11 Early-Phase Trials and Design Frameworks to Reduce Design Risk

The average likelihood of approval for a drug that enters Phase I is now 6.7%, according to a 10-year Citeline analysis covering 2014 through 2023 [1]. That number has declined in every successive analysis period. Citeline's analysis attributes declining success rates to multiple interacting factors: programs targeting increasingly difficult mechanisms, growing Phase II attrition, and the rising biological complexity of modern therapeutic hypotheses. Within that broader picture, protocol-level decisions made at the earliest stages of development, specifically dose selection, endpoint strategy, and patient population definition, remain among the variables that development teams can actually control.

Twenty-eight percent of clinical trials were cancelled during Phase II in 2023, up from a long-term average of roughly 20% before the pandemic [2]. A portion of those failures trace to protocol decisions that were made, or deferred, during Phase I. Endpoint selection, dose optimization, biomarker strategy, and patient population definition are all choices made at Phase I that set the probability of Phase III success. When they are made with precision early, programs move more efficiently. When they are deferred, the program carries that uncertainty into more expensive stages.

The 11 trials and design categories examined here were selected because each embodies a deliberate structural choice that gives clinical development teams something concrete to learn from: adaptive phase transitions, biomarker-guided population refinement, evidence-based dose selection, master protocol infrastructure, or careful treatment window timing. None of these is theoretical. All are active programs or recently published design frameworks with direct relevance to how early-phase protocols should be written now. Three of the 11 entries are design frameworks or design categories rather than single trials (entries 5, 6, and 11); they are included because the published methodology and regulatory context they represent are as practically informative as any individual program.

Evidence note: Several trial-specific entries in this article rely on sponsor press releases, company announcements, or conference coverage as primary sources, because independent peer-reviewed publications are not yet available for programs in active early-phase development. These sources are labeled throughout. Claims drawn from sponsor-reported data should be interpreted as preliminary and company-provided until peer-reviewed results are published.

Note: The term "trial" is used broadly in this article to include both individual clinical trials and design frameworks or methodological categories (entries 5, 6, and 11 in the table). The three framework entries are included because their published evidence and regulatory context are as operationally informative as any individual program.

At a Glance: 11 Trials and Design Frameworks

Why Early-Phase Design Decisions Carry Downstream Consequences

The highest attrition point in drug development sits between Phase II and Phase III, where approximately 70% of programs fail to progress [3]. A 2020 analysis published in Drug Discovery Today, using supervised machine learning across protocol data from ClinicalTrials.gov, found that protocol characteristics visible at early phases, including the number of endpoints and the complexity of eligibility criteria, were predictive of phase transition outcome with approximately 80% accuracy [4]. That result has a practical implication: late-phase failures are not all unpredictable. Many reflect early-phase protocol decisions that, in retrospect, embedded structural risk before the first patient was enrolled.

FDA nonbinding guidance finalized in August 2024 under Project Optimus [5] makes this dynamic concrete for oncology. The guidance states that the traditional maximum tolerated dose model does not adequately characterize grade 1 and grade 2 symptomatic toxicities, dose modifications, pharmacokinetics, pharmacodynamics, or dose-response relationships when selecting doses for subsequent trials [5]. While FDA guidance is nonbinding as a matter of regulatory process, inadequate dose justification at the End-of-Phase-2 meeting can generate requests for additional comparative dosing data before Phase III can proceed. An analysis of 367 industry-sponsored Phase I protocols activated at eight Sarah Cannon Research Institute drug development units between 2021 and 2024, published by Bhamidipati et al. in JCO Oncology Advances, found that Bayesian designs increased from 48% of protocols in 2021 to 75% in 2024 (P<0.001), reflecting a shift from traditional 3+3 escalation toward Bayesian methodologies such as BOIN [6]. Consistent with that finding, reporting from Targeted Oncology confirmed a broad increase in Phase I protocols explicitly including a dose optimization plan over the same period, from under 20% of programs in 2021 to the vast majority by 2024 [7].

