Randomised Phase 1 clinical trials in oncology

Abstract

The aims of Phase 1 trials in oncology have broadened considerably from simply demonstrating that the agent/regimen of interest is well tolerated in a relatively heterogeneous patient population to addressing multiple objectives under the heading of early-phase trials and, if possible, obtaining reliable evidence regarding clinical activity to lead to drug approvals via the Accelerated Approval approach or Breakthrough Therapy designation in cases where the tumours are rare, prognosis is poor or where there might be an unmet therapeutic need. Constructing a Phase 1 design that can address multiple objectives within the context of a single trial is not simple. Randomisation can play an important role, but carrying out such randomisation according to the principles of equipoise is a significant challenge in the Phase 1 setting. If the emerging data are not sufficient to definitively address the aims early on, then a proper design can reduce biases, enhance interpretability, and maximise information so that the Phase 1 data can be more compelling. This article outlines objectives and design considerations that need to be adhered to in order to respect ethical and scientific principles required for research in human subjects in early phase clinical trials.

Background

The specific aims and overall purpose of Phase 1 clinical trials have undergone substantial changes over the past several years. Until 2013, Phase 1 trials in oncology had a single, simple objective—to show that the investigational agent or regimen was adequately well tolerated in a relatively heterogeneous patient population, across a broad range of solid tumours.¹ The working assumption was that, overall, patients would have a similar tolerance or reaction to the drug or regimen regardless of the tumour origin, disease stage, or prior lines of therapies allowed within protocol eligibility requirements. The data from a small Phase 1 trial of 20–30 patients would be used to provide a single recommended Phase 2 dose (RP2D) and address whether the drug or regimen is safe and which of the available doses should be taken forward for further testing. Phase 2 trials would then be carried out in a more homogeneous population, with the aim of showing clinical activity at specific disease sites. However, the past several years have seen these straightforward objectives evolve into a rather more ambitious set of objectives: to identify the most effective dose, the right treatment schedule and the right patient population, in terms of the likelihood of anti-tumour activity. These days, early phase trials also aim to screen a potentially active agent from a number of potentially less active agents and to halt studies on drugs that, on the basis of the information observed so far, are unlikely to be successful in terms of anti-tumour activity (futility),² or investigate new combinations of drugs with a history of use either individually or in other combinations.³

Thus, the distinction between the different phases of clinical trials has become less clear, as have the criteria for determining whether or not a trial is successful.⁴ A trial with multiple hypotheses and objectives can be successful in addressing one objective but not necessarily all.² In addition, setting out to address numerous issues that fall under the remit of early phase trials, Phase 1 trials now also aim to obtain reliable evidence regarding clinical activity to lead to drug approval via the Accelerated Approval approach or Breakthrough Therapy designation in cases where the tumours are rare, prognosis is poor and an unmet need exists. Early Phase 1 trials have therefore increased in scope, size and access, and now enrol a large number of patients and aim to address the aforementioned objectives in a single protocol.^5,6 The prevailing reason for doing so is that a single trial that simultaneously addresses multiple objectives can reduce cost and duration and thus save resources, which can then be reallocated to screening and developing other novel treatments in the pipeline. Patient advocates also favour large early phase trials that provide access to potentially successful novel drugs or therapies, such as targeted or immune-based therapies, to patients with advanced disease without the risk of being randomised to receive standard of care treatment in Phase 3 trials.⁷ However, although these trials aim to address multiple objectives within the cost and infrastructure of a single trial as well as to reduce the duration of drug development, it is essential that they maintain rigour and interpretability.

This article articulates the role or randomisation in early phase trials where the goal is to go beyond the simple comparison of different drugs. We provide examples of relevant, timely clinical questions where randomisation can help avoid potential biases in data collection and their interpretation. These designs may have the potential to lead to definitive larger phase trials or trials that can lead to drug approvals.

