FDA-AACR Strategies for Optimizing Dosages for Oncology Drug Products: Early Phase Trials Using Innovative Trial Designs and Biomarkers

Abstract

Dosage selection for oncology drugs has traditionally relied on initial dose-finding trials to determine a maximum tolerated dose (MTD), which is then further evaluated in approval-supporting registrational trials. While this approach may have established optimized dosages for cytotoxic chemotherapeutics, many modern oncology drugs developed through this approach have been poorly optimized, requiring additional dosage optimization efforts in the post-market setting. Recent initiatives of the U.S. Food and Drug Administration outline the unsustainability of this approach, instead recommending the identification of a potentially optimized dosage at earlier stages through direct comparison of multiple dosages before marketing application submission. The selection of dosages for further investigation outside of the MTD requires fit-for-purpose techniques that address the specific promises and concerns of the drug under investigation. Although such strategies have been developed, they are currently rarely applied in favor of the MTD paradigm. Innovative trial elements, including various integral, integrated, and exploratory biomarkers as well as backfill and randomized dose expansion cohorts represent potential avenues to create and leverage additional data, and thereby make more informed dosing decisions. Additionally, modeling approaches such as clinical utility index can integrate these disparate datatypes into a single metric, facilitating more quantitative selection. This article, the second in a series of three articles addressing different stages of dose optimization, outlines best practices and areas for further development regarding innovative techniques for the selection of dosages for further evaluation prior to final dosage selection for registrational trials in oncology.

Introduction

Oncology drug development programs have routinely utilized dose escalation trials conducted in a small number of patients to identify a maximum tolerated dose (MTD) based on dose-limiting toxicities. The identified MTD, derived from limited short-term clinical data, is frequently moved forward in larger clinical trials, including those that support drug approval. Many drug development programs that have taken this approach have achieved FDA approval, but with the caveat of necessary post-marketing commitments or requirements to identify an optimized dose for the approved indication (1,2). This approach is unsustainable, subjecting patients to inadequately characterized dosages while these fundamental studies take place. Notably, patient advocates have been calling for change to this MTD-centric paradigm in recent years, highlighting the need for doses that support both patient wellbeing and efficacy (3).

In 2023, the U. S. Food and Drug Administration (FDA) issued a draft guidance titled “Optimizing the Dosage of Human Prescription Drugs and Biological Products for the Treatment of Oncologic Diseases” to encourage a more deliberative approach to dose selection and optimization for oncology drugs. This guidance, finalized in 2024 (4), recommends the comparison of the activity, safety, and tolerability of multiple dosages either prior to or as part of a registrational trial to support the recommended dosage for a marketing application. While such an undertaking has been performed in registrational trials, such as in the development of the PD-1 inhibitor pembrolizumab (5), this can be associated with significant sample size and cost. A more pragmatic approach may be the evaluation of two or more active dosages in an earlier trial, but the identification of the dosage(s) to explore can be challenging given the limited clinical data available at initial stages of development.

In February 2024, the FDA Office of Clinical Pharmacology and American Association for Cancer Research (AACR) convened a workshop titled “Optimizing Dosages for Oncology Drug Products: Quantitative Approaches to Select Dosages for Clinical Trials” to discuss best practices and open questions in the field of optimizing dosages for oncology drugs (6). This article is the second in a series of three consolidating learnings from each session of the workshop, which covered selecting dosages for first-in-human (FIH) trials (7), selecting dosages for further exploration in early-phase trials, and selecting dosages for registrational trials (8), respectively. Herein, we highlight best practices and areas for future development for certain clinical trial design features and modeling approaches that have the potential to facilitate the selection of dosages for additional investigation prior to final dosage decision-making for the registrational trial in hopes of spurring increased awareness, usage, and regulatory science research in this area.

