What FDA requires to let a first-in-human oncology trial proceed, and what puts the IND on hold

A first-in-human (FIH) oncology trial sits at the point of maximum uncertainty: no human has taken the drug, the patients are often out of standard options, and the sponsor is asking FDA to accept a package built entirely on animal and in vitro data. The Agency's review of the initial IND turns on four linked questions that a regulatory team should be able to answer before submission: is the nonclinical package adequate, is the starting dose defensible, is the escalation and dose-selection plan sound, and is there enough information for FDA to conclude the risk to subjects is reasonable. Failure on any one of them is what converts a 30-day review into a clinical hold.

The nonclinical package: ICH S9 sets a deliberately lighter bar for advanced cancer

For pharmaceuticals intended for patients with advanced cancer, FDA applies ICH S9 rather than the general M3(R2) nonclinical paradigm, and S9 is explicit that some studies routinely expected for other indications are not needed to start dosing.

• General toxicology is the core requirement. For small molecules, repeat-dose studies are generally conducted in both a rodent and a non-rodent species, with the toxicology dosing schedule approximating the intended clinical schedule; for biopharmaceuticals, species selection follows ICH S6 ⁶⁶. A single rodent species can suffice for a genotoxic drug directed at rapidly dividing cells where the rodent is a relevant species ⁶⁶. Safety pharmacology and toxicokinetic endpoints can be folded into the general toxicology studies rather than run as stand-alone studies ⁶⁷.
• Several studies are explicitly not essential to begin FIH dosing in advanced cancer. Embryo-fetal developmental toxicity studies are not considered essential to support trials in this population (they should be available by the marketing application), fertility and early embryonic development studies are not warranted, and full recovery/reversibility and formal NOAEL/NOEL determinations are not treated as prerequisites to starting ⁶⁶⁷⁰.

The practical message for a reviewer: the FIH oncology package is scoped to what is needed to pick a safe starting dose and characterize the near-term toxicologic risk, not to a complete developmental toxicology dataset. What must still be airtight is the general toxicology that anchors the dose.

Justifying the starting dose

FDA's starting-dose expectations run on two parallel tracks depending on what drives the drug's risk.

The NOAEL-to-HED-to-safety-factor pathway

The default quantitative method comes from FDA's guidance on estimating the maximum safe starting dose. The Agency recommends identifying the no observed adverse effect level (NOAEL) in the most appropriate animal species, converting that NOAEL to a human equivalent dose (HED) using body-surface-area (BSA) correction factors because BSA scaling is conservative and increases clinical-trial safety, and then dividing the HED by a safety factor (generally at least 10) to obtain the maximum recommended starting dose (MRSD) ^{106105107111119143}. The pharmacologically active dose (PAD), the lowest animal dose producing the intended pharmacologic effect, is to be weighed alongside the full toxicology and PK dataset, and a safety factor remains warranted even when PK/PD modeling is informative ¹¹⁹¹⁰⁷. The guidance provides the standard BSA conversion factors (for example, divide a mouse mg/kg dose by 12.3, rat by 6.2, dog by 1.8, monkey by 3.1 to obtain the HED) ¹³⁷.

The oncology-specific derivation for cytotoxic agents

For anticancer pharmaceuticals, the starting dose is more commonly anchored to toxicity rather than to a no-effect level. The historical and still-referenced approach is to take a fraction of the dose that produces serious toxicity in the test species, one-tenth of the severely toxic dose to 10% of rodents (STD10) and one-sixth of the highest non-severely toxic dose (HNSTD) in non-rodents, convert each to a human equivalent dose by BSA scaling, and carry forward the more conservative (lower) value ^7137105. The scientific basis FDA cites is that rodent lethality endpoints and non-rodent MTDs have historically correlated with the human MTD when expressed on a mg/m² basis ¹⁰⁵.