The 11 Early-Phase Trials Teaching Us How to Design Better Studies

1. AFFINITY DUCHENNE (RGX-202): Integrated Phase I/II/III Design and Pre-Negotiated Surrogate Endpoint

REGENXBIO's AFFINITY DUCHENNE trial (NCT05693142) enrolled ambulatory boys aged one and older with Duchenne muscular dystrophy across three integrated phases, with the pivotal dose, 2 x 10^14 GC/kg, selected from Phase I/II safety and biomarker data before the Phase III portion opened [8]. This is sponsor-reported data; independent replication has not yet been published in peer-reviewed literature.

The pivotal primary endpoint, the proportion of participants achieving at least 10% microdystrophin expression at 12 weeks post-treatment, was accepted by FDA as a surrogate endpoint reasonably likely to predict clinical benefit, established through formal End-of-Phase-2 alignment for the accelerated approval pathway. In May 2026, the company reported that the trial met that endpoint with high statistical significance (p<0.0001) in 93% of 30 enrolled participants [9]. Earlier Phase I/II functional data, also sponsor-reported, demonstrated that treated participants exceeded natural history benchmarks on the North Star Ambulatory Assessment [10].

The design lesson here is specific and transferable. By establishing the surrogate endpoint through formal FDA alignment before the Phase III portion opened, the program avoided the need to run a separate confirmatory trial before accelerated approval consideration. That pre-Phase III regulatory engagement, often underinvested in smaller development programs, determined the entire design's efficiency. Programs that engage FDA on endpoint and dose selection strategy before Phase III enrollment begins are positioned to avoid the additional evidence generation that might otherwise be required at End-of-Phase-2 review.

2. QUADvance (AVC-203): Allogeneic Dual-Targeting CAR-T and the Antigen Escape Design Problem

AvenCell Therapeutics dosed the first patient in its Phase I/II QUADvance study (NCT07284433) in April 2026, evaluating AVC-203, a CRISPR-engineered allogeneic CAR-T candidate targeting both CD19 and CD20, in relapsed or refractory B-cell malignancies [11]. The company received FDA IND clearance and EMA CTA approval for this study on first submission in late 2025 [11].

The dual-targeting approach addresses one of the documented failure modes of single-antigen CAR-T therapies: antigen escape, where tumor cells downregulate the targeted surface receptor and evade treatment. The structural decision to engineer simultaneous CD19 and CD20 targeting into the construct is not embellishment. It is a response to a known, observed clinical failure mode in the preceding generation of approved therapies.

For protocol designers across modalities, the question this trial poses is generalizable: what are the known failure modes of this mechanism, and can the Phase I study design explicitly stress-test resistance to those modes? Waiting until Phase II to discover an escape or resistance mechanism is an avoidable design error in any program where preclinical or competitive data already suggest the risk.

3. CAR-PRISM (Ciltacabtagene Autoleucel in Smoldering Myeloma): Treatment Window Selection as a Design Variable

The CAR-PRISM trial (NCT05767359) is the first prospective study of CAR-T therapy as primary intervention in a precursor plasma cell disorder. Results were published in Nature Medicine in April 2026 and presented at AACR 2026 [12a]. The Phase 2, single-center study enrolled 20 patients with high-risk smoldering multiple myeloma (identified using the 20/2/20 model) and delivered a single cilta-cel infusion after lymphodepleting chemotherapy, without any induction therapy. All 20 patients achieved measurable residual disease (MRD) negativity at 10^-6 within two months and remained MRD-negative at a median follow-up of 15.3 months. No dose-limiting toxicities were observed, and no progression or deaths occurred during the study period [12a].

The underlying hypothesis is immunological: the immune system in smoldering disease is more intact than in heavily pretreated symptomatic myeloma, and therefore more capable of cooperating with CAR-T activity. Early data suggest this may be correct, though the study is small and results require prospective confirmation in larger trials.

What CAR-PRISM makes visible is that treatment window selection, meaning where in disease course to conduct first proof-of-concept work, has direct consequences for what the data can tell you. Many Phase I trials enroll heavily pretreated patients partly for ethical reasons and partly because that population is available. That enrollment decision shapes the therapy's observable ceiling. When a mechanism depends on intact immune function, systematically enrolling immunocompromised patients at proof-of-concept may produce underestimates of efficacy that drive premature program discontinuation. CAR-PRISM's 100% MRD negativity rate at 10^-6 in smoldering disease, a setting no prior CAR-T trial had entered, makes this point empirically rather than hypothetically [12a].