Novel Trial Design Features to Enhance Dose-Selection Decision Making

A shift towards innovative trial designs which maximize data gathered at various dosage levels in a deeper manner than the traditional MTD-centric 3+3 dose escalation design is needed to further enhance early-stage dose selection and optimization. There are many innovative FIH trial designs currently in use that are facilitating this shift, including model-based and model-assisted designs (9–11). For example, the Bayesian Optimal Interval (BOIN) design, a class of model-assisted dose finding design granted the FDA fit-for-purpose designation for dose finding in 2021 (12,13), allows the treatment of more than 6 patients at a dose level, the potential to return to a dose level multiple times if not excluded by the design or the safety stopping rules, and the ability to escalate and de-escalate across different dose levels via a spreadsheet design/table. The benefits of these designs have been thoroughly reviewed previously, and considerations for their usage are a central focus of the first manuscript in this series. However, the integration of certain clinical trial design features such as biomarkers, backfill, and expansion cohorts into innovative early-stage trial designs can also prove essential to dosing optimization efforts by providing even further decision-informing evidence (Figure 1) (14,15).

Figure 1. Trial design features to enhance dose selection.

(A) Schematic representation of the traditional 3+3 dose escalation design, which focuses on identifying the maximum tolerated dose (MTD). In this example, six dose levels (DL) were evaluated, with significant dose-limiting toxicities observed at DL6. The MTD was determined to be DL5, and was recommended for future studies. (B) Backfill cohorts at dose levels with established safety and preliminary efficacy can enhance the understanding of the effects of a dose, thereby improving the likelihood of identifying a biologically effective dose (BED) and recommending optimized doses for future study. In this example, six dose levels were tested, with dose-limiting toxicities observed at DL6. The MTD was defined as DL5, while DL4 and DL3 were selected for backfilling. Based on the totality of data generated during the study, including biomarker data showing biological activity, DL2 and DL3 were selected for further evaluation given preliminary efficacy and safety signals. (C) Randomized expansion cohorts across multiple dose levels previously evaluated in the dose-escalation phase allows for a more robust assessment of efficacy and safety, thereby helping to identify BED ranges. In this example, the MTD was defined as DL5, while DL2, DL3, and DL4 were selected as dose levels of interest for expansion starting with DL2, which was the minimal reproducibly active dose. Based on the totality of data from dose escalation based on the totality of data, including biomarker data, DL2 and DL3 were ultimately identified as doses for further exploration. (D) The Bayesian Optimal Interval (BOIN) design is one example of a model-assisted dose-finding design that can incorporate innovative design elements such as backfill. BOIN allows for the treatment of more than 6 patients at a dose level and allows the return to previous dose levels based on pre-determined cutoffs and stopping rules. (Created with BioRender. AACR Science [2025]. https://BioRender.com/upstlrt.)

Utility of Biomarkers for Dosing in Early Phase Clinical Trials of Investigational Agents

While the establishment of MTD remains important to determine the safe upper limit of dosing, it has also become essential that modern trials establish the biologically effective dose (BED) range of a drug. Identifying BED ranges, including potential doses lower than MTD, will provide opportunities to optimize the dose and schedule of investigational agents. The usage of biomarkers, with their capacity to identify early signs of potential biological activity, safety, and even efficacy, is one clear way to establish the BED range of drug products and overall gain more knowledge during early-stage trials. Numerous types of biomarkers are used in oncology clinical trials, with examples and definitions (16,17) provided in Table 1. Key biomarkers for consideration in oncology drug development have traditionally included pharmacodynamic/pharmacokinetic, surrogate endpoint, safety, and predictive biomarkers. From a regulatory standpoint, biomarkers can be divided into integral, or required for the trial design, integrated, or associated with the scientific question tested by the trial but not necessary for its overall success, and exploratory biomarkers, encompassing all other biomarkers used to generate novel hypotheses (18). Such biomarkers are typically studied in a range of tissues collected at baseline and on treatment, including samples such as paired tumor biopsies and longitudinally evaluated non-tumor samples such as blood. Given that any biopsy inherently includes some degree of risk, regardless of tissue choice, the association of biomarker studies with a specific question or hypothesis and a predefined analysis plan is pivotal to ensure ethical use of patient time, resources, and trust. A recent FDA draft guidance is in accordance with this philosophy, indicating that tissue biopsies required for enrollment in a clinical trial should be carefully considered, and that every biopsy should be thoroughly scientifically justified (19).

