ncAdios

On first-in-human oncology

What a First-in-Human Trial Is Actually For

Published 2026-09-01 · Regulatory references are FDA guidance documents, identified below as final or draft. Guidance is non-binding and does not establish legally enforceable responsibilities. Not legal or regulatory advice.

Every first-in-human oncology trial produces data. Doses, exposures, adverse events, the first hints of activity. By that standard almost no FIH trial fails — the numbers almost always arrive.

Every patient who enrolls in a first-in-human oncology trial makes a wager — not just on this molecule, but on the intentions of the people who designed it. They are trusting that someone asked: what does this experiment need to resolve, and is this the right way to resolve it?

That question is also, precisely, the developer's question and the investor's question. An FIH trial is worth exactly the decisions it lets those three people resolve: the patient who said yes to the trial, the person spending the next development dollar, and the person spending the next investment dollar. When all three reach the same answer from the same data, the program is honest. When any of them would reach a different answer, the divergence is the finding — it usually means the program's story and its evidence have quietly parted company, and the trial has told you so before anyone at the table admits it.

The reason the developer's question and the investor's question converge is that both are downstream of a prior question — the one the patient answered when they said yes.

That is why the useful question is not what an FIH trial produced but what it resolves. A program can complete escalation, declare a recommended Phase 2 dose, and present a tidy waterfall plot — and still be unable to answer the question its next committed dollar depends on: is the next planned experiment the most capital-efficient way to resolve the uncertainty that most drives this asset's value? That question reads naturally from every chair. That it does is not a coincidence — it is the test of whether an FIH program was designed to learn or merely to advance.

The algorithm is not the program

Discussion of FIH design tends to collapse into a debate about escalation methods — 3+3 versus model-based, the merits of one design against another. The debate is real but subordinate. An escalation algorithm decides how to step between doses. It does not decide the questions that determine what the trial can learn:

  • Who should be treated, and does that population let a signal be interpreted?
  • What exposure range is biologically relevant, and how fast can clearly non-informative exposure be traversed?
  • Which evidence — toxicity, PK, pharmacodynamics, activity — should influence dose assignment, and is the trial structured to collect it when it matters?
  • Which doses must survive escalation, and will they need to be compared?
  • What uncertainty must be resolved before the next development or investment commitment — and does the design actually resolve it?

A program can hold a sophisticated answer to the algorithm question and no answer to these. The reverse is rarely true: teams that have thought through the decision sequence tend to choose sensible algorithms, because the algorithm is then serving something.

The practical test: before you lock an escalation design, write the one sentence describing what each cohort is meant to resolve. If a cohort has no sentence, you are asking patients to enroll in an experiment that has no answer it is trying to reach — and that is the cohort to redesign first.

The regulator already moved

For decades, oncology could treat dose-finding as a formality on the way to the maximum tolerated dose. That era is closed. The consequence of that shift is a liability most FIH assets carry and few price: the installed base of oncology programs was designed before the standard now used to judge them. Project Optimus was not a capital-efficiency initiative. It was the agency's response to watching patients receive doses they would not have received if sponsors had asked the right question first. FDA's dose-optimization guidance — final since August 2024 — asks sponsors to justify dosage with PK, pharmacodynamics, safety, tolerability, activity, and dose- and exposure-response, and identifies randomized comparison of more than one dosage as a recommended approach. The expectation here is expressed as guidance, not regulation; but programs are now read against it regardless, and the ones designed on the older maximum-tolerated-dose assumption meet it late — after a single dose has been carried into commitments that assumed it. The expensive part is not the expectation; it is retrofitting it.

That guidance governs the development program, not the first dose. For the starting dose itself, FDA's June 2026 draft guidance on QSP-based selection of the Minimum Anticipated Biological Effect Level formalizes model-based justification of the first human dose — the first FDA document to do so directly. It is draft, non-binding, and comment-closed as of July 2026; but it signals that the agency now expects a mechanism-grounded account of where escalation begins, not only where it lands.

