For targeted agents, treating the MTD as the dose ceiling can carry a compound into Phase 2 at a dose that adds toxicity without maximizing benefit. Project Optimus, formalized by FDA in 2021, challenges four decades of oncology convention by separating tolerability from clinical optimization. The distinction is especially important when exposure-response is non-monotonic or plateaus below the MTD, as often occurs with targeted agents and immunotherapies. Most ongoing Phase 1 studies were not built to answer that question.
The MTD Paradigm and Where It Breaks Down
The 3+3 design and model-based approaches, including BOIN, mTPI, and the continual reassessment method, emerged largely from cytotoxic chemotherapy, where dose and cell kill tend to rise together across the therapeutic range. For pathway-specific compounds, additional drug may not increase target engagement. Once dose level 3 saturates the pathway, levels 4, 5, and 6 can add toxicity without added efficacy. A Phase 1 that reaches the MTD and names it the recommended Phase 2 dose has found a ceiling, not necessarily the optimum.
The optimal biological dose question is not new. For more than a decade, publications on immunomodulatory agents have shown response plateaus at exposures below the MTD, while the MTD generated more immune-related adverse events without improving tumor response. Project Optimus changed the regulatory expectation. FDA now asks sponsors to define the dose-response relationship in dedicated dose-optimization studies, often expansion cohorts within Phase 1/1b, before fixing the Phase 2 dose. Many programs therefore need parallel dose cohorts rather than one escalation ladder ending at MTD.
This requirement lengthens an already demanding development path. A Phase 1 that finds the MTD and immediately expands that dose into Phase 2 is insufficient for many oncology indications. Sponsors must anticipate what dose-optimization cohorts will resolve and determine cohort sizes before enrollment. Four cohorts of 15 patients each require substantial time and cost to define dose response. The design question is whether the most informative cohort structure can be established before the first patient enters.
Patient Heterogeneity Across the Development Timeline
Population variability creates a second problem at the Phase 1 to Phase 2 transition. Phase 1 oncology participants are typically heavily pretreated and have few remaining options. Patients entering Phase 2 may have fewer prior treatment lines and materially different pharmacokinetics. Reduced renal function after platinum therapy, hepatic enzyme induction from anticonvulsants, and albumin changes that alter protein binding for highly bound compounds can each shift exposure at the same nominal dose.
A Phase 2 dose based on Phase 1 MTD data from patients with a median of three to four prior lines may yield different exposures in a Phase 2 population with a median of one to two prior lines. The nominal dose remains unchanged, but the pharmacokinetic result may not. Population PK models that omit this shift can underestimate Phase 2 exposure, leading to inadequate dosing or tolerability patterns absent from Phase 1 expansion data.
Without explicit population PK analysis across this transition, the selected Phase 2 dose assumes that the two populations are comparable. That assumption may fail. Phase 1 tolerability findings remain clinically meaningful, but their relevance to the intended Phase 2 population's PK behavior must be tested rather than inferred from equivalence between heavily and less heavily pretreated patients.
Seamless Phase 1/2 Designs: Efficient but Demanding
Adaptive seamless Phase 1/2 designs provide one response. BOIN-ET extends dose escalation by combining early efficacy information with toxicity, so both endpoints inform dose selection. Instead of taking the highest tolerable dose into a separate Phase 2, the design seeks the dose most likely to satisfy an efficacy criterion while keeping toxicity acceptable. The statistical framework is established, and FDA has accepted such approaches when implemented with prospective type I error control.
The operational burden is substantial. In a seamless design, the Phase 2 protocol generally must be nearly complete before Phase 1 enrollment begins because both stages use shared infrastructure, randomization systems, and statistical analysis plans. That differs from common practice, in which Phase 2 planning starts after Phase 1 DLT data arrive. Without an upfront decision to use a seamless design, retrofitting one during Phase 1b expansion can require major amendments and create timeline consequences.
Where Simulation Addresses the Gap
Virtual patient simulation can inform the decision before the Phase 1 to Phase 2 transition. Using compound PK parameters from preclinical work and early Phase 1 data, a model can estimate exposure distributions in a virtual population resembling the intended Phase 2 cohort. Pharmacodynamic models based on mechanism and available biomarkers can then estimate response probability across doses as a distribution, rather than a single point estimate.
In QuantHealth's early-access pilot, a team developing a targeted kinase inhibitor for second-line solid tumors planned to advance 200 mg from Phase 1 expansion into Phase 2. Simulation of population PK in virtual patients with Phase 2-typical treatment burden and organ function found that 200 mg exceeded the target EC90 in approximately 85% of patients. At 150 mg, equivalent target engagement coverage reached 78%. The 50 mg difference offered no meaningful population-level efficacy probability advantage at the upper tail, but it shifted predicted Grade 3+ on-target adverse events in a way relevant to the planned safety run-in. Before protocol lock, the team changed the Phase 2 starting dose and DLT monitoring rules.
The model did not forecast the Phase 2 result. It assessed what the available PK/PD evidence implied about the exposure-response tradeoff between two candidate doses. That narrower question is more tractable and can be addressed before the first Phase 2 patient is enrolled.
Scope and Limits of Simulation in Oncology Dose Finding
Simulation does not replace Phase 1 clinical evidence. It extracts additional decision-relevant information from the data already generated. When Phase 1 provides preliminary PK and some pharmacodynamic signal, simulation can extend dose-response analysis across a broader virtual population and dose range than enrollment directly covered. Its strongest use is comparative, helping rank candidate doses rather than forecasting the absolute result at one dose.
Simulation cannot identify unexpected toxicities driven by mechanisms absent from the model. It also cannot characterize immune-related adverse events in immuno-oncology without explicit immune activation modeling or replace the direct safety observation supplied by DLT monitoring. A model expresses current knowledge from preclinical and Phase 1 evidence. Safety surprises are not part of that knowledge base and therefore will not appear in a model fixed at protocol lock.
Our methodological boundary is explicit: simulation outputs are probability distributions, accompanied by documented model assumptions, not forecasts of trial outcome. If the model gives a 70% probability of meeting the primary endpoint at the recommended dose, that 70% P(success) means the inputs and model structure imply success with that probability, conditional on the assumptions. It does not mean the trial itself has a 70% chance of success. Assumption uncertainty is quantified separately and reported so the clinical team can identify what requires validation before protocol lock.
Practical Questions for Protocol Design Teams
At the Phase 1 to Phase 2 dose decision, simulation can organize several questions: which Phase 1 candidate dose has the greatest probability of meeting the Phase 2 primary endpoint given the exposure-response evidence; how that probability changes when organ function differs between populations; whether the recommended dose's simulated response distribution supports the planned sample size compared with the original power calculation; and what change in the compound's PD profile would make the lower dose preferable.
Project Optimus makes dose-optimization an expected part of development for most new oncology compounds. Adding dose-optimization cohorts without specifying what they must distinguish can increase patient exposure and extend timelines without sharpening the decision. Simulation can define the dose contrasts, cohort sizes, and interim rules needed for expansion or stopping. Building that framework before Phase 1 is generally less costly than revising the protocol after Phase 1 data arrive.