1 Page summary
Gsanc1
Week6_2_Spring20202.ppsRandomized Trials (RCTs) and Alternative Clinical Trial Designs
Week # 6
Chapters 10 & 11
Clinical Trials
- Generally, investigator applies an intervention and observes the effect on one or more outcomes.
Key Attributes
- Ability to demonstrate causality
- Random assignment minimizes the influence of confounding variables
- Blinding ensures that the apparent effects of the intervention are not due to differential use of other treatment in the intervention and control groups or to biased measurement of the outcomes
Definition
- Randomized comparison of two or more groups to evaluate medical interventions
- Interventions can be:
Drugs
Surgeries
Devices (e.g., pacemaker)
Programs (e.g., screening for cancer, telephone nursing hotline)
Definition
- RCTs are conducted for any of three reasons
To evaluate a new intervention before it is given regulatory approval
To gain regulatory approval for a new intervention
To evaluate interventions that are controversial or that are widely used without adequate evidence
Definition
- In the regulatory process leading to new drug approval, the word ‘trial’ can be applied to many designs, including RCTs
- Phase I trial: small study of 20-80 healthy volunteers – all get the new drug – purpose is to assess toxicity and pharmacologic effects
Definition
- Phase II trial: study of 100-200 persons with disease of interest – all get the drug – purpose is to assess safety and efficacy
- Phase III trial: RCT (required for regulatory approval)
Definition
- Phase IV trial: continuation of follow-up of RCT subjects past the ‘official’ end of the RCT
Sometimes an observational study using medical databases where thousands of persons taking the intervention are followed for years
Useful for detecting rare side effects
Definition
- RCTs can be single- or multi-centre
Single-centre: all patients are recruited from the same clinic or hospital
Multi-centre: patients are recruited from more than one clinic or hospital
Often necessary to recruit enough patients to meet sample size requirements
May span several countries
Definition
- Comparative nature of RCTs is important!
We cannot judge an intervention unless we compare it to something else
A new surgical procedure might cure stress urinary incontinence in 50% of patients
However, is 50% good?
The question cannot be answered without a comparison group (e.g., standard surgery)
Definition
- Same issue with case-control and cohort studies
- Testing a hypothesis requires a comparison group
Definition
- Typical evaluation in an RCT:
New versus standard intervention
- Drug trials:
New drug versus placebo
Problem: new drug may be better than placebo, but is it better than the standard drug or a competing new drug?
Definition
- New Drug Versus Placebo
US FDA will approve a drug for public consumption if it is shown to be statistically significantly better than placebo
Other countries follow the FDA’s lead
Pharma industry likes these regulations because drugs can get approval even if a competitor’s product is better!
Definition
- New Drug Versus Placebo
Researchers are aware of the problems with placebo-controlled drug trials, but most drug trials are funded by industry
Most drug trials are therefore placebo-controlled
What other role(s) can placebo-controlled trials play?
Hint: Safety consideration?
