Stefano Garzon

Research · Oct 23, 2026

Ethics in clinical trials: equipoise, consent, conflict of interest, and publication bias

Concept

A clinical trial has obligations to participants and to science. Friedman adopts seven ethical requirements: social or scientific value, scientific validity, fair participant selection, favorable benefit-risk balance, independent review, informed consent, and respect for enrolled participants.

These requirements have methodological consequences. A trial too small to answer its question exposes participants to risk without producing knowledge. An unpublished trial has the same effect.

Equipoise

Randomization is acceptable when there is real uncertainty about which treatment is better. Requiring each investigator to have no preference at all makes most trials impossible. Freedman proposed clinical equipoise: the requirement is met when there is genuine uncertainty in the expert medical community about the preferred treatment, not necessarily in the individual investigator.

Equipoise can change during a trial. In ISIS-2, with 17,187 patients with suspected infarction, streptokinase reduced 5-week vascular mortality from 12.0% to 9.2%. During recruitment the monitoring committee told investigators there was proof of benefit for patients treated within 4 hours of pain onset. Randomization continued for physicians who remained uncertain. Friedman notes that many would argue enrollment in that subgroup should have stopped.

By the same logic, a trial designed to prove harm is unacceptable. After two CAST arms were stopped, the investigators rejected a proposal to test quinidine in order to show harm.

Placebo and choice of control

If a life-saving treatment exists, comparing a new drug with placebo in its place is unethical. Placebo is acceptable when added to standard care, when existing therapy has no proven benefit or is poorly tolerated, or when withholding treatment causes only mild, treatable discomfort.

Placebo also has informational value. Open trials of renal denervation showed systolic reductions of 22 to 32 mmHg. In SYMPLICITY HTN-3, with a sham procedure and 535 patients, the reduction was 14.1 mmHg with denervation and 11.7 mmHg with sham, a difference of 2.4 mmHg.

Consent is a process, not a signature on a form. Basic elements include a statement that the study is research, its aims, procedures, foreseeable risks and benefits, alternatives, confidentiality, and the right to withdraw without penalty.

The most cited failure is therapeutic misconception: the participant perceives research as treatment. It occurs in serious diseases with no alternative and when the treating physician is also the investigator.

SUPPORT randomized 1,316 extremely preterm infants to saturation targets of 85–89% or 91–95%. The lower target reduced retinopathy and increased mortality before discharge. The US regulator concluded that the consent form should have mentioned a possible increase in mortality.

Some settings allow no prior consent, such as cardiac arrest, under specific rules: community consultation, a safety committee, and deferred consent.

Conflict of interest

Financial and intellectual conflicts can bias design, conduct, analysis, and publication. Friedman considers disclosure and ethics committee oversight sufficient for early phase studies but not for late phase trials. In these, analysis should be done or validated by a group independent of the sponsor.

Heres and colleagues examined industry-sponsored head-to-head trials of second-generation antipsychotics. The result favored the sponsor’s drug in about 90% of studies, including when the same drugs were compared by different sponsors.

Publication bias

Turner and colleagues compared 74 FDA-registered antidepressant trials with the literature. Of 38 positive trials according to the FDA, 37 were published. Of 36 negative or questionable trials, 22 were not published and 11 were published in a way that conveyed a positive result. Prospective registration in public databases makes unpublished trials traceable.

Common errors when reading a paper

What to check in a trial

  1. Whether the community was uncertain about the compared treatment when the trial started.
  2. The chosen control and whether it deprives participants of effective treatment.
  3. Registration number and agreement between registered and published outcomes.
  4. Funding source, sponsor role, and who performed the analysis.
  5. Whether the analysis was validated by an independent group.
  6. How consent was obtained in vulnerable or emergency populations.

Related posts: interim analyses, meta-analysis.

References