4. RaPTR-101 (RPTR-1-201): TCR Bispecific Therapy in Solid Tumors, and the Phase 1/2 Patient Selection Hypothesis

Repertoire Immune Medicine's RaPTR-101 trial (NCT07293754) opened in December 2025, evaluating RPTR-1-201, a T cell receptor bispecific therapy, in advanced solid tumors [13]. The study is currently in active recruitment, with an estimated primary completion in December 2028.

TCR bispecific therapies occupy a distinct immunological niche: they engage endogenous T cells through the T cell receptor complex rather than engineering patient-derived cells, and they can target intracellular antigens that conventional antibody-based therapies cannot access. Phase I work in this class carries outsized importance because dose-response and toxicity relationships for TCR bispecifics in solid tumors are not yet well characterized.

The specific design value of this trial is in what its early data will teach about patient selection. Solid tumor cell therapy programs have historically enrolled molecularly heterogeneous populations where efficacy signals are diluted across patients with different tumor biology. RaPTR-101's explicit tumor biology selection criteria are designed to test whether any observed signal is mechanism-specific or whether it reflects patient-selection confounding. That distinction will determine how the Phase 2 expansion should be defined and will provide reference data for the broader class.

5. Bayesian BOIN Trials Post-Project Optimus: The Shift from 3+3 to Evidence-Based Dose Selection

Not a single trial but a design category that Phase I oncology programs must now address explicitly, the move from traditional 3+3 dose escalation to Bayesian Optimal Interval (BOIN) and related model-based designs represents the most structurally significant shift in early-phase oncology trial design in recent years.

The 3+3 design, still used in 73.85% of Phase I oncology trials reviewed in a systematic literature analysis covering 2020 to 2022 [14], treats dose-limiting toxicity at a single threshold as the primary decision criterion. FDA's Project Optimus guidance, finalized in August 2024 and explicitly framed as nonbinding guidance, states that this approach does not capture the totality of data, including PK/PD, biomarker evidence, and low-grade symptomatic toxicity, needed to identify an optimal biological dose [5]. The guidance recommends sponsors compare multiple dose levels for safety, tolerability, pharmacokinetics, and pharmacodynamic biomarkers, either prior to or as part of a registrational trial [5].

More than half of the Phase I oncology trials reviewed in the 2020 to 2022 literature did not reach a maximum tolerated dose [14]. In those programs, the traditional design's central organizing principle was not even informative. Sponsors designing Phase I programs today without a documented dose optimization rationale may encounter questions at End-of-Phase-2 meetings that require additional comparative dosing studies before Phase III is authorized. That is a timeline and cost consequence, not a regulatory sanction, but it is a consequence that better Phase I design prevents.

6. Adaptive Biomarker-Guided Enrichment Trials (Serra et al., 2025): Population Refinement Without Early Commitment

A methodological study published in Statistics in Medicine in October 2025, by Serra and colleagues at the University of Cambridge MRC Biostatistics Unit and collaborating institutions, proposed a one-arm two-stage early-phase biomarker-guided design [15]. The design allows, at a pre-specified interim analysis, a decision to stop the trial, continue enrollment in the full biomarker-unselected population, or narrow enrollment to a biomarker-positive subgroup. Simulation results showed that the approach produces better decision-making than a classical design that ignores subgroup structure, even when sample sizes at the interim are small [15].

A 2024 review published in Therapeutic Innovation and Regulatory Science by Tu and Renfro documented that molecular tumor characterization has substantially changed the practical requirements for early-phase trial design in oncology, requiring frameworks that can test biomarker subgroup hypotheses without locking eligibility criteria before any clinical pharmacodynamic data exists [16].

For sponsors, this work is not just methodological. Locking biomarker eligibility too early, before any pharmacodynamic data from the study is available, can produce over-inclusion (enrolling patients unlikely to respond, diluting the efficacy signal) or over-exclusion (missing the active population because the biomarker threshold was set without clinical evidence). Designs that pre-specify adaptive enrichment rules before enrollment opens, with explicit interim analysis criteria, are already an expectation in some FDA precision oncology guidance documents.