When pharmacology, not toxicology, sets the dose (MABEL)

For biologics and other agents where the pharmacology itself is the principal driver of risk, a NOAEL-derived dose can be dangerously high. FDA's draft guidance recommends a minimum anticipated biological effect level (MABEL) approach, supported by quantitative systems pharmacology modeling of the biological interaction, for FIH dose selection in these products, including biological products licensed under section 351(a) of the PHS Act ¹⁰. For antibody-drug conjugates specifically, FDA emphasizes that dosing be driven by integrated PK/PD, safety, biomarker, and receptor-occupancy data because small increases in exposure can produce significant adverse reactions ¹¹. In oncology, sponsors frequently present both a toxicology-based and a pharmacology-based estimate and justify the choice.

Dose-escalation and dose-optimization design

Escalation schemes and stopping rules

FDA recognizes the traditional 3+3 design, in which three patients are enrolled per dose level, escalation proceeds when no dose-limiting toxicities (DLTs) occur, a level is expanded to six on a single DLT, and the MTD is considered exceeded when more than one of six patients experiences a DLT, a scheme aimed at avoiding a phase 2 dose with treatment-limiting toxicity in more than roughly 17% of subjects ¹². The Agency also notes that 3+3 is not always the best choice and that alternatives such as accelerated titration or continual reassessment may be appropriate ¹². Regardless of scheme, the protocol must clearly define the DLTs, off-treatment criteria, and study stopping rules that protect subjects, and, for model-based designs, the acceptable parameters for the dosing endpoint, supported by data ¹².

Project Optimus: moving past the MTD

FDA's oncology dose-optimization guidance (the core of Project Optimus) is now central to how the Agency reviews FIH-through-early-development dosing. The guidance states that many modern oncology drugs have dose-response relationships in which doses below the MTD deliver similar activity with less toxicity, so dose-finding should not simply seek the MTD ⁹². FDA recommends selecting the dosage using the totality of nonclinical and clinical data across a range of doses, PK, PD, safety, tolerability, and activity, plus dose- and exposure-response relationships, and designing trials that adequately evaluate more than one dosage rather than deferring optimization until after approval ⁹³⁹⁵. Waiting until post-approval, FDA warns, risks exposing many patients to a poorly tolerated or non-optimal dose ⁹³.

Expansion cohorts and safety oversight

Multiple expansion cohorts appended to a single FIH protocol receive specific scrutiny. FDA recommends that such a protocol combine an initial dose-escalation phase with three or more cohorts that each carry cohort-specific objectives, that cross-cohort comparisons not be planned absent a prespecified randomization and analysis plan, and that new safety information be disseminated to investigators, IRBs, and FDA no later than 15 calendar days after the sponsor determines it should be reported ³⁵. Cohort sizing is expected to be justified: safety-confirmation cohorts at the presumptive recommended phase 2 dose are small (for example, roughly 6-12 subjects for AML safety confirmation, about 20 for AML efficacy estimation, and generally no more than 20 for MDS), and activity-estimating cohorts should carry a statistical analysis plan with a maximum-sample-size justification and stopping rules for lack of activity (generally up to about 40 subjects for solid tumors under a Simon two-stage design) ^13141617. Where the therapeutic index is narrow and DLTs may be severe, expansion should be delayed until enough data exist to define the RP2D, or the confirmatory trial should build in a very early interim safety analysis ¹³¹⁴¹⁷.

FDA also expects structural safety oversight for these designs: a systematic safety-reporting plan with more-frequent-than-annual cumulative summaries, real-time serious-adverse-event review, prespecified toxicity stopping rules with the exact bounds, statistical method, and clinical assumptions stated, and an independent data monitoring committee or comparable entity able to recommend eligibility, dose, or schedule changes ¹⁶²⁴²⁵. FDA suggests setting stopping bounds to give a high probability (greater than 60-70%) of pausing enrollment at the lowest unacceptable toxicity rate while keeping the probability of a pause below 30% when toxicity is acceptable ¹⁶²⁴.

