
60Cross-over designs
- Both treatments are administered sequentially to
all subjects - Subject serves as own control, random order
- Compare treatment period vs. control period
- Increases power by reducing person-to-person
variability
61Cross-over designs
- Diuretic vs. beta blocker for blood pressure
- 1/2 get d followed by bb
- 1/2 get bb followed by d
- Migraine prophylaxis
- Rx x 3 months followed by placebo
- Placebo x 3 months followed by Rx
- Atrial Overdrive Pacing in sleep apnea
- AOP x 1 month followed by low-rate pacing
- Low-rate pacing x 1 month followed by AOP
62Cross-over assumptions/limitations
- Transient outcomes only
- No order effects
- No carry-over effects
- Need quick response and quick resolution
- Wash out period helpful
- More commonly used in phase I/II
63Matched Pair Randomization
- One of each pair to each treatment
- Reduces risk of imbalanced randomization
- Allows smaller sample size
64Matched Pair Randomization
- Example two eyes within an individual (one to
each treatment) - Diabetic Retinopathy study
- Example Filter for carotid arteries in patients
with atrial fibrillation - Random side of intervention
- Example Quality Improvement in Stroke
Prevention (QUISP) trial - Randomize paired Kaiser facilities
65Matched Pair Randomization Issues
- Not necessary when sample size makes balanced
randomization likely - Loss in efficiency
- May not be feasible
- Arguments about how to analyze the data
- Maintaining the match vs. breaking it
66Cluster or grouped randomization
- Randomize groups to treatments
- Often useful especially for public health-type
interventions - May be only way to study a question
- Intervention is at the group level
- Cross-contamination between individuals
67Cluster or grouped randomization
- Examples
- Quality Improvement in Stroke Prevention (QUISP)
trial - DTC Advertising Trial
- Randomize matched cities
- Cities to public health risk factor reduction (5
Cities Project)
68Cluster or grouped randomization
- Sample size complex true n is between n
clusters and n individuals (closer to clusters) - Tendency to underestimate necessary sample size
- Analysis complex
- Must account for clustering
- Mixed models best
69Randomized Trials the Evidence in
Evidence-Based
- Today
- Randomized trials why bother?
- Randomization
- Adaptive designs
- Selection of participants (Inclusion/exclusion)
- Design options for trials
- Factorial designs
- Cross-over designs
- Matched pairs
- Cluster or group randomization
70Previews of coming attractions
- Statistical issues in randomized trials