Role of randomisation in early phase trials

Can Phase 1 data help to address the more ambitious set of aims and, if so, how can this be achieved? If a trial is meant for discovery and hypotheses generating or to accumulate evidence in order to inform the design of future trials, then the Phase 1 scope will be more focused. However, if the data emerging from the study are not sufficient to address a more extensive set of objectives, then an effective design will be needed to reduce biases, enhance interpretability, and maximise information to provide more compelling data. The objectives and design considerations that need to be addressed are outlined in Table 1.

Table 1.

Objectives of early phase trials and three potential designs.

Design 1	Design 2	Design 3
Simple dose escalation	Dose escalation with DEC (dose expansion cohorts)	Dose escalation with DEC and randomisation
Objectives
1. Identify the MTD and establish the safety profile in a heterogenous patient population (safety only)	1. Identify the MTD in a heterogenous patient population (dose escalation; safety) 2. Identify whether the drug shows promising efficacy and in which disease groups (DEC; efficacy) 3. Identify the appropriate patient population for drug development (DEC and dose escalation)	1. Identify the MTD in a heterogenous patient population (dose escalation) 2. Identify whether the drug shows promising efficacy and in which disease groups (DEC) 3. Assess whether the drug works uniformly or whether there are differences in response within subgroups (disease heterogeneity and drug activity) 4. Identify patient populations, dose and treatment schedule 5. Identify which drugs need to be eliminated early because they are ineffective and which drugs to take forward because of promising activity (DEC and dose escalation)
Protocol development
Patient population/schedule of treatment administration
Heterogeneous population with single schedule	Heterogeneous population with single schedule (dose escalation) followed by narrower eligibility criteria (DEC) and possibly multiple schedules	Heterogeneous population with single schedule (dose escalation) followed by narrower eligibility criteria (DEC) and multiple treatment schedules, or subgroups

These examples are meant for illustrative purposes only and in practice can overlap and not be distinct.

MTD maximum tolerated dose, DEC dose expansion cohort.

Current protocols are adaptive and may get amended⁸ in order to adjust the eligibility criteria, patient population, dose levels or treatment schedule under study as investigators learn from the data. Amendments can change the eligibility criteria to focus on a specific sub-population or add/drop cohorts to study new populations as emerging evidence points to potential clinical activity.⁷ However, those protocols need to adhere to ethical, safety and scientific criteria, and their design must take into account both patient safety and the trial’s objectives. An important distinction compared to previous Phase 1 trials is the attempt by current trials to eliminate ineffective drugs, in various disease indications, in an early-phase, single protocol trial while simultaneously aiming to address whether heterogeneity exists between diseases (primarily in terms of efficacy). In the absence of any prior clinical data, or with preclinical data that might not translate well to humans, it is also important to enrol the minimum number of patients in order to answer the questions as quickly as possible by designs that are efficient regardless of the level of activity. While we aim to address multiple objectives with the infrastructure of a single trial and reduce the duration of drug development, which may be too ambitious a goal, it is essential that we maintain rigor, efficiency and interpretability. Randomisation in early phase trials is a sensible approach since it can achieve the goal of obtaining information at more than one dose level, treatment schedule or drug combination while adhering to ethical and safety criteria. A number of early phase trials that involve randomisation currently exist (e.g. NCT03939897, NCT04045613, NCT03819465, NCT03837509 and NCT03989414).

Which questions can be addressed by the trial?

The premise of innovative, adaptive designs is that multiple objectives can be achieved with a smaller sample size as a result of data borrowing or information sharing. For example, Phase 1 model-based dose-escalation designs enable sharing information on dose levels, treatment schedules, patient groups (e.g. paediatrics and adults). The use of a model to guide the dose-escalation allows for data sharing between groups, dose levels or cohorts, thus resulting in efficiencies in terms of smaller sample sizes and high accuracy in finding the right dose.⁹ These designs have been shown to be safe, ethical and scientifically valid (in addressing the objectives); while they minimise sample size and trial duration, and patients do not receive ineffective treatment.^10–13

