Randomized Controlled Trials: The Basics

1
Randomized Controlled Trials The Basics
Dr. PadmaMohan. J
2

• Therapeutic reports with controls tend to have
  no enthusiasm, and reports with enthusiasm tend
  to have no controls.
• David L Sackett, 1986
• Chest 1986 Feb 89(2 Suppl) 2s-3s

3
Contents

• Introduction
• History
• Randomization
• The Use of Blinding and Placebos
• Size of Trials
• Monitoring Trial Results
• Reporting Trial Results
• Alternative Designs

4
Clinical Trials-Types

• Phase 1 Done to assess drug safety and dosage,
  in volunteers.
• Phase 2 Small scale investigations for
  effectiveness and safety (initial clinical
  investigation for effect).
• Phase 3 The rigorous full scale RCT
• Phase 4 Post marketing surveillance- this has
  limited scientific value.

5
What is RCT?

• A randomized controlled trial (RCT) is a type of
  scientific experiment most commonly used in
  testing the efficacy or effectiveness of
  healthcare services (such as medicine or nursing)
  or health technologies (such as pharmaceuticals,
  medical devices or surgery).
• A well-designed trial provides the most rigorous
  method for evaluating treatments.

6
Settings for RCTs

• There are many different types of treatment or
  intervention that can be evaluated with a
  clinical trial.
• Drug therapy
• Medical procedures
• Educational intervention
• Screening trial
• Vaccine trial

7
RCT-Types

• The primary aim of RCT is to provide reliable
  evidence of treatment efficacy (or effectiveness)
  and safety.
• Efficacy of an intervention is the benefit it
  achieves under ideal conditions (for example,
  when all patients take the treatment as intended)
  a trial which focuses on measuring efficacy is
  often known as an explanatory trial.
• Effectiveness of an intervention is the benefit
  it achieves when delivered through routine
  clinical practice a trial which focuses on
  measuring effectiveness is often known as a
  pragmatic trial

8
Essential Design Features

• Controlled Any intervention needs to be compared
  to one or more other interventions. This is often
  a new treatment compared to standard treatment.
  In drug therapy trials there will often be a
  placebo control.
• Unbiased There needs to be a fair comparison
  between the treatments with no bias, whether
  deliberate or accidental. Randomization is
  crucial patients are randomly allocated to a
  particular treatment group.
• Large Patients vary considerably in their
  response to treatment. In order to obtain a
  precise estimate of any treatment effect,
  sufficiently large numbers are required.

9
History (before 1950)

• James Lind
• In 1747, surgeon on HM Bark Salisbury, James Lind
  did trial on scurvy.
• On 12 persons he gave various food supplements
  like cider, garlic,fresh lime and oranges,vinegar
  etc
• Scurvy was cured in those who took citrus .
• Subjects' cases "were as similar as I could have
  them"

10
History (before 1950)

• Louis 1834 founding figure for clinical trials
  with scientific footing- he applied numerical
  method for comparison.
• Sutton (1865) First experiment where a placebo
  was used. (in Rheumatic fever giving just mint
  water).
• Similar reports Use of sulphonamides
  Antimalarials (1930), Penicillin (1941)
• First trial with randomized control was
  streptomycin trial for TB in 1948

11
History (after 1950)

• Sir Austin Bradford Hill (1962) pioneered in the
  field and was prime motivator for Medical
  Research Council.
• Field trial of Salk Polio Vaccine 1954
• University group diabetes program-1964
• Several cancer chemotherapy trials
• Cardiovascular drug trials etc

12
RCT Components

• RCTs ..help to minimize bias and the
  possibility that any factor, other than the
  therapy or therapies intervention under
  investigation, account for differences in
  outcome between the control group and the test
  group or groups.
• Susan Bressler Neil Bressler
• American Journal of Ophthalmology 2001 April
  131(4) 503-4

13
Bias

• Nonrandom (systematic)error that distorts
  estimates of treatment effects
• Selection effect
• Observer bias
• Bias reduction refers to balancing out the
  unknown prognostic factors Achieving balance
  refers to balancing the known prognostic factors

