Navigating Go/No-Go Decisions in Early-Phase Clinical Trials: A Data-Driven Framework

Go/no-go decisions in early phase clinical trials work only when the decision rule is written down before unblinding, then tested against evidence that was collected to answer it. Treat a stop as a result, not a negotiation. Regulators are converging on the same discipline: the FDA published a request for information on AI-enabled optimization of early phase clinical trials on April 29, 2026, followed by a second notice on May 28, 2026 — two notices in eight weeks on the same pilot program.

Key takeaways

• A go/no-go call is a capital allocation decision, not a scientific verdict. Frame it as "what do we spend next, and on what evidence?"
• Pre-specify the decision rule — effect size, biomarker direction, safety margin, enrolment feasibility — before you see unblinded data.
• p-values alone are the weakest input at early phase, where samples are small and confidence intervals are wide.
• Biomarker and exposure-response data usually carry more decision weight than the primary endpoint at Phase 1/2a.
• Two of the vendor resources most often cited on this topic now return 404s. Build your evidence base from retrievable sources.

What Are Go/No-Go Decisions in Early-Phase Clinical Development?

Go/no-go decisions are staged investment checkpoints where a sponsor commits or withholds the next tranche of capital and effort — typically at the transitions from Phase 1 to Phase 2, from 2a to 2b, and from 2b to Phase 3. They differ from safety-driven holds because the trigger is a pre-agreed evidence threshold, not an adverse event.

The distinction matters operationally. A DSMB stop is reactive and usually binary. A go/no-go review is scheduled, has a named decision-maker, and produces one of four outputs: continue as planned, continue with a protocol amendment, pause for more data, or stop. If your governance cannot produce the fourth output, you do not have a decision framework — you have a launch sequence with extra meetings.

Why Are Early Go/No-Go Decisions More Critical Now?

Later-phase trials consume the large majority of a program's total development spend, so a wrong Phase 2 decision is expensive in a way a wrong Phase 1 decision is not. Capital conditions since 2022 have made that asymmetry harsher: fewer programs get a second chance to correct a bad assumption, and investors price in the option value of stopping early.

The regulatory environment is shifting too. The FDA's AI pilot program drew two Federal Register notices within eight weeks — the April 29, 2026 RFI and the May 28, 2026 follow-up. Whatever the pilot ultimately permits, the direction is clear: sponsors will be expected to justify early-phase design and stopping choices with documented reasoning, not intuition.

What Criteria Should Drive a Clinical Trial Go/No-Go Decision?

Clinical trial go/no-go criteria should cover four domains — biology, exposure, safety, and operability — and each domain needs a threshold written before unblinding. A program that clears three domains and fails one has not passed; it has a specific, addressable problem, which is a different conversation.

The operability row is the one teams skip. It is also the one that most often converts a scientific "go" into a financial "no" — a topic I covered in more depth in optimizing patient recruitment beyond traditional approaches.

Which Metrics Matter Beyond p-values?

At early phase, p-values are the least informative output you have, because small samples produce wide confidence intervals and a "significant" result can sit on a fragile effect size. The metrics that actually move decisions are effect size relative to your Phase 3 assumption, exposure-response shape, durability of response, and the safety margin at the proposed Phase 2 dose.

Add two more that rarely appear in statistical analysis plans: screen failure rate by site, and the competitive landscape at the time of the decision. A signal that looked differentiated at protocol design can be unremarkable by the readout if two competitors have reported since. Neither of those is a statistical question, and both change the answer.

How Do You Build a Quantitative Go/No-Go Framework?

A quantitative go/no-go framework is a written decision rule plus a governance process that cannot quietly amend it. Build it in this order, and finish before the first patient is dosed.

1. Name the decision. One sentence: "At end of Phase 2a, we will decide whether to fund a 300-patient Phase 2b in this indication."
2. State the alternatives. Continue, amend, pause, stop — with the budget attached to each.
3. Pre-specify thresholds for each domain in the table above, including the exact analysis population.
4. Define the minimum effect size that justifies the next phase, derived from the Phase 3 power calculation rather than from the observed data.
5. Assign a decision owner who is not the program lead. The program lead has an obvious conflict of interest.
6. Fix the review date and the data cutoff before unblinding.
7. Document the outcome and the reasoning, including for stops. This is the artifact regulators and investors will ask for.
8. Re-run the framework at each subsequent checkpoint rather than reusing the Phase 1 version.

Step 4 is where most frameworks break. If you set the threshold after seeing the data, you have not made a decision — you have rationalised one.

What Role Does Biomarker Data Play in Go/No-Go Decisions?

Biomarker data in go/no-go decisions serves two distinct purposes, and conflating them causes errors. A target-engagement biomarker tells you the drug reached the biology. A downstream or predictive biomarker tells you the biology responded. The first is a pharmacokinetic-dynamic question you can usually answer with confidence; the second is an efficacy signal with all the uncertainty that implies.