In any scientific investigation, the number and complexity of the issues that can be addressed depend heavily on the available resources, and might need to be restricted, for example, if the number of patients being enroled is limited. However, if the design is adaptive, as information evolves and it could be decided that the study of a particular subgroup, or treatment schedule, that perhaps shows more promise than others, warrants a greater allocation of resources than would occur via simple randomisation.¹⁴ In this case, flexible adaptive randomisation schemes,¹⁵ such as unbalanced randomised blocks and bandit-type algorithms,¹⁶ can play an important role¹⁷ in terms of allocating patients to dose levels or treatment arms in a structured way. The use of blocks allows better control of the influence of several factors on the treatment effect. In general, designs with a maximum sample size (upper bound) but no fixed sample size, as is the case in sequential designs that allow early stopping rules for lack of activity or safety concerns, are preferred in early phase trials.^12,13

Incorporating a dose-expansion cohort

In the more classical Phase 1 designs, randomisation plays no real part in the dose-escalation phase, especially in the absence of any dose-expansion cohort (Fig. 1a). In this situation, allocation to a particular dose level (which occurs sequentially, depending on the outcome of the previously treated cohort) is primarily guided by the simultaneous effort to either move in an upward direction, away from levels offering little chance of therapeutic benefit or to move in a downward direction, away from levels appearing to be strongly associated with the occurrence of serious adverse events. Different Phase 1 designs might address this issue in slightly different ways, but the common goal has always been to identify a single recommended RP2D that makes the best compromise between the two goals. Deciding to incorporate a dose-expansion cohort, which considers efficacy alongside toxicity, represents a fundamental change.² There is no obvious reason to limit investigation to the recommendation of a single, best candidate dose. Instead, it should be recognised that there might be two or three doses, following the escalation phase, that could be considered in the dose-expansion cohort. Indeed, not only is it possible to consider more than a single dose as a candidate for the RP2D, but the ethical restrictions that tended to rule out or reduce the role of randomisation in early phase chemotherapy trials have now been largely lifted.⁴ This is because there is no compelling reason to favour one candidate dose over another—one might be associated with a lower risk of inducing an adverse event but might also be less effective. In oncology, in the absence of clinical activity, we cannot plan to enrol many patients upfront just for the purpose of collecting safety data. It would be advantageous to learn much more about the relative benefits of such competing doses, and the most suitable tool for this is that of randomisation.

Fig. 1. Dose escalation (carried out in cohorts of 3 patients) followed sequentially by dose expansion after the maximum-tolerated dose (MTD) has been determined.

Dose expansion randomises subjects equally to two dose levels in molecular- or disease-specific patient populations (denoted by B: breast, P: prostate and C: colon cancer). a shows the dose escalation. b shows the dose expansion.

Optimising the effort to identify the best dose

Sample sizes have traditionally been too small to obtain any information other than an approximate estimate of the maximum tolerated dose (MTD). However, larger sample sizes in early-phase trials with dose-expansion cohorts now enable us to consider the accuracy, precision and variability of the treatment effect estimate. Typically, dose expansions have been carried out after the MTD has been established, with the goal of examining efficacy in various disease or molecularly defined populations treated with a fixed dose. Thus, the dose-escalation phase is carried out first (MTD determination; Fig. 1a), followed by a dose-expansion phase, which is randomised to two dose levels (Fig. 1b). More recent protocols, however, allow for dose expansions concurrently with the dose-exploration phase (Fig. 2) by expanding lower dose levels to a higher number of patients, sometimes as many as 20–40,¹⁸ with the aim of collecting data and comparing various dose levels or treatment schedules in terms of both safety and activity. The assumption in certain clinical settings, such as immunotherapy, is that the dose toxicity curve is expected to be rather flat or to reach a plateau in terms of the therapeutic index. Thus, various dose levels can be comparable in terms of safety, but the question is whether the dose levels are equally active or comparable in terms of efficacy.