7. ADI-100 (Adicet Bio): CAR-Gamma-Delta T Cells in Autoimmune Disease and the Modality Translation Problem

Adicet Bio dosed the first patient with ADI-100 (prulacabtagene leucel), an allogeneic chimeric antigen receptor gamma-delta T-cell therapy, in a Phase 1 study (NCT06375993) in November 2024, targeting autoimmune diseases including lupus nephritis, systemic lupus erythematosus, systemic sclerosis, and related conditions [17, 28]. In October 2025, the company reported positive preliminary safety and efficacy data from the trial in lupus nephritis and SLE patients (data cutoff August 31, 2025), citing sponsor-reported reductions in disease activity scores and improved renal function with a favorable safety profile; all seven autoimmune disease cohorts were actively enrolling as of early 2026, with a full clinical update targeting mid-2026 [33].

The design challenge this trial highlights is modality translation: moving a cell therapy from oncology, where it was originally developed and where the safety threshold framework was established, into a non-oncology population where acceptable toxicity is defined by a fundamentally different risk-benefit calculus. In oncology, grade 3 adverse events may be accepted in exchange for tumor response. In autoimmune disease, a patient who might otherwise manage their condition for decades faces a risk-benefit assessment that changes the DLT definitions, stopping rules, patient-reported outcome requirements, and the monitoring frequency that the Phase 1 protocol must specify.

Phase 1 designs in cell therapy for autoimmune disease must address this population-specific safety threshold question in the protocol itself, not leave it for the safety monitoring committee to resolve on a case-by-case basis during enrollment. ADI-100 could provide an early reference dataset for allogeneic CAR-T safety in this population that other autoimmune cell therapy programs can use to calibrate their own DLT framework and stopping-rule assumptions.

8. RIDGE-1 (TN-401, Tenaya Therapeutics): Cardiac Gene Therapy and the Sequential Dosing Requirement

Tenaya Therapeutics dosed the first patient in the Phase 1b RIDGE-1 trial (NCT06228924) in November 2024, evaluating TN-401, an AAV-based gene therapy, for PKP2-associated arrhythmogenic right ventricular cardiomyopathy (ARVC) [17]. The first patient received 3 x 10^13 vg/kg, with sequential dosing of two additional patients at that level before any potential escalation, pending independent safety review committee evaluation.

RIDGE-1 captures a design challenge that is specific to cardiac gene therapy: the delivery mechanism and the target organ are the same, and both must survive the intervention. The gene therapy must achieve sufficient transduction of cardiac muscle cells to produce functional expression, while the immune response to the viral vector must not damage cardiac tissue in the process. That narrow margin between therapeutic effect and organ toxicity, which cannot be fully characterized from preclinical models, is the reason the protocol requires sequential patient dosing with mandatory independent safety review before escalation.

For sponsors developing early-phase gene therapy programs in rare cardiac indications, RIDGE-1's design, its cardiac monitoring schedule, safety committee structure, biomarker panel, and dose escalation criteria, will be informative even before TN-401's data are published. The regulatory expectations for cardiac monitoring and immune response documentation in Phase 1 gene therapy are becoming clearer with each new program.

9. STAAR (Sangamo's ST-920): Phase 1/2 Surrogate Endpoint Negotiation for Accelerated Approval

Sangamo Therapeutics reached alignment with FDA following a Type B interaction in late 2024, pursuing accelerated approval for isaralgagene civaparvovec (ST-920), an AAV-based gene therapy for Fabry disease, using one-year post-treatment eGFR slope from the Phase 1/2 STAAR trial (NCT04046224) as an intermediate clinical endpoint [17, 29]. The Phase 1/2 STAAR study enrolled 32 patients, all of whom have rolled into long-term follow-up [35]; positive mean annualized eGFR slope data at 52 weeks across all dosed patients were presented at WORLDSymposium in February 2026, with company data suggesting improvement in renal function [35]. Sangamo initiated a rolling BLA submission in December 2025 and reported advancing preclinical and clinical modules to FDA in March 2026 [34]. All these are sponsor-reported milestones.

