In 1978 the National Commission for the Protection of Human Subjects of Biomedical and Behavioral Research issued the Belmont Report, which has since become a pivotal document in medical research ethics. This report pointed to three unifying ethical principles that must be followed in clinical research: respect for persons, beneficence, and justice, which are considered fundamental for ethical clinical research.
However, ensuring that study design aligns with these fundamental principles isn’t always easy. The good news? Adaptive design methodologies can help.
Respect That Goes Beyond Informed Consent
Trials evolve from initial uncertainty to obvious differences between treatment arms. In the case of progressive fatal diseases, patient risk from placebo or active comparator may not be minimal. In such cases, to maintain respect for persons, investigators may recognize a duty to share interim data with their patients in keeping with principles of informed consent and minimizing harm. The dilemma becomes whether to continue withholding interim analysis information to maintain the scientific integrity of the trial (also an ethical imperative) or to risk compromising trial integrity in favor of prioritizing study participant health and well-being.
Case Study: BIG 1-98
The Breast International Group (BIG) 1-98 Trial was a randomized, double-blind, phase 3 trial comparing letrozole and tamoxifen as adjuvant endocrine therapy in post-menopausal women with operable invasive, receptor-positive breast cancer. The primary endpoint of the trial was disease-free survival, specifically as a comparison between those who started with letrozole and those who started with tamoxifen.
The three randomized treatment groups in BIG 1-98 were:
- Monotherapy with either letrozole or tamoxifen for five years.
- Letrozole for two years followed by tamoxifen for three years.
- Tamoxifen for two years followed by letrozole for three years.
Planned interim analysis found that, after a median follow-up of 26 months, letrozole monotherapy was superior to tamoxifen therapy. Based on these results, patients in the tamoxifen monotherapy group who were disease free were allowed to selectively cross over to letrozole treatment. However, this adaptation was not pre-specified and, with 25.2 percent of patients in the tamoxifen monotherapy arm electing to begin treatment with letrozole, this development confounded the interpretation of long-term follow-up data.
Appropriately planned and pre-specified adaptive design could have mitigated this problem, both ensuring valid data analysis and satisfying regulatory requirements while maximizing patient survival and health.
Outcome Adaptive Randomization as a Tool of Beneficence
Outcome adaptive randomization (OAR) is a popular tool for ensuring patient well-being, especially in life-or-death situations. OAR is designed to minimize harm and maximize benefits, allowing investigators to:
- Change sample sizes based on blinded or unblinded treatment outcomes using pre-specified rules for adaptation.
- Refine doses being studied by dropping treatment arms deemed ineffective and/or overly toxic.
- Add patients to alternate treatment arms that show early signs of patient tolerance and treatment activity.
Case Study: ECMO
Following promising results of a Phase 1 trial of extracorporeal membrane oxygenation (ECMO) for newborns with respiratory failure, researchers designed a trial using a randomized “play-the-winner” methodology. In it, the first patient had an equal probability of being assigned to either ECMO or optimal conventional treatment, while the next patient who enrolled had a greater chance of being assigned to the more successful treatment. This method could be used because the outcome of each case would be known very soon after randomization.
Researchers anticipated that most ECMO patients would survive, while most control patients would die and were understandably reluctant to withhold potentially lifesaving treatment. However, the small size of the control group necessitated a second trial using block randomization in its first phase. There, adaptive design halted randomization after four deaths, minimizing the number of newborns assigned to less effective conventional treatments.
Too Good to Be True?
While a powerful tool, OAR is not without drawbacks:
- More patients are indeed assigned to the presumptive superior therapy, but some continue to be randomized into inferior treatment arm.
- The additional explanations required for OAR may make truly informed consent more difficult to obtain, as many patients already have issues understanding even standard randomization.
- Possible enrollment bias if patients who enroll later are qualitatively different from early enrollees, which can be difficult to anticipate.
Additionally, some researchers argue that OAR actually increases the number of unfavorable outcomes for patients overall, requires larger trials, and can delay the introduction of effective treatments into routine clinical care. As such, OAR is most useful in rare disease settings where a majority of patients with a given condition take part in clinical trials.
Two Sides of Justice in Adaptive Design Trials
At the surface, randomization in clinical trials may appear to be unethical, as patients are given unequal treatment, with preliminary safety and efficacy data from open-label trials and anecdotal reports indicating the potential superiority of the investigational product. However, it’s important to remember that even Phase 3 trials fail.
Adaptive design methodologies offer the opportunity to expose fewer patients to ineffective and/or toxic treatments, ensuring a more fair distribution of superior treatments. However, adaptive design can also lead to operational bias: Interim analyses and trial adaptations may influence an investigator’s sense of justice in the middle of a trial, creating a shift in enrollment bias as the study continues. Adaptive allocation based on response means that it is difficult (if not impossible) to fully blind investigators to treatment arms.