Definition
- New Drug Versus Placebo
Placebo-controlled trials do have a role to play in research
They help assess the severity of side effects
If 34% of people taking aspirin have gastrointestinal side effects, we cannot know whether this is a large or small percentage until we compare it to the percentage in people not taking aspirin
Definition
- Ideally, evidence for the efficacy of a medication, and regulatory approval, should be based on placebo-controlled AND new versus standard intervention studies
Design
Begin with a defined population (e.g., persons who are depressed)
Randomize these persons to intervention groups
New anti-depressant medication
Standard anti-depressant medication
OR
Placebo
Design
- Can have more than two groups
e.g., new drug, standard drug A, standard drug B
- Follow-up is for a pre-defined period of time (e.g., 24 weeks)
- Compare groups at end of follow-up
- Comparisons may also be done at pre-defined points during follow-up
Example - Comparison
- New anti-depressant drug A
- Standard anti-depressant drug B
- 200 depressed people randomized to treatment
- Half get drug A and half get drug B
Example - Comparison
- After 24 weeks, the mean scores on the Hamilton Depression Scale were:
16 for the group that got drug A
19 for the group that got drug B
p < 0.05 (difference was statistically significant)
Higher scores mean greater depression
People who got new drug A were less depressed than people who got standard drug B
Randomization
- The strength of the result lies in randomization
- Randomization = random allocation of study subjects to study groups
- Randomization is typically performed by a statistician using a computer program
Randomization
- If randomization is done correctly, then…
Each subject has an equal chance of being assigned to either group
A roughly equal number of subjects will be assigned to each group
The subjects will be similar to one another on all factors except the intervention to which they are assigned
Randomization
- This ‘similarity’ between groups helps ensure that the only factor influencing the comparison is the intervention to which subjects are assigned
- For this reason, RCTs are considered the ‘gold standard’ of medical research
Stratified Randomization
- Ensure that variables considered important to prognosis are equally distributed amongst the intervention groups
- If prognosis is worse in older males, then extra measures should be taken to distribute age and sex equally between treatment groups
Stratified Randomization
- Divide (stratify) the study population by age and sex and then randomize within each grouping
Stratified Randomization
Study Population = 1,000
Males = 600
Females = 400
Old = 240
Young = 300
Old = 100
Young = 360
Treatment
A
n = 500
Treatment
B
n = 500
Stratify by sex:
Stratify by age:
Randomize within each subgroup
Blinding
- Not allowing study subjects to know what treatment they received
- Important because knowledge of treatment could affect patient response (esp. for subjective outcome measures such as pain)
Blinding
- e.g., patients receiving a new treatment for headache might think ‘new’ means ‘better’
If they know they are taking the new treatment, then they may report greater pain relief than what is really the case
Blinding
- How to blind
Use identical-looking placebo as comparison
In new drug versus standard drug comparisons, ask manufacturers to make the drugs identical
Not likely to happen (cost, corporate pride, inability due to pharmacological reasons)
Blinding
- ‘Double-blind’
Blinding of data collectors and data analysts
Done to prevent knowledge of treatment from influencing how data are collected or analyzed
- ‘Triple-blind’
Blinding physicians and hospital staff who treat study subjects
Blinding
- Blinding of patients and physicians is not always possible when there are obvious differences between interventions
Surgery versus collagen injection to treat female stress urinary incontinence
Laparoscopy versus abdominal surgery to remove ovarian cysts
Institutionalized versus home care for the elderly
Use of walkers versus scooters for persons with mobility impairment
- Try and blind data assessors and analysts
Blinding
- Does inability to blind preclude the use of an RCT?
- NO
Look for problems (quality control)
Potential problems:
Data collection is more rigorous for patients in one group versus another
Healthier patients disproportionately receive one intervention versus another
Other RCT Designs
- Planned crossover
Patients are randomized to receive intervention A or B
After a pre-defined period of time, they are switched to the other intervention
Advantages:
Patients serve as their own controls
Sample size is smaller
Other RCT Designs
- Planned crossover
Cautions:
Washout period: the time between discontinuance of the first intervention and start of the second intervention must be long enough to eliminate any carry over effects from the first intervention
Ordering effect: patients may react differently to the first intervention because of the psychological effect of being studied
Are there situations where planned crossovers are not possible?
Other RCT Designs
- Planned crossover
Not possible for surgical interventions or interventions that cure the disease
Other RCT Designs
- Factorial design
Economical means of using the same population to test two different drugs
Drugs must have…
Different outcomes
Independent modes of action
Otherwise, drug interactions would prevent the independent study of the effects of each drug
Other RCT Designs
Randomized
Aspirin
Placebo
Β-carotene
Placebo
Placebo
Β-carotene
1st comparison – aspirin versus placebo
2nd comparison – B-carotene vs. placebo
Factorial
Problems
- Unintended crossovers
Some subjects receive the intervention to which they were NOT randomized
In the surgery versus collagen trial, some patients randomized to collagen later received surgery
Problems
- Unintended crossovers
What to do?