The design relevance is in how that negotiation became possible and what early-phase data supported it. FDA's acceptance of eGFR slope as an intermediate endpoint was contingent on the quality of the biomarker data from the Phase 1/2 study, the consistency of the treatment effect across enrolled patients, and the biological plausibility of the surrogate-to-clinical-outcome relationship. None of those conditions were established at study initiation. They were built through the conduct of the Phase 1/2 trial itself.

Programs in rare disease often cannot enroll the patient numbers required for conventional randomized Phase 2 trials. Understanding what endpoint structure, surrogate, clinical, or composite, will support regulatory engagement before Phase 2 design is locked can compress the timeline to approval by years. STAAR's path through that negotiation is a live reference point for other gene therapy and rare metabolic disease programs where similar discussions are still in early stages.

10. ComboMATCH and MyeloMATCH: NCI Master Protocols and the Infrastructure Requirements of Multi-Arm Design

ComboMATCH, launched by the National Cancer Institute in 2023 as a direct successor to NCI-MATCH, is a group of precision medicine Phase 2 treatment trials testing drug combinations guided by tumor genetic alterations rather than single agents [18]. NCI launched the program specifically because NCI-MATCH identified a structural limitation: single targeted agents produced limited durable responses in heavily pretreated patients, and combination approaches require different evidence-generation infrastructure [19]. Each ComboMATCH treatment trial evaluates a drug combination, typically two targeted drugs or a targeted drug plus chemotherapy, against patients selected by specific cancer genetic alterations [18].

MyeloMATCH, the companion umbrella trial for newly diagnosed AML and MDS, had enrolled approximately 600 patients as of a report from the Frederick National Laboratory for Cancer Research, with interim molecular profiling data presented at ASH 2025 covering the first 400 enrolled patients [20, 21].

What these programs demonstrate for sponsors designing Phase I/II programs for targeted combinations is specific and operational. The design requirements include pre-specified molecular eligibility criteria with standardized assay validation, shared infrastructure for cohort-level futility rules, coordination of companion diagnostic development across sub-studies, and version control of eligibility criteria as protocol amendments are filed over time. Site selection for multi-arm programs also requires identifying investigator sites with the NGS infrastructure and patient throughput to support multiple cohort-specific eligibility streams simultaneously; tools like KScout (Kitsa's site selection platform) are designed for exactly this kind of biomarker-matched site feasibility assessment. Any sponsor considering a basket or multi-arm early-phase program faces these same documentation and infrastructure challenges. Watching how the MATCH successor programs address them at scale is informative before committing to a similar architecture.

11. Integrated Early-Phase Protocols with Pre-Specified Transitions: The German Multi-Stakeholder Consensus (2026)

A January 2026 consensus document published in Frontiers in Pharmacology, produced by a multi-stakeholder group spanning German academic, regulatory, and industry institutions, published concrete operational guidance on designing integrated early-phase protocols that reduce substantial modifications [22]. Substantial modifications are the protocol amendments filed when a single study needs to transition from healthy volunteers to patients, or from dose-finding to dose expansion, and those amendments were not explicitly anticipated in the original protocol.

The consensus identifies the core problem: frequent substantial modifications slow development timelines, reduce regulatory transparency, and create documentation burden across the trial master file. The EMA first-in-human guideline supports pre-defined transitions within a single integrated protocol, but only when the transition criteria, safety thresholds, and decision rules are explicitly documented in the original submission [22]. The German guardrail concept clarifies when a pre-specified transition does not require a new substantial modification notice, giving sponsors operational planning certainty.

What this consensus makes concrete is that integrated Phase I-to-Phase I/II transitions are not just a statistical methodology choice. They require specific protocol architecture decisions made at the time of the initial IND or CTA submission. Governance structures, safety monitoring committee mandates, and adaptive decision criteria must be written into the protocol before the first patient is enrolled, or the regulatory pathway for adaptation narrows considerably once the trial is underway.

Common Design Principles Across These Eleven Programs

Several themes run through these trials that apply regardless of therapeutic area or modality.