Has a net direction and magnitude
14
Unknown Sources of Bias

• Subjects mental/physical condition
• Substance use
• Childcare needs
• Not measured not available to analysis

15
Known Sources of Bias

• Age
• Sexual orientation
• Education
• Substance use (over 3 mos)
• Can be quantified and used in analysis

16
Reduce bias

• Prognostic factors should not be related to
  treatment assignment
• Investigator should not assign treatment
• Assignment process should not be discoverable
• Equal allocation of subjects Achieving Balance

17
Systematic Errors Reduction

• Achieved by
• Randomization
• Use of control groups
• Blinding

18
Randomization

• Random allocation of different interventions
  (treatments or conditions) to subjects.
• Principal method for reducing treatment selection
  bias.
• Both known and unknown confounding factors are
  evenly distributed between treatment groups
• Guarantees that observed unobserved baseline
  difference is due to chance.
• Remaining difference can be attributable to
  treatment effect

19
Simple Randomization

• Make treatment assignment without regard to those
  already made. Does not take into account already
  made allocations.
• Not discoverable
• 2 treatments flip a coin for A or B
• Does not guarantee the same proportion of
  subjects in each treatment. Thus other methods
  e.g. Blocked randomization, Minimization etc

20
Blocked Randomization

• A block contains randomly permuted treatment
  assignment.
• A sequence of blocks forms the randomization
  schedule
• Block size must be a multiple of the of
  Treatments Eg 2 treatments (A, B), block size4
• 6 possible blocks (AABB) (ABAB) (ABBA) (BAAB)
  (BABA) (BBAA)

21
Stratified, blocked randomization

• Control imbalance (blocking)
• Produce same proportion of different types of
  subjects in all treatments (stratification
  blocking)
• Stratify on known/suspected prognostic factors.
  Thus can be used only when prognostic factors are
  known.

22
Comparison groups

• Comparisons between treatments are much more
  reproducible and are always preferable.
• Often one compares against a standard or
  traditional treatment that acts as baseline.

23
Control

• Placebo
• No treatment
• Historical (not ideal)
• Dose related
• Active

24
Placebo effect

• The positive effect of a treatment due to the
  patients state of mind relating to the treatment
  (e.g. belief in the treatments efficacy ) or
• The effect of observing a (usually) beneficial
  effect of treatment, even though the treatment
  has no actual therapeutic qualities, is known as
  a placebo effect.

25
Blinding

• Randomization alone does not guarantee that all
  bias is eliminated.
• There is potential bias that might exist if
  patients, the investigators responsible for
  treatment, or those responsible for evaluating
  patient outcome know which treatment the patient
  is receiving.
• Thus apply blinding wherever possible

26
Who is blinded

• Patients
• Persons responsible for interventions
• Persons responsible for outcome measurements or
  evaluation.
• Usually persons responsible for interventions and
  evaluations are same.
• Thus usually we have double blinded trials

27
Types of Blinding

• Open trial
• Single-blind trial
• Double-blind trial
• Triple-blind trial

28
Other Components of RCT

• Patients Subjects for study to represent the
  target population. So clear inclusion and
  exclusion criteria.
• Intervention Clearly defined.
• Outcome Clearly defined, preferably objective.

29
End points

• Precision in estimating treatment effects
• Endpoint ascertainment
• Objective endpoint preferable
• Usually the end points are
• Event Cure, Death, cancer recurrence
• Time to event

30
Ethical Issues

• Individual ethics (individual subject benefit or
  harm) vs Collective ethics (benefit or harm to
  mankind)
• We have to bear in mind patients subjected to
  trial and also future patients.

31
Ethics

• Depriving a control group from a treatment
• Should all patients be subjected to potentially
  beneficial treatments
• Should all patients be subjected to potentially
  harmful treatments?
• What if a potentially beneficial intervention was
  available only in short supply (limited resources

32
Size of Trials

• Single centre or Multicentre
• Calculate sample size to get best evidence that
  could influence policy makers to change the
  practice based on evidence.
• Estimating treatment difference that is
  clinically meaningful is important.

