Would You Randomize Before Consent? Lessons from EDI and Zelen's Design Dilemma

What happens when the ideal design for scientific clarity clashes with our instincts for patient autonomy? The Early Detection Initiative (EDI) for pancreatic cancer started with a bold design—then changed course. Here’s why it matters.

Act I: The Trial That Got Ethics Committees Buzzing

Imagine you’re reviewing a clinical trial proposal. It wants to randomize patients before getting their consent. The logic? Patients in the control group won’t receive any experimental intervention—so why burden them with paperwork and stress?

That was the core idea behind Marvin Zelen's 1979 proposal for what became known as "randomized consent" or the Zelen design.

Fast forward to the EDI trial for pancreatic cancer—a real-world example of this design in action, and a case study in how bold ideas bump into practical and ethical resistance.

Act II: What Is Zelen’s Design, and Why Is It Controversial?

In a traditional RCT, you:

1. Approach the patient
2. Get informed consent
3. Randomize them to a treatment arm

Zelen flipped that sequence:

1. Randomize eligible patients first
2. Only ask for consent from those assigned to the intervention arm
3. Leave control patients under standard care with no trial-specific consent

The logic: minimize selection bias, preserve real-world representativeness, and reduce dropouts from disappointed control arm assignees.

Critics, however, cried foul. How could you ethically include patients in research without their knowledge, even if it was minimal-risk?

Act III: Enter the Early Detection Initiative (EDI)

The EDI trial (NCT04662879) aimed to determine whether a dual intervention—a risk stratification algorithm (ENDPAC model) followed by CT imaging for high-risk patients—could detect pancreatic cancer earlier than usual care among people with newly diagnosed diabetes.

Its original design? Post-randomization consent a.k.a. Zelen design:

• All eligible patients were first randomized.
• Only those in the intervention arm with a high ENDPAC score were approached for imaging and consent.
• The control arm (no imaging) did not receive any trial-directed care or contact; participants continued with their usual clinical care and were followed passively via their electronic medical records (EMR).

This was a clever fit: the imaging was non-standard, the control group followed existing care, and the study needed scale.

But the design didn’t hold.

Act IV: When Good Designs Meet Real-World Constraints

After two years, EDI faced a tough reality: only ~20% of intervention-assigned participants who were approached for imaging consented.

That wasn’t enough to meet the study’s primary endpoint under the original RCT framework. Consent fatigue, logistical delays, and discomfort with the design chipped away at feasibility.

The trial was revised:

• It became a prospective observational study with optional imaging at certain sites.
• All patients are now enrolled via EMR under a waiver of consent for passive follow-up.
• Only high-risk participants (ENDPAC > 0) at designated imaging sites—chosen based on operational feasibility, not randomization—are approached for consent to imaging.

Zelen's bold design gave way to a pragmatic compromise.

Act V: The Ethical Debate, Revisited

Zelen's design walks a razor-thin line:

• Pros: Reduces selection bias, mimics real-world care, improves power by avoiding dropout
• Cons: Undermines patient autonomy, raises concerns about transparency, complicates IRB review

In the EDI trial, this tension was addressed head-on:

• Ethicists were consulted in protocol development.
• Only a limited dataset was collected, primarily minimal demographic and clinical variables, to reduce re-identification risk. This supported the waiver of informed consent for passive EMR follow-up.
• Sites used existing opt-out policies to honor patient preferences.

But in the end, the design was still too unwieldy to maintain.

Act VI: Would You Use It?

Let’s say you're running a public health trial where the intervention is a postcard reminder, and the control is... nothing. Would you randomize before consent?

What if the intervention is more involved—a blood draw, a CT scan, or a referral for specialist follow-up? Would a Zelen-style design help you reach the people who wouldn’t otherwise engage, or would it erode the transparency expected of research?

EDI pushes us to grapple with these questions. The original design was efficient and arguably more representative of real-world behavior, but only if enough people agreed to proceed once invited. When fewer than one in five said yes, the statistical advantages unraveled.

Today, researchers must weigh not only scientific rigor and feasibility, but also participant trust, perceived intrusion, and IRB culture. The Zelen model still has power, especially in pragmatic or public health settings, but it demands a deeper reflection on what kind of consent truly protects participants, and what kind may actually exclude those most at risk.

Zelen’s design is not dead. But it’s no longer simple.

Final Act: Designing in the Gray

The best trial designs don’t always fit neat ethical or operational boxes. The original EDI design was courageous, efficient, and scientifically sound. But it also revealed the limits of what IRBs, patients, and sponsors are ready to accept.

Designing studies today means navigating not just statistical power and endpoints but also trust, communication, and feasibility.

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.