Biomarker decision timing. Nearly every program uses a biomarker, either as an eligibility criterion, an adaptive decision variable, or a surrogate primary endpoint. The design question in each case is not whether to use a biomarker, but when in the development process the biomarker relationship is characterized, validated, and embedded in the protocol's decision architecture. Early commitment to a biomarker threshold that has no clinical pharmacodynamic support is a structural risk.

Regulatory alignment before Phase 2. AFFINITY DUCHENNE, STAAR, and the Project Optimus-influenced BOIN programs each illustrate the value of End-of-Phase-2 or Type B meetings as active design inputs rather than administrative milestones. They are data-informed negotiations about what evidence standard the program is building toward. Sponsors who enter those meetings with pre-specified endpoint rationales, dose optimization data, and biomarker pharmacodynamic evidence are in a stronger position to avoid requests for additional studies before Phase III proceeds.

Population definition as a testable scientific hypothesis. Whether a trial enrolls broadly (to detect signals in any subpopulation) or narrowly (to maximize signal clarity in a preselected group) is a scientific decision with downstream consequences. The biomarker enrichment literature and the adaptive designs reviewed here are both attempts to make that decision empirically rather than arbitrarily.

Protocol architecture as the first risk mitigation tool. The German multi-stakeholder consensus makes explicit what experienced clinical development teams already understand: a protocol that does not pre-specify its own adaptive decision rules is a protocol that will face amendment burden. Amendments are not only operational delays. They create regulatory documentation complexity that cascades through the entire trial master file.

Regulatory and Documentation Considerations

ICH E6(R3), adopted by ICH on January 6, 2025, came into effect for EU clinical trials on July 23, 2025 following formal EMA adoption [23, 23a]. FDA issued its companion final guidance on E6(R3) in September 2025 [27]. The guideline restructures GCP into overarching principles, Annex 1 for interventional trials, and a forthcoming Annex 2 for non-traditional designs. E6(R3) emphasizes quality-by-design: sponsors are expected to document quality management plans, risk assessments, stopping rules, and Bayesian thresholds prospectively in the protocol and statistical analysis plan before trial initiation, not reconstructed during or after conduct. This does not create a blanket requirement that every adaptive feature be formally approved in advance, but it does mean that pre-specified adaptations are treated differently from post-hoc modifications under inspection.

FDA's Project Optimus guidance, finalized in August 2024, is explicitly framed as nonbinding guidance [5]. However, sponsors designing Phase I oncology programs without a documented dose optimization rationale face the practical risk of being asked to generate additional comparative dosing evidence at End-of-Phase-2 review before Phase III may proceed.

The EU Clinical Trials Regulation (CTR), which first became applicable on January 31, 2022, completed its three-year transition period on January 31, 2025 [24]. As of that date, all clinical trials in the EU, including ongoing trials previously authorized under the Clinical Trials Directive, were required to operate under the CTR and submit through the CTIS portal. Any early-phase program planned for EU sites must now account for CTIS submission requirements from the initial CTA stage.

FDA finalized its guidance on conducting clinical trials with decentralized elements in September 2024 [25]. Early-phase trials incorporating remote monitoring, digital biomarkers, or electronic patient-reported outcomes must document their validation approach in the protocol and specify how those data sources will be integrated into safety and pharmacodynamic assessments.

Operational Impact for Sponsors, CROs, and Sites

The designs reviewed here impose specific operational demands that differ from conventional Phase I studies.

Integrated Phase I/II/III structures require sites to be qualified and contracted for all three phases before enrollment begins, because pre-specified transition criteria may be met sooner than expected and site activation delays at transition cannot be introduced. CROs managing these programs need to review master service agreement structures and site activation workflows against the possibility of early-phase transitions.

Biomarker-adaptive enrichment designs require assay readiness: analytical validation of the biomarker assay, defined specimen handling protocols, specified turnaround times from collection to result, and a pre-specified plan for what happens when biomarker data is unavailable for a patient at the interim analysis window [16]. These requirements must be documented before enrollment opens.

Bayesian dose designs require biostatisticians who can implement and monitor them in real time, and data management systems that can feed accumulating dose-toxicity data into the model continuously rather than in cohort batches. A site's data entry timelines, which may have been adequate for 3+3 cohort review schedules, may not support continuous Bayesian updating. CROs supporting Bayesian adaptive Phase I programs should audit data entry SLAs before study start.