Fig. 2. Dose escalation and dose expansion carried out concurrently prior to determination of the maximum-tolerated dose (MTD).

Dose expansion randomises equally to three dose levels (levels 1, 2 and 3) in molecular- or disease-specific patient populations denoted by B: breast, P: prostate and C: colon cancer. Lighter-coloured circles indicate backfill patients whom are treated separately from the dose-escalation patients at levels 1, 2 or 3. a shows the dose escalation; b shows the dose expansion.

The estimated MTD has a probability of not much greater than around 50% of being the true MTD.^9,19 The remaining approximate 50% probability mass will mostly be distributed between the two adjacent dose levels either side of the best estimate of the MTD. This means that after treating the 20–40 patients typically included in dose-escalation trials, the level we calculate as most likely to be the MTD has a lower or equal percentage chance of being the true MTD than the two adjacent levels combined.⁹ Given this, the argument that further experimentation should be carried out using a single dose—the estimated MTD — is not a compelling one. Hence, there is a need to learn more about the adjacent levels.²⁰

Obtaining information on the probability of experiencing a dose-limiting toxicity (DLT) at these adjacent levels requires experimenting there which will be most effective when carried out using randomisation.^12,13,20

In rare patient populations, a long-term drug development process that studies the drug in distinct phases of three clinical trials including a comparative follow-up study might not be feasible.²¹ Thus, results from early phase trials, in terms of dose, treatment schedule and patient population, will inform registrational studies.²² Although the sequential entry of patients eliminates some biases, randomisation will enhance our chances of avoiding imbalances in patients from specific categories being treated at particular doses or schedules. In small sized studies, however, the advantages of such randomisation compared with sequential enrolment would be less clear, in particular when the patient population remains heterogeneous, with respect to prior treatment, comorbidities or advanced disease. We need, for example, to reduce the biases that can occur due to not accounting for the status of advanced disease patients, heavily pre-treated patients, or patients with different comorbidities even within the allowed eligibility requirements.²³ Unlike preclinical studies, in which allocation to dose levels is straightforward, randomising patients to dose levels must be achieved within the ethical, scientific and safety restrictions on trials in human subjects, and needs to be accomplished while eliminating the chance of patients receiving an ineffective or unsafe dose, answering the question(s) using the minimum number of patients, and being able to stop a futile drug early or expand a safe/active drug to more patients if feasible.

Controllable and uncontrollable factors

The goal of randomisation is to reduce biases due to known but uncontrollable factors such as prior treatment, age, gender and/or comorbidities. Known and controllable factors such as treatment dose and schedule allow the techniques of randomisation. Randomisation to different doses, or schedules can provide more complete Phase 1 data. Assessment of safety and efficacy across multiple dose levels can be more efficiently investigated by incorporating randomisation. Isolating particular effects, as well as interactions between treatment and disease or biomarkers, is complex. Each case is different, and a few examples help to illustrate the role of randomisation in this effort.

Groups defined by dose schedule, dose levels and genetic markers

In this example, the factors that can be controlled are the dose and schedule, which we will refer to as treatment groups (groups/cohorts in Phase 1 are also constructed because of disease and mutation, which we cannot randomise). For example, assume four treatment groups comprising two different dose levels and two different schedules. Within a specific disease—for example, lung cancer—suppose we have 48 patients with a block size of 12 every block of 12 patients is equally distributed among the four treatment groups: three patients receive dose 1/schedule1 (group A), three receive dose 1/schedule 2 (group B), three receive dose 2 schedule 1 (group C) and three receive dose 2/schedule 2 (group D). This process is then repeated for four other diseases which, viewed as study variables, are fixed and thus not open to randomisation. For example, if we have breast, prostate, colon, melanoma and uterine cancer, such that we have five diseases with 48 patients each, the total sample size will be 240.

