Project Optimus: Why Oncology Dose Finding Moved Beyond the Maximum Tolerated Dose

Project Optimus is the US Food and Drug Administration Oncology Center of Excellence initiative that changed the question behind oncology dose finding. The old question was often, “What is the highest dose patients can tolerate?” The newer question is, “Which dosage has the most favorable balance of activity, safety, and tolerability, and what evidence supports that choice?”

The distinction is not semantic. A maximum tolerated dose can be above the point where a molecular target is already saturated or antitumor activity has plateaued. Additional drug may then add adverse effects, dose interruptions, or discontinuations without a corresponding improvement in benefit, and the FDA's August 2024 final guidance sets out a more complete development strategy, commonly including randomized comparison of two or more dosages before approval.

Why the maximum tolerated dose became the default#

Traditional cytotoxic chemotherapy often has a steep relationship between dose and both tumor killing and toxicity. Early phase studies escalated doses in small groups until prespecified severe toxicities appeared. The result was an estimate of the maximum tolerated dose, which frequently informed the recommended phase 2 dose.

That strategy was pragmatic for its era, but several assumptions do not transfer neatly to targeted agents, antibodies, and some immune therapies. Biological activity may reach a plateau before severe toxicity. Toxicity may accumulate over months rather than appear within a short first-cycle window. A dose that looks manageable in a small, selected early phase group may become difficult to sustain in a broader population or in combination treatment.

The maximum tolerated dose is therefore a safety landmark, not proof of the best benefit-harm balance. Project Optimus asks drug developers to characterize the full dose-response relationship rather than treating escalation as a race to one upper boundary.

What an optimized dosage means#

An optimized dosage is supported by the totality of evidence for the intended use: how much drug reaches the circulation over time, whether a biological target is affected, whether tumor activity changes, and which adverse effects occur. It also rests on two things that only show up later, which are how long treatment can be sustained and what patients report about their own symptoms and function.

No single endpoint automatically wins. A lower dose with slightly less early tumor shrinkage might not be preferable if disease control is materially worse, and a higher dose with a small activity difference might not be preferable if it causes frequent interruptions that reduce the amount of therapy patients can actually receive. The relevant tradeoff depends on disease setting, treatment duration, and alternatives. It depends on combination partners and the seriousness and reversibility of harms.

The FDA guidance also distinguishes a dose from a dosage. Dosage includes amount, frequency, and sometimes schedule or treatment duration. Two regimens can deliver the same nominal amount per cycle while producing different concentration patterns, recovery time, or tolerability.

Build the evidence across development#

Dose optimization is not a single trial placed at the end of a program. It starts with nonclinical pharmacology and first-in-human planning. It then develops through pharmacokinetic sampling, pharmacodynamic measures, and preliminary activity. It develops through safety and longer observation.

Early work should characterize more than dose-limiting toxicities during one short window. Reviewers need to see adverse events over time, cumulative effects, and reasons for missed or reduced doses. They need to see discontinuations and supportive care. The FDA Oncology Dosing Tool Kit highlights the value of describing serious and fatal events, events by grade, dose modifications, and the rationale for the starting dose and escalation range.

Pharmacokinetic analysis asks how administered dosage relates to concentrations in the body. Pharmacodynamic analysis asks what those concentrations do to a target or biological pathway. A biomarker can support dose choice when it is analytically sound and clearly linked to the drug's mechanism, but a biomarker plateau is not automatically a clinical efficacy plateau. Tumor response, time-to-event outcomes, and tolerability still matter.

Why randomized dose comparison is powerful#

Comparing outcomes from separate dose-escalation groups can be misleading, and people receiving a later dose may differ because eligibility changed, investigators learned how to manage toxicities, supportive care improved, or the mix of tumor types shifted. Small cohorts also produce unstable estimates.

Randomization makes the dose groups comparable on known and unknown baseline factors on average; it provides a cleaner estimate of how outcomes differ because of dosage rather than because of who happened to receive it. The 2024 guidance generally favors randomized evaluation when more than one dosage remains plausible.

A randomized dose-optimization study is not simply a miniature efficacy trial. Its design should specify what evidence will drive selection. That can include tumor response, duration of response, and time-to-event outcomes when feasible. It can include grade and timing of adverse events, dose modifications, and treatment discontinuation. It can include drug concentrations, biomarker change, and patient-reported tolerability. Selection rules should be prespecified enough to prevent a favorable narrative from being assembled after results are known.

Patient-reported outcomes add information clinicians cannot infer#

Clinician-graded adverse events are essential, but they can understate symptom frequency or burden. Fatigue, diarrhea, and neuropathy may determine whether long-term therapy is sustainable. So may pain, sleep disruption, and impaired daily function. Patient-reported outcome measures capture the patient's account directly and can separate two regimens that look identical to you in a conventional safety table.

Completion and missingness need close reading. If people stop answering questionnaires because one dose is difficult to tolerate, the responses that remain can make that dose look better to you than it was. The protocol should define collection timing, handling of missing observations, and how patient reports contribute to the dose decision.

Common interpretation errors#

“The lower dose had fewer adverse events, so it is optimal”#

Lower toxicity is only one side of the balance. The activity evidence must remain adequate for the intended disease and setting.

“The higher dose produced more responses, so it is optimal”#

A numerical response difference can be due to chance, short follow-up, or imbalanced baseline risk. Durability, serious harms, modifications, and uncertainty around the estimates must be included.

“No maximum tolerated dose was reached, so any high dose is acceptable”#

Failure to reach a toxicity boundary does not establish that further dose has value. Pharmacology and clinical outcomes may already have plateaued.

“Target saturation proves the clinical dose”#

Target activity can narrow the range, but assay limitations and incomplete links between a biomarker and patient benefit prevent it from deciding the dose by itself.

“Project Optimus means every approved dose should be reduced”#

The initiative concerns evidence generation and reasoned selection; it does not presume that lower dosing is universally equivalent, and it does not authorize anyone to change your cancer treatment without supervision.

Combinations and multiple indications are harder#

When two active drugs are combined, each drug's contribution and interaction matter. A tolerable combination may still contain more of one component than needed. Conversely, reducing one component could compromise the joint effect. Programs may need factorial, response-adaptive, model-assisted, or other efficient designs, but their assumptions must remain transparent.

The optimized dosage can also differ by cancer type, biomarker group, or combination partner. It can differ by organ function or treatment setting. FDA's research program begun in 2025 specifically addresses statistical design for combinations and multiple indications. That work signals an unresolved methodological area, not a single endorsed algorithm.

How to appraise an oncology dose claim#

Ask six linked questions before you accept a dose:

1. Which dosages and schedules were actually compared?
2. Were groups randomized, and were important baseline factors balanced?
3. Did the study observe patients long enough to capture cumulative toxicity and durable activity?
4. Were pharmacokinetic, pharmacodynamic, efficacy, safety, and patient-reported data considered together?
5. Were dose reductions, interruptions, discontinuations, and missing reports shown by assigned group?
6. Was the selection rule prespecified, and does the chosen dosage fit the population and indication in the proposed label?

If the only evidence is first-cycle toxicity, the answer is incomplete. If the only evidence is a response rate, it is also incomplete. The strength of Project Optimus is the insistence that dosage be justified as a multidimensional clinical decision.