Table 1. Biomarker categories for potential use in clinical trials.

Functional biomarkers with a direct role in dose-related decision-making at early stages highlighted in green.

Biomarker Category	Subtype	Purpose	Example
Functional	Susceptibility/risk	Indicates predisposition to the development of a disease or condition.	BRCA1/2 mutation to identify persons at elevated risk of developing breast cancer
	Diagnostic	Used to detect/confirm presence of a disease or a subtype of a disease	Gene expression testing to determine cell-type of origin in B-cell lymphoma
	Monitoring	Assess disease status or effect of an intervention/exposure over time.	Molecular residual disease testing for recurrence of chronic myeloid leukemia
	Prognostic	Establish the likelihood of a clinical event such as recurrence or progression.	Gleason score to assess the likelihood of cancer progression in prostate cancer
	Predictive	Identify patients who are more or less likely to experience an effect from a treatment or exposure.	BRCA1/2 mutations predicting sensitivity to PARP inhibitors
	Response	Either indicates biologic activity of a medical product or environmental agent without necessarily drawing conclusions about efficacy/outcome (pharmacodynamic) or is used as a substitute for patient experiences including survival (surrogate endpoint).	Phosphorylation of proteins downstream of target (pharmacodynamic) Overall response rate to treatment (surrogate endpoint)
	Safety	A biomarker capable of indicating likelihood, presence, or degree of toxicity around exposure to a potentially toxic treatment or environment.	Neutrophil count for patients on cytotoxic chemotherapy
Regulatory	Integral	A biomarker fundamental to the design of a trial that might define eligibility, stratification, study endpoints, or other characteristics.	BRCA1/2 mutations for inclusion in PARP inhibitor trials
	Integrated	A biomarker with pre-planned collection and analysis protocols that aims to help test a specific hypothesis, but isn’t required for the trial to proceed.	PIK3CA mutation detected by NGS as an indicator of response to anti-PI3K and ERα inhibition in breast cancer
	Exploratory	All other non-integral or integrated biomarkers - frequently those which are analyzed retrospectively, not collected for all patients, and with unclear relationship to other variables of interest.	ctDNA testing to identify resistance mutations

In addition to identifying the BED range, the advantages of incorporating biomarkers in early-stage trials also include earlier assay validation, hypothesis-testing, and identification of effective/ineffective treatments, thereby enabling accelerated evaluation of a drug’s potential for long-term success. Early incorporation may be particularly relevant when the drug at hand does not yet have a well validated mechanism of function or biomarker. Such a co-development approach can keep costs lower than if developed later on, solidify pathways involved in activity, and provide additional data that may contribute to future decision-making. The Pharmacological Audit Trail (PhAT) is one such way biomarkers can be leveraged to facilitate decision-making throughout development. The PhAT lays a roadmap for the usage of biomarkers in the context of drug development, connecting key questions asked at different development stages to various go/no-go decisions (20,21). Adhering to procedures such as the PhAT that serially interrogate a drug’s biologic activity will be especially helpful in ensuring that the totality of potential data is being collected and considered, which can enable informed dosing decision-making and beyond.

Circulating tumor DNA (ctDNA) is an example biomarker with multiple applications in drug development. Already in regular clinical use as a predictive biomarker to enroll patients into molecularly targeted trials, other potential uses include as a pharmacodynamic and surrogate endpoint biomarker to aid in dosing selection (22). Retrospective analyses have shown that changes in ctDNA concentration in blood over the course of treatment correlate with radiographic response which together with other biomarkers, clinical efficacy data, and safety data, can enable the determination of biologically active dosages (23–25). Several studies have also reported early evidence of a correlation between molecular responses, observed with ctDNA studies, to response to investigational agents and patient benefit, including progression-free survival and overall survival (26–29). Additionally, ctDNA benefits from its ease and safety of collection, making it a desirable candidate for biomarker development.

