4  Clinical trials for cancer treatments

4.1 Questions to answer in designing a clinical trial (Article)

D.1.1 What is the current standard of care?

D.1.2 What care outcomes do we want to improve, and how much improvement would count as a real advance? (Here risk-benefit comparison becomes important)

D.1.3 What is the proposed alternative?

D.1.4 How do we determine a safe “dose” of the proposed alternative? (In the case of surgery this question may not make sense)

D.1.5 How do we verify that the proposed alternative has any effect at all?

D.1.6 How do we measure the overall benefit conferred by using the alternative method?

4.2 The typical strategy: Three phases

  • Phase I: enroll healthy volunteers to do dose finding, assessing toxicity and side effects
  • Phase II: enroll patients with the disease and administer the maximum tolerated dose to establish that the new treatment has some beneficial effect
  • Phase III: enroll patients with the disease, but set up a comparison of the new treatment versus the standard of care
    • a design plan is produced detailing how many patients are to be enrolled, how the treatments will be administered, and how the data will be analyzed
    • patients are randomly assigned either the new treatment, or the standard of care
    • a Data Safety Monitoring Board will review study data periodically
      • studies may end prematurely if great benefit is seen early on, or if it appears no benefit could be documented by continuing the trial

4.3 Experimentation with human subjects is carefully regulated

The concept of informed consent is central in the implementation of clinical trials.

The Belmont report defined principles and guidelines for all human subjects research.

4.4 Equipoise: when medical knowledge can only advance by experimenting

4.5 Equipoise: individual and clinical

Distinguished in AMA ethics, 2015

What is the state of medical knowledge? Why is this trial asking an important question? How does the likelihood of benefiting in study A compare with that of benefiting in study B or with the standard of care? – attributed to B. Freedman

  • Belief in the new treatment

  • Belief in the patient’s suitability for the new treatment

  • Interpretation of new evidence

  • Risk that new evidence will “regress”

  • Alternative: fully informed patient consent

4.6 Blinding and randomization to reduce bias

The concept of the double-blind clinical experiment is well-established

  • The investigator does not know what treatment has been given to the patient

  • The patient does not know what drug they have received

This is accomplished by providing treatments that are identical in appearance, taste, etc., but differ in their chemical composition and hypothesized effects. An independent party has knowledge of the treatment that has been administered, and can decode the treatment assignments as needed.

The role of blinding in experimentation is reviewed in a Nature commentary, and the Wikipedia article has many useful references.

4.6.1 Randomization to reduce bias

Once blinding has been accomplished it may seem unnecessary to take further steps to eliminate bias in the evaluation of an experimental comparison.

Blinding helps to eliminate specific forms of bias rooted in (possibly un- or sub-conscious) investigator or participant actions that may affect experimental outcomes or interpretation.

Other forms of bias can be mitigated using random allocation of treatments to participants. As an illustration suppose an investigator decides to use “alternating assignments”, and typically sees two eligible patients a day. We could then have

                Day 1    Day 2   Day 3 ....
Treatment      A    B    A   B   A   B
Time of day    AM   PM   AM  PM  AM  PM

A problem would arise if the characteristics of patients coming in the morning were different from those coming in the afternoon. For example, the patients arriving in the AM slot could tend to be older. The comparison between treatments A and B would then be confounded with a difference in average ages of patients in the experiment.

If, instead of alternation, the treatment assignment were randomized, the patients receiving treatment A would have characteristics that are on average indistinguishable from those receiving treatment B.

Details of the approach to randomization will have an impact on its effectivness. Challenges can arise in multicenter studies with many centers. The following display plots the excess numbers of patients assigned to one arm in a trial, against the total number enrolled.

Excursions 1

There are long stretches of time during which more patients in the study are receiving one of the treatments than the other.

The use of a different approach to randomization, in which small blocks of allocations are randomized independently, can reduce the size and duration of periods of imbalance over time.

Excursions 2

Randomization is a well-accepted tool in experimental design, but must be conducted with care. The wikipedia article has some interesting references on assessment of the overall effect of use of randomization in health research.

Back to top