Treat a positive engagement marker as permission to continue dosing, not as evidence of efficacy. Treat a predictive biomarker as a hypothesis for the next trial until it has been prospectively tested.

How Does Risk Assessment Work in Early Drug Development?

Risk assessment in early drug development is about identifying which assumptions, if wrong, would invalidate the whole program — then buying information cheaply before the expensive trial. Rank risks by cost-of-being-wrong, not by likelihood, because the cheap-to-fix risks rarely sink a program.

The IQVIA white paper on optimizing early phase clinical trials for rapid evidence-based decision-making is a reasonable starting point on structuring that evidence flow; note that the Applied Clinical Trials webcast page on the same theme is registration-gated, so plan for that if you want the recording rather than the summary.

Proof of Concept Go/No-Go: What's Different?

A proof-of-concept go/no-go is the highest-stakes checkpoint in early development, because it is the last one where stopping is cheap. Everything after it is a commitment to a population, an endpoint, and a manufacturing scale. The PoC decision should therefore be judged against a higher bar than a Phase 1 continuation: not "is there a signal?" but "is this signal large and durable enough that a Phase 3 powered on it would plausibly succeed?"

That reframing changes what you measure. It pushes you toward effect size and durability over statistical significance, and toward a frank read of whether the enrolled population resembles the one you intend to treat.

How Do Investors Make Investment Decisions in Clinical Trials?

Investors make clinical trial investment decisions on the same four domains sponsors use, but weighted toward operability and competitive position, because those are the risks that survive a good scientific result. A differentiated mechanism with an unrecruitable population is not a differentiated asset.

For diligence, ask for the pre-specified decision rule and the date it was written. If the sponsor cannot produce a rule that predates the data, price the program as though the readout is unaudited. The same scepticism applies to tooling — I wrote about separating hype from high-impact AI in clinical trial feasibility for exactly this reason.

How Do You Accelerate Early Phase Trials Without Cutting Corners?

You accelerate early phase trials by shortening the time between data availability and decision, not by shrinking the data collection. Most of the calendar time in early phase is spent waiting for analysis, debating interpretation, and re-running feasibility — not enrolling.

Adaptive elements help, provided they are pre-specified and the regulatory path is agreed in advance; the trade-offs are set out in adaptive trial design: balancing flexibility and regulatory rigor. What does not help is adding arms to avoid making a decision. That is delay dressed as optionality.

Why Case Studies Are a Weak Substitute for Your Own Data

I have deliberately not included named case studies here, because a published account of someone else's go/no-go call is a narrative reconstructed after the fact, with the decision rule usually invisible. You cannot tell from the outside whether the sponsor followed a framework or got lucky, and survivorship bias means the failures are the ones nobody writes up.

What transfers between programs is the structure: pre-specified thresholds, a named decision owner, and a written record. Copy the structure, not the outcome.

One practical note on sourcing. Two vendor resources frequently cited on this topic — Incertive's explainer on go/no-go decisions and MMS Holdings' webinar on go/no-go decision-making for early-stage studies — currently return "page not found." Link rot is not a reason to skip the reading, but it is a reason to cite sources you can retrieve when an investor asks for your evidence base six months from now.

Leveraging Technology for Enhanced Decision Support

Decision-support tooling is worth paying for when it compresses the interval between "data available" and "decision made," and not otherwise. The honest test: does it produce a verdict with citations you can hand to an investor or a regulator, or does it produce a dashboard that still requires three weeks of interpretation?

Clinical Trial OS runs a six-pillar feasibility assessment and returns a go/no-go verdict in hours, with every number cited to its public source — which is the format a diligence conversation actually needs. If your current process already produces that artifact reliably, you do not need new software. If it produces a slide deck that everyone argues about, the problem is the framework, not the tooling.

FAQ

What are the key go/no-go criteria in early phase clinical trials?

Four domains: target engagement, efficacy signal size and durability, safety margin at the proposed dose, and enrolment operability. Each needs a numeric threshold written before unblinding, plus a named decision owner who is not the program lead.

How is a go/no-go decision different from a DSMB safety review?

A DSMB review is triggered by accumulating safety data and can halt a trial at any time. A go/no-go review is a scheduled capital decision at a predefined milestone, producing one of four outputs: continue, amend, pause, or stop.

Can you make a go/no-go decision on biomarker data alone?

No. A target-engagement biomarker supports dose selection and permission to continue dosing, but it is not evidence of efficacy. Predictive biomarkers are hypotheses for the next trial until prospectively validated in the intended population.

What is the most common mistake in early-phase go/no-go frameworks?

Setting the effect-size threshold after seeing the data. If the decision rule postdates the readout, you have rationalised a decision rather than made one, and investors and regulators will treat the result accordingly.

When should a biotech stop a program rather than amend the protocol?

When the failure is in the mechanism or the exposure-response relationship, not the design. Design problems — wrong dose, wrong population, insensitive endpoint — are amendable. A flat exposure-response curve at tolerated exposure usually is not.