Where a program enters the clinic is now part of the same calculus. As Alex Harding has argued (an experienced operator's argument, not a regulatory instrument), the US IND pathway — and the pre-IND feedback channel it opens — is an advantage with no direct equivalent in the leaner Australian CTN route, where trial review sits with an ethics committee rather than the regulator and no formal pre-submission feedback mechanism exists. Other jurisdictions do offer scientific advice; what the US couples uniquely tightly is that advice to the gate the trial actually passes through. The choice between pathways moves first dose by months and reshapes how much data a program must generate before it starts. A global program that has not chosen its development geography deliberately has usually chosen by default. Anyone evaluating an FIH-stage asset today — as its developer or its potential owner — is therefore reading it against a standard, and a set of pathway choices, that many running programs never consciously made.

Two ways to misread the same trial

Early oncology data invites two opposite mistakes, and both are made at the same tables — board approvals, term sheets, partnering decisions.

The first is reading too much. A response rate at an uncontrolled dose, in a small selected population, becomes the thesis for the next cohort and the next financing — and the exposures, denominators, and dose rationale underneath cannot carry the weight placed on them. Nothing about this looks like failure while it is happening; that is what makes it structurally easy to miss.

The second is reading too little. A program with a coherent biological story and an interpretable early signal is discounted — deprioritized internally, or marked down in diligence — because its data package is disorganized, the evidence real but never assembled into a form a committee can trust. Assets are mispriced in both directions by the same underlying failure: nobody separated what the data show from what the documents assert.

That failure is concrete, and so is the discipline that prevents it. "The drug is active at 200 mg" is three claims wearing one coat: a response was observed (fact), at a dose assumed to equal the exposure that mattered (inference), in patients who may not represent the development population (uncertainty). A data room that lets those three travel together is how a program gets mispriced — in either direction. The discipline is to state what is known, what is inferred, what is unknown, and what could kill the asset — and to keep those four categories from bleeding into each other. Most data rooms do not enforce that separation. Most committee debates dissolve without it.

The same read, from either chair

Two of those three chairs commit money. A developer deciding the next cohort and an investor deciding the next tranche need the same thing: an honest account of what the program has actually learned, at what confidence, and whether the next committed step — a cohort, a manufacturing lot, a go/no-go to expansion, a tranche — is aimed at the uncertainty that matters most. Both are answerable to the third chair's prior question, which is why their agreement is a test rather than a coincidence. When they reach different answers from the same data, that disagreement is the finding, not a nuisance to be reconciled away.

This is judgment work, not computation. But it is judgment that improves with structure: explicit gates a program must pass for the rest of the analysis to matter, and a stated hierarchy of which uncertainties drive value. Structure does not replace the thirty years of decisions behind a senior read; it makes that read auditable — which is what a board, a committee, or a skeptical co-investor is entitled to.

And the read that matters is not "is this good science." It is "will the people who decide this program's fate — the review division and the next investor — accept what the data actually support." That is a clinical judgment and a regulatory judgment at once, and they are rarely the same person's.

The mechanics above are the floor. The judgment — what this program has learned, and whether its next dollar is aimed correctly — is the substance of the FIH Diligence Read: ten business days, fixed scope, a gate-by-gate account of what the data support, and a board-ready synthesis that keeps fact, inference, and uncertainty separate. Where the evidence does not support a recommendation, the Read says so — refusal is part of the output. If the decision in front of you is an FDA interaction rather than a capital one, that is the Calibration Sprint's terrain — same discipline, different door.

Book a 20-minute fit call See the FIH Diligence Read

Sources

Jesús Gómez-Navarro, M.D., is a medical oncologist and drug development executive, and founder of OncAdios LLC. He advises AI-oncology and biotech companies as a fractional CMO, board director, or co-founder.

← Back to Writing  ·  What a pre-IND meeting actually decides →  ·  The FIH Diligence Read →