An important question to ask when beginning the trial design process is ‘how does this treatment work?’ The term ‘cytotoxic’ may be used to describe chemotherapeutic agents, where tumour shrinkage or response is widely accepted as reflecting anti-cancer activity. Many new cancer therapies are, however, targeted at specific molecular pathways relevant to tumour growth, apoptosis (programmed cell death) or angiogenesis (new blood vessel formation). Such ‘targeted therapies’, including tyrosine kinase inhibitors, monoclonal antibody therapies and immunotherapeutic agents, may be ‘cytostatic’. Here, a change in tumour volume may not be the expected outcome: in such cases, tumour stabilisation or delay in tumour progression may be a more anticipated outcome.

The mechanism of action of the agent under investigation will inform many subsequent decisions, including the choice of outcome measure and whether or not the trial should be randomised.

The aim of the treatment under investigation should be considered both in the context of its mechanism of action and the specific population of patients in which the treatment is being considered.

It is important to consider the ultimate aim of treatment, which would inform the outcome measures in future phase III studies, and how this relates to shorter term aims that can be incorporated into phase II trials. For example, in a population of patients with a relatively long median progression-free survival (PFS) and overall survival (OS), the aim of a phase III trial may be to prolong further PFS and/or OS. These would, however, be unrealistic short-term outcomes for a phase II trial; tumour response, which may reflect PFS or OS, can be an appropriate shrinkage aim in a phase II trial. By contrast, where the prognosis is less good PFS may provide a realistic short-term outcome in phase II.

It is essential to consider how the longer term and shorter term aims of treatment are related, to ensure an appropriate intermediate outcome measure is chosen in phase II that provides a robust assessment of potential efficacy in subsequent phase III trials.

It is important to ascertain whether the treatment under investigation will be given as a single agent or in combination with another novel or established intervention. This distinction can inform the decision as to whether or not randomisation should be incorporated. Where an investigational agent, be it a conventional cytotoxic or a targeted agent, is used in combination with another active treatment it can be very difficult to distinguish the effect of the investigational agent from that of the standard partner therapy; this distinction can be made easier by incorporating randomisation (see Section 2.2.2 for further discussion).

Similarly, the assessment of toxicity for combination treatments should also be addressed. Where the addition of an investigational therapy is expected to increase both activity and toxicity to a potentially significant degree, dual primary endpoints may be considered to assess the ‘trade-off’ between greater activity and increased toxicity (see Section 2.1.4 for further discussion).

Biomarkers are an increasingly important part of clinical trials. They can be defined as ‘a characteristic that is objectively measured and evaluated as an indicator of normal biological processes, pathogenic processes, or pharmacologic responses to a therapeutic intervention’ (Atkinson et al. 2001).

Biomarkers may be considered in the design of phase II trials in two ways. First, a biomarker may serve as an outcome measure. The biomarker may be an intermediate (primary) endpoint in a phase II trial provided it reflects the activity of a treatment and is associated with efficacy; this may form the basis for a stop/go decision regarding a subsequent phase III trial. Decisions regarding the use of biomarkers as primary outcome measures will feed into the decision regarding use of randomisation, considering whether any historical data exist for the biomarker with the standard treatment and the reliability of such data. Where a change in a biomarker reflects the biological activity of an agent, but is not predictive of the natural history of the disease, this alone may be an appropriate endpoint for a proof of concept phase II trial; in such cases a second, go/no-go phase IIb trial may be required to assess the impact of the treatment on the cancer prior to a decision on proceeding to a phase III trial. The use of biomarkers as outcome measures is discussed further in Section 2.2.1.

Second, in the era of targeted therapies a molecular characteristic of the tumour that is relevant to the mechanism of action of the treatment under investigation may serve as a biomarker to define a specific subgroup of patients in whom an intervention is anticipated to be effective. This has been done especially successfully in studies of small molecules and monoclonal antibodies targeting HER-2 and related cell surface receptors (Piccart-Gebhart et al. 2005; Slamon et al. 2001). The potential for a biomarker to identify a subpopulation of patients may, however, only become apparent after phase III investigation, as in the case of the monoclonal antibody cetuximab in colon cancer where efficacy is limited to patients with no mutation in the KRAS oncogene (Bokemeyer et al. 2009; Tol et al. 2009; Van Cutsem et al. 2009). Where available, using a biomarker to enrich the population in a phase II trial in this way can increase the likelihood of anti-tumour activity being identified, and thus speed up drug development. By definition, when using a biomarker for population enrichment, the resulting phase II population is not representative of the general population. Interpreting outcomes in the enriched population may, therefore, be more challenging as historical control data may be unreliable; randomisation incorporating a control arm should be considered in such situations.