What triggers a clinical hold on a first-in-human oncology IND

A clinical hold is FDA's statutory tool to stop or prevent a trial, and the grounds for a Phase 1 IND are set out in 21 CFR 312.42(b)(1). The guidance identifies four bases that recur in FIH oncology holds:

• Subjects are or would be exposed to an unreasonable and significant risk of illness or injury ^99184187.
• The clinical investigator is not qualified by scientific training and experience to conduct the trial ¹⁸⁷.
• The investigator's brochure is misleading, erroneous, or incomplete ¹⁸⁷.
• The IND contains insufficient information for FDA to assess the risks to subjects ^187101104.

The last ground is the one most FIH oncology holds actually cite, because a Phase 1 package can be internally consistent yet still leave FDA unable to conclude the risk is reasonable. In practice, the "insufficient information" and "unreasonable risk" grounds are populated by deficiencies in three sections:

Pharmacology/toxicology deficiencies

The nonclinical package must provide the data FDA needs to conclude it is reasonably safe to proceed. If the absence of a study leaves insufficient information to assess the risks to human subjects, that study is needed, and single- and repeat-dose general toxicology studies are what support both the dose and duration of the trial and the initial safe starting dose ¹. Deficiencies that can support a hold therefore include inadequate animal safety data such that FDA cannot assess human risk, a proposed clinical or starting dose not adequately supported by the nonclinical data, and missing pharmacology/toxicology studies the Agency considers necessary for its safety assessment ¹. FDA's IND-content expectations reinforce this: the pharmacology/toxicology section must include an integrated summary of the toxicologic effects, full data tabulations suitable for detailed review, adequate information on all animal safety studies whether adverse or supportive, and, for each study subject to Good Laboratory Practice regulations, a statement of GLP compliance or a brief explanation of any noncompliance ^{157162158166168160}.

CMC deficiencies

Chemistry, manufacturing, and controls information for a Phase 1 IND is graded to the phase, but it must still assure the drug's identity, quality, purity, and strength, with the initial submission focused on what is needed to evaluate subject safety ³⁹⁴⁰. FDA is explicit that a safety concern or insufficient data to evaluate safety is the only CMC basis for a clinical hold, and it names the recurring triggers: a product made with unknown or impure components, a structure of known or highly likely toxicity, a product that cannot remain chemically stable through the proposed study, or an impurity profile that indicates a potential health hazard or is too poorly defined to assess one ³⁹. Supporting elements FDA expects include the product and dosage-form description, route of administration, sterilization method where relevant, manufacturing process and container-closure information that can affect quality, and stability data adequate for the planned trial duration ⁵⁰⁵¹.

Deficient IND information and the investigator's brochure

Beyond nonclinical and CMC content, a hold can rest on the IND simply not containing adequate information to let FDA assess whether the risks are reasonable, and on an investigator's brochure that is misleading, erroneous, or incomplete, for example one that understates known toxicities or omits relevant nonclinical findings ^101104187.

How the pieces fit for a reviewer

The four elements are not independent checklist items; they are a single argument. The nonclinical toxicology defines the toxicity that sets the starting dose; the starting-dose method (NOAEL/HED/safety factor, or STD10/HNSTD for cytotoxics, or MABEL for pharmacology-driven biologics) has to be justified from that toxicology; the escalation and dose-optimization plan has to protect subjects while generating the dose- and exposure-response data FDA now expects under Project Optimus; and the clinical-hold grounds are, in effect, the failure modes of that argument, an unjustified dose, a thin toxicology package, an unstable or impure product, or an IND that simply does not give FDA enough to conclude the risk is reasonable. A sponsor who can trace a clean line from the general toxicology study through the starting-dose derivation to a protocol with clear DLT definitions, stopping rules, and independent safety oversight has addressed the same questions a hold would otherwise ask.