While the incorporation of biomarkers into dosing and other drug-development decisions is key, challenges exist. For example, there may be no available analytically validated and clinically qualified biomarker assays relevant to the drug in question, and the development of a new biomarker is costly and complex. ctDNA assays have been a clear example of these difficulties, as the variable levels of ctDNA shed between cancer types and the variable performance between the sometimes-bespoke myriad of high-sensitivity ctDNA assays in different clinical settings reflect these challenges (30–33). Timing of an assay is also a critical consideration, illustrated by the growing push to build consensus regarding the timing and frequency of testing required to optimize the detection of changes in ctDNA while on treatment (34). While the benefits of molecularly characterizing tumor in real time through a non-invasive ctDNA assay procedure is clear, the analyses of ctDNA samples have mainly been limited to retrospective studies, leaving its usage as a real-time biomarker to assess the effects of drug on tumor subject to ongoing and future prospective investigations before regular regulatory usage. Further work regarding analytical/clinical validity and utility should continue to be performed even as ctDNA is incorporated into the totality of evidence necessary for making decisions during development such as dosing. This is supported by a recent FDA draft guidance indicating confidence in ctDNA as a potential biomarker, but confirming that increased assay standardization and prospective validation studies remain essential for the promise of ctDNA as a biomarker to be fully realized (35).

Apart from ctDNA, there are other emerging biomarker technologies analyzing platelets, peripheral blood mononuclear cells, circulating tumor cells, and tumor response, among others, which each have potential to become important datapoints for consideration in dosing decision-making. In addition to assay considerations, another important step in biomarker development for usage in dosing optimization is understanding the relationship between biomarker response to drug and clinical activity or safety. Given that many of these biomarkers can be measured serially, modeling and simulation, including the use of semi-mechanistic longitudinal models, can be leveraged to understand and explore various dosage regimens. For example, investigating longer dosage intervals or step-up and step-down dosages in silico prior to conducting a large clinical study could maximize the chances of selecting a more optimized dosage. Hence, further attention to the understanding and development of key biomarkers should play an influential role in dose selection.

Utilizing Controlled Backfill Cohorts and Expansion Cohorts to Gain Greater Understanding of Dose-Exposure Relationships for Safety and Activity

In addition to the classic goals of dose-escalation and dose ranging, many early-phase trials now address unique research questions based on a development program of specific preclinical, clinical and other emerging data including biomarker results, and seek to derive answers as early as possible. The implementation of novel trial designs, such as backfill cohorts, can help inform and accelerate this process during dose escalation and even contribute to future dosing decisions. Backfilling incorporates small patient cohorts at different dose levels of interest once the respective dose/schedule is cleared during dose escalation. These backfill cohorts can be initiated in parallel to dose escalation, providing increased clinical data for safety, tolerability, pharmacokinetics, pharmacodynamics, and antitumor activity, while adhering to ethical and scientific considerations (14,36). Backfill cohorts can even be a tool for more complex biomarker data collection including, when appropriate, paired tumor biopsies or other additional studies that evaluate target and pathway modulation, investigate mechanisms of action, and/or determine the biologically effective dose range for a specific drug. Necessary cohort sizes will vary on a case-by-case basis and depend on the specific drug development questions being asked. This may include, among other parameters, the number of dosages being backfilled to, the amount of variability in the patient population, the degree of drug activity observed, and any relevant biomarker specific subgroups being employed. The clinical data collected from backfill patients can thereby add to the totality of data beyond the usual ~3 patients treated at each dose level in the primary dose escalation cohorts typical of many traditional trial designs.