33
Monitoring bodies

• The committee or group that is responsible for
  the conduct of the trial.
• An independent ethical committee.
• A external monitoring body usually referred to as
  Data Monitoring Committee.
• Rationale for DMC Protocol compliance, Adverse
  effects, Data Processing and Interim analysis.

34
Reporting of Trials

• The two important components to critically
  appraise and RCT are
• 1. Consort statement
• 2. Intention to treat analysis

35
What is Number Needed to Treat?

• NNT or Number Needed to Treat- is a summary
  measure in clinical trials
• It indicates the number of people who should
  undergo the new treatment so that at least one is
  cured
• It is derived from the relationship NNT
  1/Risk difference between test and control groups

36
Interaction E.g. Trial Data
37
Interaction

• The observed risk ratios of 1.05 and 0.59 are
  suggestive of a difference in the impact of CABG
  (compared with angioplasty) among non-diabetics
  and among diabetics.
• Suppose this difference in the impact of CABG was
  true, i.e. that CABG really was beneficial in
  diabetics but not non-diabetics. Then we can then
  say there is an interaction between treatment and
  diabetic status.

38
Interaction

• However, these observed results alone are not
  enough to conclude that a true interaction
  exists. We need a more formal analysis and we do
  this using a significance test. First we need to
  state an appropriate null hypothesis. The next
  stage is to assess the probability that the
  results could have arisen by chance if the null
  hypothesis is true, i.e. we produce a p-value,
  and the smaller the p-value the more evidence
  against the null hypothesis. This is the
  interaction test.

39
Interaction

• Cautious Overinterpretation due to lack of power
  risk of false positive result.
• Different types of interaction can occur between
  treatment and some other patient characteristic.
• Qualitative interaction
• All or nothing
• Quantitative

40
Interaction Qualitative

• This is where the treatment effect is in the
  opposite direction according to the subgroup
  category. For example, in the gemfibrozil trial,
  the observed risk reduction for smokers was a 16
  increase in risk whereas for the non-smokers it
  was a 34 decrease. However, such an interaction
  is usually implausible, and a true interaction of
  this type is rare.

41
Interaction All or Nothing

• This is where the treatment is effective only in
  a particular subgroup of patient. For example, a
  treatment may be effective in men but not in
  women. We will see a possible example of this
  later in the section. This type of interaction is
  more plausible (e.g. treatment might work in
  higher risk patients) and would be important to
  know.

42
Interaction Quantitative

• This is where the treatment is effective in more
  than one category of patient but the magnitude of
  the benefit differs. This type of interaction is
  very plausible but less clinically important
  since there is benefit in all categories.

43
Interaction Key points

• Keep the emphasis on the overall result
• Pre-specification of subgroups
• Limit the number of subgroup variables
• View subgroup analyses as exploratory
• Lack of power of subgroup analyses
• Exaggeration of subgroup claims
• Other Issues Chance, disease severity,
  Complications, Referral bias, Other trial
  results, Treatment effects over time.

44
Confounding

• A Confounder refers to an extraneous variable
  that satisfies both of two conditions 1. it is a
  risk factor for the disease being studied, and
• 2. it is associated with the exposure being
  studied but is not a consequence of exposure.

45
Confounding

• For example, alcohol its relation to lung
  cancer.
• Adjusting for the effects of confounding factors
  is evidently important.
• Dealt with in the study design by matching or
  stratifying sampling of study subjects, or in the
  data analysis by stratified or multivariate
  analyses (4,10,12).

46
Alternative Designs

• Factorial Design Treatment A vs. the control of
  treatment A Treatment B vs. the control of
  treatment B.
• Non-inferiority or Equivalence trials
• Crossover trials
• Cluster randomization trials E.g. Comparing
  effects of health education in adolescents by
  selecting schools as clusters.