Genetic tests can be carried out at baseline, but the prevalence of mutations is mostly unknown and hence a study ensuring that a given percentage of mutated patients is enroled in these early phase trials is not always feasible. Patients with mutations may have a different drug tolerance distribution to those without mutations. Making use of this information, can help us obtain more accurate dosing for the respective groups. Factors that can be controlled in this randomisation scheme are known prognostic factors for each disease, such as age, or prior lines of therapy. The dose level to which patients are allocated is balanced with the idea of obtaining more precise dose estimates across the several groupings (see Supplementary Information, example 1).

Bridging between completed and current studies

So far we have discussed the use of randomisation in large early phase trials with dose expansions. Large, early phase dose-expansion trials are distinct from small early phase dose-confirmation trials. Dose-confirmation trials have prior data that support an existing MTD—for example, a MTD from another population or similar formulation of the drug—and investigate adjacent doses, or groups of doses, using toxicity and, possibly, efficacy to make a recommendation for the population under study.

Paediatric trials following a completed adult trial

Paediatric dose-confirmation studies often follow a completed, or nearly completed, trial in a different population (often adult) that examined a wider dose range; the paediatric studies investigate a reduced dose range, perhaps just one or two dose levels, no more than three (Table 2; Supplementary Information, example 2). In this case, dose-expansion comes after dose-escalation and might include a single dose. If the dose is fixed and controlled for in the expansion study, randomisation in the expansion is carried out to determine only the schedule. If the dose is not fixed, but a narrow range—for example, 1–2 levels—needs to be explored just for confirmation of the adult MTD in the paediatric population, then statistical considerations described previously in bridging studies can be applied, and no further methodological considerations are required.^24,25

Table 2.

Examples of the use of randomisation in early-phase dose-confirmation trials.

Clinical Setting	Prior/available data	Assumptions
Paediatrics	Extrapolating from completed adult studies	Drug profile in adults has been established and it is safe. Paediatric patients are expected to tolerate the treatment as well as or better than the adult patients. This has been true with most cytotoxic therapies, whether this is true for novel classes is uncertain.
Paediatrics	Extrapolating from ongoing adult studies	Drug’s safety profile in paediatrics and adults is unknown. Paediatric patients are expected to tolerate the treatment as well as or better than the adult patients and this will be tested.
Combination Regimens	Experimenting at the MTD of each drug individually and 1–2 levels below	Combination of the MTDs of each single agent is expected to be tolerable and this is to be confirmed.
Emerging evidence	Obtaining reliable control estimates on safety parameters	Historical data and data from ongoing trial are anticipated to be in agreement and this will be monitored.

MTD maximum tolerated dose.

Combination regimens

Typically, trials of combination regimens require a small number of patients per dose-level in order to obtain adequate pharmacokinetic/pharmacodynamic (PK/PD) or efficacy data in addition to safety data and other exploratory correlative endpoints (e.g. NCT04148937, NCT04188548). If there are one or more potential dose levels for the MTD, then randomisation over these levels (Supplementary Information, example 3) with some protection—rules to avoid overdosing or underdosing—will allow a more informed decision regarding the combination regimen.²⁶ Table 2 provides a few examples of such trials.

Assessing backfill doses

Backfill is used in current Phase 1 protocols mainly to explore dose levels below the MTD, and it is part of the dose-exploration phase (Fig. 2). For example, if the MTD is known to be level 3, enrolment at level 3 is expanded to more patients and possibly to levels 1–2 with the aim of comparing levels in terms of safety and efficacy or determining whether a lower level is equally efficacious but possibly better tolerated. The possibility of randomising to two levels below the MTD allows us to obtain more information at the backfill dose level(s). It should be made clear in protocols whether treating patients to more than two levels below the MTD is scientifically justified. Including randomisation during the backfill phase can be reasonable and the observed outcomes can identify the RP2D in the same context as ‘pick the winner’ or selection designs.²⁷ The data at backfill doses might suggest further expansion; if not, lower dose levels should not be expanded. Further expansion can be achieved with sequential designs that have been established in the dose-expansion setting.²⁰ However, expanding dose levels or backfilling simultaneously before finding the MTD, can be potentially unethical as patients cannot be treated in large numbers far away from the MTD with the rationale of spreading experimentation in order to obtain information.