At the end of dose escalation, one way to compare the merit of certain doses is to conduct dose comparison expansion cohorts to study similar patient populations across multiple potentially optimized dosages. Such direct comparison of safety, tolerability, pharmacokinetics, pharmacodynamics, and antitumor activity between the different dosages is critical, and can inform dose selections for more in-depth and resource-intensive studies. Pharmacokinetic and pharmacodynamic modeling and/or backfill cohorts can increase the patient experience at each dose level to aid in the selection of dosages for such studies. One key consideration when selecting dose levels to compare is the degree of direct overlap in pharmacokinetic drug exposures, particularly when exploring different dosing schedules. While the decision to randomize either backfill or expansion cohorts may be considered on a case-by-case basis, in general, randomization is recommended to ensure that the results of such dose comparisons are due to the specific doses and schedules being compared, as opposed to differences in patient characteristics (36). Rich data can be both considered and collected from these cohort studies, including potentially patient reported outcomes (PROs) in addition to more traditional data. While it may be challenging to determine which framework to use, how frequently to collect data, or even what data to collect, any relevant information collected can contribute to the totality of data necessary to select an optimized dose.

Evaluating and Modeling All Early Data to Select Doses for Further Evaluation

Typically, early clinical studies in patients with oncologic diseases are conducted over a range of doses with small numbers of patients in each dosage cohort. However, it is important to note that these studies are multi-dimensional and provide data well beyond the primary or secondary endpoints, especially when designed with features such as biomarker collection and backfill cohorts in mind. These early clinical studies, when appropriately designed, can provide useful data over a range of dosages. This may include differences in pharmacokinetic (PK) parameters such as drug exposure over time, among other metrics, as well as modulation of depth and duration of pharmacodynamic (PD) biomarkers including antitumor activity measures such as response rates and safety measures such as adverse events, all in a longitudinal fashion. The totality of these data can be analyzed in a systematic fashion to inform subsequent dosages and studies instead of simply relying on MTD. The use of integrated clinical pharmacology analysis from early clinical and nonclinical studies is not uncommon in oncology (37). This is despite frequently limited dose-ranging data, low confidence with exploratory PD endpoints, and a degree of uncertainty around traditional empiric exposure-response analyses. For example, nonclinical data such as the half maximal inhibitory concentration of kinases of interest, as well as PK data such as maximal plasma concentration, area under the plasma-concentration-time curve, and target inhibition PD data from early clinical studies have been used to support the selection of effective doses for late-stage clinical studies, such as for the second generation covalent irreversible BTK inhibitor zanubrutinib (38,39). Alternatively, longitudinal semi-mechanistic models were used to characterize the dynamic changes in biological activity over time relative to drug exposure for asciminib, supporting efforts to optimize its dosing in specific subpopulations (40,41). While such modeling, especially with respect to time, is critical for understanding a particular product’s risk/benefit profile, the simultaneous consideration of a myriad of factors and models, including various safety data and efficacy data from clinical and nonclinical sources, is necessary for the identification of optimized dosages.

In most analyses, efficacy and safety endpoints are analyzed separately and qualitatively compared to support dosage decision-making at early stages. Given the need for quantitative, integrative, and transparent approaches for dosage selection, the concept of clinical utility index (CUI) was introduced as an alternative endpoint to more qualitative dosage selection techniques (42,43). In practice, a CUI sums metrics created for each chosen clinically meaningful endpoint composed of weighted importance and scored exposure-response relationships to generate a single score for each dose, with the highest summed utility score providing the best overall benefit/risk profile (Figure 2). While the CUI approach is not currently commonplace in oncologic drug development, there have been several cases of its application. For example, the usage of CUIs was essential for selecting dosages for further study during development of the investigational TIGIT blocking monoclonal antibody SEA-TGT, especially given previous clinical experience that drugs in its class are not expected to reach MTD or display clear safety and efficacy signals during dose-escalation trials (44,45). In this instance, various PK and PD endpoints measured during the SEA-TGT dose escalation trial were selected a priori for inclusion in the CUI based on nonclinical or literature derived data. The integration of this data, with PD endpoints based on previously validated mechanisms of action and PK endpoints derived from a semi-mechanistic translated PKPD model, allowed for the determination of CUI at each dosage and therefore dosages for further exploration.

