Would You Randomize Before Consent?

Lessons from EDI and Zelen’s Design Dilemma

Imagine reviewing a clinical trial protocol that randomizes patients before asking for consent. Patients assigned to the control arm will receive nothing beyond standard care, so they will never be contacted. Only those randomized to the intervention will be informed that a trial exists.

Your first instinct is probably to reject it. It feels like a trick—an end run around informed consent dressed up as efficiency. Even if the intervention is minimal risk, the idea of enrolling people in research without their knowledge triggers every ethical reflex we’ve been trained to trust.

That instinct, however, is exactly what Marvin Zelen challenged in 1979. His proposal—later known as randomized consent, or Zelen’s design—argued that in some settings, asking for consent before randomization introduces more distortion than protection. If only the intervention group ever has to decide, the resulting sample may be closer to real-world care than anything produced by a conventional randomized trial.

In a traditional randomized controlled trial, the sequence is familiar:

1. Identify an eligible patient
2. Obtain informed consent
3. Randomize to treatment or control

Zelen inverted this order:

1. Randomize eligible patients first
2. Seek consent only from those assigned to the intervention
3. Leave control patients in standard care with no trial‑specific consent

The appeal is clear. The design reduces selection bias, preserves real‑world representativeness, and avoids the demoralizing dropout that can follow assignment to a “do nothing” control arm.

But critics raised an obvious concern: how can patients be included in research—even minimal‑risk research—without their knowledge? At what point does methodological cleanliness become ethical opacity?

The first time I encountered a post‑randomization consent design in a protocol, my instinct was to reject it outright. It felt like a procedural trick, a way of enrolling people in research without fully owning that choice. It took time—and exposure to settings where scale, passivity, and representativeness actually matter—for me to understand the counter‑argument: that in some cases, asking for consent first does not protect autonomy so much as quietly reshape who ends up represented at all.

A version of Zelen’s design was later used in the Early Detection Initiative (EDI) trial for pancreatic cancer, a large study evaluating whether earlier detection could be achieved among people with newly diagnosed diabetes.

The trial tested a two-step intervention: a risk stratification algorithm (the ENDPAC model), followed by CT imaging for those who crossed a predefined risk threshold. To preserve scale and avoid contaminating usual care, EDI randomized eligible patients up front and sought consent only from those assigned to imaging. Control-arm patients were never contacted and continued routine care, with outcomes captured passively through electronic medical records.

On paper, the design was well matched to the problem. Imaging was not part of standard care, the control arm reflected real-world practice, and the study required large numbers to observe a rare outcome. In practice, the design did not survive.

After roughly two years, the numbers told an uncomfortable story. Only about 20% of intervention‑assigned participants who were approached actually consented to imaging.

That figure was not merely disappointing; it was decisive. At that level of uptake, the randomized comparison no longer reflected the population it set out to study. The gains Zelen’s design promised—reduced bias, preserved scale, cleaner inference—collapsed along with participation. What remained was the administrative and ethical complexity, without the statistical payoff.

The trial was restructured.

EDI transitioned into a prospective observational study:

• All eligible patients are now enrolled via electronic records under a waiver of consent for passive follow‑up.
• Imaging is optional and offered only at select sites.
• High‑risk participants are approached based on operational feasibility rather than randomization.

What began as a randomized experiment became a pragmatic compromise.

Zelen’s design has always lived on a knife edge.

The design promised reduced selection bias, closer alignment with real-world care, and fewer control-arm dropouts. It also invited concerns about autonomy, transparency, and resistance from institutional review boards.

EDI attempted to mitigate these concerns:

• Ethicists were involved during protocol development.
• Only a limited, minimal dataset was collected for passive follow‑up to reduce re‑identification risk.
• Sites relied on existing opt‑out mechanisms embedded in clinical care.

Even so, the design proved too brittle to sustain.

Consider a low‑stakes public health intervention—a postcard reminder, a default referral, a nudge embedded in clinical workflow. Would randomizing before consent feel acceptable?

Now raise the stakes. A blood draw. A CT scan. A specialist referral triggered by an algorithm. Does a Zelen‑style design help reach people who would otherwise disengage—or does it undermine the trust that research depends on?

EDI forces this question into the open. Its original design was efficient and arguably more representative of real behavior—but only if enough people agreed to proceed once invited. When fewer than one in five said yes, the logic unraveled.

Today, trialists must weigh not only statistical rigor and feasibility, but also participant trust, perceived intrusion, and institutional culture. Zelen’s design still has a place—particularly in pragmatic and public health settings—but it demands far more care than its elegance suggests.

The best trial designs rarely fit neatly into ethical or operational boxes.

EDI’s original approach was courageous, efficient, and scientifically defensible. It also revealed the limits of what patients, IRBs, and sponsors are prepared to accept.

Designing trials today means navigating power calculations and human psychology; endpoints and trust.

Would you randomize before consent?

Sometimes, asking the question is as important as the answer.

📬 Want more design dilemmas like this? Subscribe to the newsletter or check out the Evidence in the Wild archive.