Master protocol programs, including any basket or multi-arm Phase I/II structure, create version control and eligibility documentation challenges that cascade into the eTMF. Regulatory inspectors reviewing a multi-arm early-phase program will expect clean, traceable eligibility criteria documentation for each cohort from original protocol through each amendment. Maintaining that consistency across adaptive protocols, IBs, and ICFs is a core use case for protocol generation tools like KScribe (Kitsa's regulatory document generation platform).

AI and Automation Perspective

Machine learning models trained on clinical trial protocol data have demonstrated approximately 80% predictive accuracy for phase transition outcomes in published research [4]. Those models are most useful as decision support tools when a development team is comparing design configurations before a trial opens. They require historical training data and expert interpretation; they do not substitute for the biostatistical and regulatory judgment needed to write a protocol.

In early-phase dose selection, model-informed approaches integrating PK/PD measurements, biomarker data, and safety observations into dose-optimization frameworks are now supported both by published methodology and by FDA-AACR workshop guidance on implementing Project Optimus principles [26]. These quantitative approaches, including the PEDOOP design and related pharmacometrics-enabled dose optimization frameworks, help identify optimal biological doses by integrating the multi-dimensional data that Project Optimus guidance requires, in a way that is difficult to accomplish with traditional cohort-level review.

For protocol authoring, the challenge in early-phase programs is maintaining consistency across multiple adaptive decision criteria, cohort-specific eligibility requirements, and evolving regulatory cross-references across documents. AI-assisted drafting tools must be able to handle the bidirectional dependencies between decision criteria, amendment history, and guidance citations that accumulate in complex early-phase programs. All outputs require human expert review, particularly for regulatory language distinguishing when a transition is pre-specified (not a substantial modification) versus when it constitutes an amendment requiring submission.

How Kitsa Fits Into This Problem

Protocol documentation for complex, adaptive early-phase designs is not a routine drafting task. KScribe, Kitsa's regulatory document generation tool, is designed to support protocol authors working on eligibility criteria, adaptive decision rule language, biomarker decision architecture sections, and cross-document consistency between protocols, IBs, and ICFs, precisely the areas where documentation ambiguity in early-phase programs generates amendment burden and regulatory questions at End-of-Phase-2 meetings.

Key Takeaways

• The average likelihood of FDA approval for a drug entering Phase I is now 6.7%, driven by multiple factors including biological complexity, Phase II cancellations reaching 28% in 2023, and the increasing difficulty of targeted therapeutic hypotheses [1, 2].
• FDA's Project Optimus final guidance (August 2024), though nonbinding, has measurably shifted Phase I design practice: Bayesian designs increased from 48% to 75% of protocols at eight SCRI drug development units between 2021 and 2024 (P<0.001), reflecting a shift from 3+3 toward Bayesian methodologies such as BOIN [6].
• Integrated Phase I/II/III designs, exemplified by AFFINITY DUCHENNE, depend on pre-Phase III FDA regulatory alignment on surrogate endpoint validity; that alignment is not achieved by waiting for interim efficacy data.
• Biomarker enrichment timing is a design decision, not a protocol afterthought. Pre-specified adaptive enrichment rules allow population refinement at interim analysis without committing to a fixed threshold before any pharmacodynamic data exists [15].
• New modalities entering non-oncology populations (CAR-T in autoimmune disease, gene therapy in cardiac indications) require DLT framework redefinition tailored to the acceptable risk threshold of that population, which differs materially from oncology standards.
• NCI master protocol programs (ComboMATCH, MyeloMATCH) demonstrate that multi-arm early-phase designs are operationally feasible at scale, but they impose specific demands on assay infrastructure, version control, and eligibility documentation that must be addressed in the protocol.
• ICH E6(R3) (effective EU July 23, 2025; FDA September 2025) expects stopping rules, Bayesian thresholds, and risk assessments to be documented prospectively in the protocol and SAP before enrollment begins; pre-specified adaptations are treated differently from post-hoc modifications under inspection [23, 23a, 27].