Another example is when pharmacokinetic drug concentrations are in the range that caused efficacy in preclinical models or pharmacodynamic assays show biological activity; lower dose levels are then expanded to test efficacy in humans. There is a pharmacodynamically active dose range over which efficacy occurs and it is often difficult to know what the best dose is, or at which point additional drug is necessary even if it is not toxic. Spreading or experimenting far away from the MTD must still respect coherence and other statistical and ethical principles.¹⁰ The requirement to treat every patient in the study in accordance with principles governing research in human subjects means keeping a focus on and balance between signs of treatment-related adverse events alongside any signs of potential treatment benefit (risk/benefit). We need to remain mindful of the context, which is not that of a potentially conclusive comparative study, but rather one in which the study enables the investigators to fully exploit all available data to make an informed decision about how to proceed further.

Discussion

Although randomisation has been used extensively in dose-finding studies in other fields, such as evaluating the dose of psychiatric drugs or in studies evaluating addiction and the effect of dose on withdrawal symptoms, randomisation in Phase 1 dose-finding trials in oncology is a new concept. However, randomisation in oncology is now often used during dose expansion in Phase 1 protocols to assess different dose levels, treatment schedules or patient populations. Dose-expansion cohorts have been included in many early phase trials with the aim of reducing the uncertainty and improving the confidence/accuracy of the estimated RP2D, although they are now becoming more common for addressing multiple clinical objectives. For example, in a previously reported trial²⁸ of ten patients, after three DLTs have been observed at level 3, there is a 10% probability that dose level 2 is the correct MTD among six levels.²⁰ However, with the inclusion of 22 patients and after eight observed DLTs, the probability that the estimated MTD is the correct MTD increases (48% probability at level 2 and 27% at level 3) and decreases at other levels. We can exploit this information from the dose-escalation phase in any randomisation scheme carried to the dose-expansion phase, and optimise both data collection and data information gathered from the totality of data from early phase trials. Model-based dose-finding algorithms sequentially update and improve our best estimate of the MTD. Scientific and statistical uncertainty regarding the MTD can be our basis for allocation. It provides the justification for randomising to more than just the best current estimate of the MTD.

The Pharmacological Audit Trail has documented the shift in the way Phase 1 trials have changed in a series of biomarker-driven questions that are used to support pharmacological understanding and evidence-based decision-making in drug discovery and development in an attempt to expedite the regulatory approvals of new anticancer drugs.^29–31 Other authors have discussed that there is room for improvement in the way we identify what appears to be the most promising dose and treatment schedule in Phase 1 trials.³²

Given the rapidly changing landscape of the type of treatment being considered for drug development in oncology, the need for a newer and broader set of goals for Phase 1 studies is becoming apparent. We believe that we are in a position to address these ambitious goals, and that the need to develop new sets of methodologies for the task is not compelling. Careful adaptation of tried and tested approaches borrowed from later phase trials—randomisation, in particular—can make a very significant contribution to the drug development process.

Author contributions

A.I. contributed to the conception of the review; the collection, analysis and interpretation of the information included within; the drafting of the review; played an important role in interpreting the results; and critically revising the review for important intellectual content. Figures were conceived and developed by A.I. J.O. contributed to the conception of the review; the collection, analysis and interpretation of the information included within; played an important role in interpreting the results; and critically revising the review for important intellectual content. Both authors approved the final version of this review and agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Ethics approval and consent to participate

Not applicable.

Consent to publish

Not applicable.

Data availability

Not applicable.

Competing interests

The authors declare no competing interests.

Funding information

This research was funded in part through the NIH/NCI Cancer Center Support Grant P30 CA008748.

Supplementary information

The online version contains supplementary material available at 10.1038/s41416-021-01412-y.