There are, however, potential risks with an over-reliance on biomarkers in phase II trials. If the mode of action of a novel therapy has been incorrectly characterised, the biomarker chosen for enrichment may be inappropriate and could lead to a false-negative phase II trial because the wrong patient population has been treated. Likewise, if a biomarker used to demonstrate proof of principle of biological activity does not accurately reflect the clinically relevant mode of action, the outcome of a phase II trial may be misleading. When a biomarker is the primary endpoint for a trial or used to enrich the patient population of patients it is vital that the biomarker be adequately validated. Where there is insufficient evidence that a biomarker reliably reflects biological activity or identifies an optimal patient group, measurement of the biomarker in an unselected phase II trial population may be appropriate as a hypothesis-generating exercise for future studies.

Approaches to trial design that incorporates biomarker stratification are discussed further in Section 2.2.3.

Where the toxicity of the investigational treatment is believed to be modest in the context of phase II decision-making or the toxicity of agents in the same class is well known, the primary phase II trial outcome measure will usually be anti-tumour activity, with toxicity included amongst the secondary outcome measures.

If the toxicity profile of the investigational treatment, be it a single-agent or combination therapy, is of particular concern, the activity and toxicity of the treatment may be considered as joint primary outcome measures, such that the investigational treatment must show both promising activity and an acceptable level of toxicity to warrant further evaluation. Such designs allow incorporation of trade-offs between pre-specified levels of increased activity and increased toxicity, to determine the acceptability of a new treatment with respect to further evaluation in a phase III trial.

Response is usually evaluated via a continuous outcome measure, that is, the percentage change in tumour size. This is, however, typically dichotomised as ‘response’ versus ‘no response’ following RECIST criteria (Eisenhauer et al. 2009). Such binary outcomes, categorised as ‘yes’ or ‘no’, may be used for any measure that can be reduced to a dichotomous outcome including toxicity or change in a biomarker. Other outcome measures that may be expressed as continuous, such as time to disease progression, are frequently dichotomised to reflect an event rate, such as progression at a fixed time point.

In phase II studies of cytotoxic chemotherapy the biological rationale for response as an indicator of anti-cancer activity is based in part on the natural history of untreated cancers which grow, spread and ultimately cause death. Administration of each cycle or dose of treatment kills a substantial proportion of tumour cells (Norton and Simon 1977) and as such is linked to delaying death (Norton 2001). These principles may be applicable to chemotherapeutic agents which target tumour cell kill, and therefore the endpoint of response may be a relevant indicator of anti-tumour activity.

There are inherent issues in the assessment of tumour response, associated with investigator bias, inter-observer reliability and variation in observed response rates over multiple trials (Therasse 2002). These may, to some degree, be alleviated by the incorporation of independent central review of response assessments or the incorporation of a randomised control arm when historical response data are unreliable. The use of classical response criteria for trials of drugs that may not cause tumour shrinkage is likely to be inappropriate and raises questions over the design of phase II trials and the endpoints being used (Twombly 2006). Measures of time to an event such as disease progression or novel endpoints such as biomarkers may be more relevant when evaluating the activity of newer targeted therapies. Nevertheless, because most targeted or biological therapies are selected for clinical development on the basis of pre-clinical data demonstrating at least some degree of tumour regression, tumour response may remain an appropriate outcome measure for novel agents, as acknowledged by two Task Force publications (Booth et al. 2008; Seymour et al. 2010).

Continuous outcome measures such as tumour volume or biomarker response may be appropriate and relevant outcome measures for consideration in studies of novel agents (Adjei et al. 2009; Karrison et al. 2007; McShane et al. 2009). The use of biomarkers in clinical trials is becoming increasingly common in the development of targeted treatments with novel mechanisms of action. Only when a biomarker has been validated as an outcome measure of activity, that is, when a clear relationship has been established with a more conventional clinically relevant outcome measure, should a biomarker be used as the primary outcome measure of a phase II trial. The difficulties in identifying validated biomarkers have been highlighted (McShane et al. 2009), in addition to the need for technical validation and quality assurance of the relevant assays. As discussed above, biomarkers may be dichotomised to produce a binary outcome; statistical designs can, however, incorporate biomarkers as a continuous outcome, which may often lead to more efficient trial design.

Multinomial outcome measures may offer an alternative to binary outcomes when multiple levels of a clinical outcome may be of importance. For targeted or cytostatic therapies, an alternative to binary tumour response (i.e. response vs. no response) that remains objective may be the ordered categories of tumour response such as complete response plus partial response versus stable disease versus progressive disease (Booth et al. 2008; Dhani et al. 2009). Alternatively activity of an experimental therapy may be evidenced by either a sufficiently high response rate or a sufficiently low early progressive disease rate (Sun et al. 2009).