Figure 2. Clinical Utility Indices (CUIs) are a transparent, fit-for-purpose aid for the selection of dosages for additional exploration in subsequent trials.

To develop a CUI, broad stakeholders should be convened including various medical professionals, drug developers, and patients. These stakeholders will determine important endpoints for consideration, the relative importance of these endpoints through weighting, and optimal cutoff points to condense the measurements of endpoints into standardized utility scores. There is not a limit to the number of endpoints that may be simultaneously included in a CUI, but too many can create an unwieldy metric and overlap between endpoints should be avoided. Frequently, at least one efficacy and one safety endpoint are selected for inclusion within the context of drug development. Example endpoints might include rates of very good pathologic response and rates of grade 3+ neutropenia at a given dosage. Upon selection, stakeholders use their expert opinion to decide the relative importance or weight of each endpoint, with scores totaling to 1.0 across all endpoints. Finally, cutoffs and utility scores for each scoring level are determined, with each level representing a clinically meaningful stratification and the scoring representing its relative benefit to patients with a minimum score being 0 and a maximum score being 1.0. After stakeholders have determined these underpinning characteristics, the CUI is applied to data from preceding clinical trials or nonclinical studies to determine the dosage with the highest overall score. Each endpoint’s weight is multiplied by the utility score achieved for a specific dosage and added together across that dosage, with dosages with higher potential benefit/risk ratio receiving higher CUIs. Sensitivity testing is frequently conducted, changing out certain endpoints, weights, and cutoffs, to ensure that the choices made by the stakeholders which developed the CUI did not greatly impact the outcome and to provide knowledge about what the largest driving forces behind the CUI were. (Created with BioRender. AACR Science [2025]. https://biorender.com/fq801p5.)

More conventionally, CUIs combine clinical safety and efficacy endpoints in a single index to determine potentially recommended dosages instead of independently evaluating safety and efficacy separately. This paradigm was employed to identify dosages to be further explored for venetoclax in combination with bortezomib and dexamethasone in the treatment of pre-treated patients with multiple myeloma (46). Using a post-hoc selected efficacy endpoint of binary “yes/no” very good partial response or better and a binary safety endpoint of “yes/no” grade 3+ neutropenia in a 2×2 table, the CUI was able to recommend dosages balancing the two at varying weights. By conducting sensitivity analysis and exercising expert opinion, a specific dosage was settled on for subsequent clinical development. In addition to the datatypes described above, CUI have also been shown to incorporate quantitative PK and PD time-course data using techniques such as semi-mechanistic or QSP models to model a continuous response and examine the “what-if” situations where dosages are modified for exploration in future clinical studies (47). The combination approach, using efficacy or safety binary endpoints data alongside continuous PK response data, is not trivial and further work is needed to interpret the revised scale alongside research to further develop the utility of such a metric.

Approaches, such as CUI, that integrate many endpoints to obtain a single metric for the identification of dosage(s) to explore in future studies are attractive, but limitations exist. CUIs need to be fit-for-purpose and require consensus among the stakeholder team at every step of their development. Endpoints for potential inclusion in CUI can be drug specific, class specific, and/or even patient population specific and as such, considerable time may be required to gain alignment among stakeholders. A priori selection of these parameters can limit bias but requires knowledge of the drug being developed, which may not be available. Post-hoc selection could be logistically appealing to maintain flexibility as new information from clinical studies or from competitors becomes available, but the biases introduced ought to be accounted for. Additionally, the final metric is center/site specific due to the somewhat subjective nature of endpoint definition/construction and weighting decisions. In practice, different clinicians could potentially suggest different summary indices and/or weighting, resulting in an uninterpretable scale that can lead to the selection of different doses. Furthermore, previous studies rely on certain assumptions such as the clinicians ‘weighting is close to the true weighting.’ As such, the robustness of the CUI and the veracity of any underlying assumptions need to be re-evaluated in the context of every study.