To preserve randomization, analyze patients according to the group to which they were originally randomized
Intent-to-treat analysis
Analyze patients according to the treatment actually received
Hope the results of both analyses are similar
If not, then the study may be biased
Problems
- Noncompliance (AKA adherence)
Occurs when study subjects do not comply with the treatment to which they were randomized
Subjects may refuse the treatment and leave the study (drop outs)
Subjects may stop the treatment without telling the investigators (a problem in drug studies)
Subjects in one group may accidentally take the drug assigned to the other group on one or more occasions
Problems
- Noncompliance
Could reduce the size of observed treatment effects in the study sample
Could make the study groups appear more alike than they really are
Problems
- Noncompliance
Address noncompliance through…
Pill counts
Urine or other tests
Removal of noncompliers from the study sample before randomization (pilot study)
Issues with RCTs
- Internal versus external validity
An RCT has internal validity when it is conducted in a methodologically sound manner
Proper randomization
Blinding (if possible)
Steps taken to reduce unplanned crossovers and noncompliance
Correct analysis of results
Issues with RCTs
- Internal versus external validity
An RCT has external validity when its results can be generalized to populations other than the study population
Issues with RCTs
- RCTs are conducted to assess whether an intervention CAN work
- Investigators are therefore concerned with internal validity
To minimize drop outs and potential drug interactions, study subjects chosen to participate in RCTs often have few comorbid disorders
Issues with RCTs
Study subjects may also be chosen only if they are good compliers (e.g., AIDS drug cocktail trials)
- Thus, study subjects may be atypical of the population that would normally receive the intervention
Issues with RCTs
- This means many RCTs are not generalizable (no external validity)
They do not address the question of effectiveness (i.e., does an intervention work in the average patient)
Issues with RCTs
- Ethics
Is it ethical to withhold a new intervention from patients (randomization means some patients will not get the new intervention)?
Yes, if the medical community is generally uncertain about whether the new intervention is better than the standard intervention
Issues with RCTs
- Ethics
Promising evidence from phase I or II studies, or observational studies, may suggest that the intervention is better, but the issue will have to be resolved with an RCT
Individual physicians may strongly favour (or dislike) the new intervention, but the RCT would be ethical as long as the medical community is uncertain about whether the new intervention is better
Issues with RCTs
- Ethics
Clinical equipoise (Freedman, 1987): only the “expert” medical community needs to be uncertain
Problems
Who are the experts?
How can ‘uncertainty’ be quantified?
Issues with RCTs
- Length of follow-up
Often short (e.g., 12 weeks, 24 weeks) due to high cost of recruitment and follow-up
Short follow-ups may be inadequate to study long-term outcomes for chronic conditions
Mean survival following dementia diagnosis = 4 years
12-week trial of a dementia drug does not indicate how efficacious the drug will be after three years of use
Issues with RCTs
- Length of follow-up
Short follow-ups may be inadequate to study rare side effects
Even with 500+ patients, years of follow-up would be required detect rare side effects
Highlights importance of Phase IV studies
RCTs vs. Observational Studies
- Both compare ‘exposed’ versus ‘unexposed’ groups
RCT: active treatment group vs. placebo group
- Obs. Studies: subjects NOT randomized
Subjects are assessed for exposure status and assigned to the appropriate group (assignment based on observation)
Investigator does not assign the exposure
Why can’t we use RCTs all the time if they are the gold standard?
o Ethics unethical to randomize persons to an exposure known to be harmful
● e.g., cannot take a group of non-smokers and assign them
to a group where they will be required to start smoking
o RCTs have relatively short follow-ups (expensive to conduct, designed for regulatory approval) difficult to study rare outcomes or adverse effects (Vioxx)
o Gold standard research design, but not gold standard for answering every question
● Poor generalizability
● Cannot randomize everything (e.g., SES)