A benefit of the discussions required to appropriately perform this approach includes the opportunity to engage with patient advocates to ensure that the concerns of all stakeholders are taken in account. In addition, quantitative modeling and simulation, longitudinal data, and PROs may be incorporated, given that the acuteness and duration of safety events may be an influential factor from a patient’s perspective but are frequently not integrated in drug development. There is much debate around how to collect PROs and what to collect, but most well-intentioned and researched approaches will provide value. Regardless of variables included, approaches with the ability to integrate all available data can lead to the rapid selection of more optimized dosages which ideally provide a greater patient benefit/risk ratio than simpler approaches that take less data into account. Such benefits however need to be quantitatively addressed at the design stage via simulation studies.

Discussion

The FDA–AACR workshop on Optimizing Dosages for Oncology Drug Products, which gave rise to this second manuscript in a set of three concerning the future of dosage optimization in oncology drug development, examined in greater depth the topics presented herein (6). Dose optimization is increasingly being taken into consideration in the planning of trials and in regulatory decisions. The approaches and principles outlined here are applicable across drug development in all geographic regions, as their implementation is likely to increase the odds of success in development and improve therapeutic outcomes for patients with cancer. In an era where oncology clinical trials are being asked to do more at earlier stages, it is imperative to collect and leverage the totality of both nonclinical and clinical data, comprising safety, tolerability, pharmacokinetics, pharmacodynamics, and efficacy to answer questions about the drug being developed in a manner consistent with FDA dose optimization initiatives such as Project Optimus (48). This flagship program seeks to facilitate the expansion and usage of dosing approaches in oncology that maximize safety in addition to efficacy by promoting tools, communicating expectations, and spurring collaboration between key parties. Project Optimus and other programs, such as the dosing-centric initiatives of the FDA Office of Clinical Pharmacology, are necessary to realign drug development with patient and physician attitudes. Now, both patients with cancer and medical oncologists see benefits to lower dosages, with 53% of patients and 85% of oncologists acknowledging that a higher dose is not always better and over 90% of both populations willing to discuss flexible dosing options to optimize quality of life (49, 50). Through increased focus on dosing optimization efforts, all parties invested in the drug development process can meet this moment and provide better care to patients.

Approaches that can synthesize many varied datatypes to aid in the identification of doses for further evaluation in a transparent and quantifiable way will be required to serve these patient-centered goals. Subjectivity is likely to be inherent to such approaches, which rely on decisions regarding variable inclusion and discretionary cutoff points, and can be a great advantage by allowing stakeholders to tailor the method for a specific trial, disease, or patient group. In the future, the investigation of strategies to incorporate patient reported outcomes and longitudinal data into CUIs and other decision-making and data-synthesizing approaches will be essential to ensure the long-term health of patients on novel oncology treatments. Modern dose finding trial designs will also need to incorporate standard and emerging biomarkers, such as ctDNA, in larger cohorts of patients through the use of backfill cohorts during dose escalation and randomized dose comparison cohort expansions to maximize the amount of data gathered from trials and to facilitate the usage of all possible relevant data in dosage decisions. Given the complexities and key differences between different classes of investigational oncology agents, designing each trial on a case-by-case basis and in close consultation with the FDA is best practice. Leveraging all available data through modern techniques is essential for selecting potentially optimized dosages for further exploration and will enable success in subsequent trials.

Abbreviations

BED

Biologically Effective Dose

BOIN

Bayesian Optimal Interval

CUI

Clinical Utility Index

PRO

Patient Reported Outcome

PD

Pharmacodynamic

PK

Pharmacodynamic

MTD

Maximum Tolerated Dose