More on first-in-man studies

Stopping rules for dose escalation

In October's ‘Editors' view’ we visited the art and science of estimating an appropriate dose for first-in-human studies [1]. In this issue of the Journal we publish a safety grading scale to support dose escalation during single ascending dose (SAD) and multiple ascending dose (MAD) early phase studies in healthy young adult men [2]. It describes how such grading can be used to help decide whether to stop the study or to continue dose escalation.

Previous grading scales have provided guidance for first-in-human studies in patients with cancer [3–5] or HIV infection [6], applications for which appropriate reference ranges differ importantly from ranges in healthy young men. A grading scale has also been proposed for adults and adolescents who are subjects in trials of prophylactic vaccines [7], but the adverse effects of vaccines are much less diverse than the adverse effects of other biological products and drugs of low molecular weights, so there is a real need for guidance and consistency in this important kind of investigation. As with earlier proposals [3–7] the authors grade the intensity of adverse effects that are discontinuous variables (e.g. symptoms such as fatigue or nausea) in terms of their impact on daily activities and whether active treatment is needed (grade 1 events do not interfere with daily activities, grade 2 events interfere with daily activities but require no treatment other than paracetamol, grade 3 events interfere with daily activities or require treatment, grade 4 events are life-threatening). Continuous variables (such as electrocardiographic QT interval, serum creatinine concentration) are scaled according to threshold values or changes from baseline. When a cohort of subjects has been dosed, these grades are used in an algorithm to support decisions regarding dose escalation.

Decisions regarding dose escalation involve balancing risks (e.g. of discomfort from nausea) to individual volunteers against the risks of failing to define the correct dose range to use in phase II studies, or even of killing off further development of a potentially useful product. This can occur as a result of mis-identifying a ‘maximum tolerated dose’ (MTD) on the basis of an intercurrent event (e.g. severe diarrhoea caused by an undiagnosed enterovirus) or within-subject random variation in vital signs or laboratory values. Variation in laboratory data is inevitable, and since many different analytes are measured repeatedly in each subject, some values outside of the reference ranges are to be expected. Deciding whether to stop dosing or to escalate is straightforward at the extremes. If after dosing there are no symptoms and no safety signals from vital signs or laboratory tests, and a predefined target drug exposure (in terms of AUC and C_max scaled to the no observed adverse effect level, NOAEL, or the minimum anticipated biological effect level, MABEL, as mentioned previously [1]) or, most importantly, a desired pharmacological effect has not been achieved, then the investigation proceeds per protocol to the next dose level. Dose escalation based solely on clinical chemistry and symptoms while neglecting pharmacology risks throwing out the baby with the bathwater, for example, when harmful effects due to collateral pharmacology occur only at doses where receptor occupancy-initiated primary pharmacology is already appreciable, especially in the case of highly novel prototypical drugs [8]. By contrast, in the exceptional circumstance of a life-threatening adverse event that is attributable to the drug, the investigation would stop, or indeed might have never be undertaken in the first place, if the preclinical pharmacology had been better understood [9]. Difficulty arises in the grey area between these extremes, and it is here that the safety grading scale proposed by Sibille and his colleagues [2] could be most useful.

Regulatory agencies and research ethics committees rightly insist on predefined stopping rules as part of the protocol: investigators have conflicting interests, and such decisions have considerable commercial implications and should be made as objectively as possible. This is more easily said than done, because combining qualitative symptoms of various intensities with quantitative laboratory values is like combining apples with pears (though with a less happy result than in a really good winter pudding!). Consequently, a definitive guideline based on a hard and fast calculation, seductive though this would seem, is not achievable. Instead, Sibille and his colleagues in ‘Club Phase I’ (CPI) offer a grading system and algorithm to generate ‘points to consider’ in the context of the clinical pharmacology of the drug in question and individual aspects of the protocol. Should dose escalation stop and the MTD be defined on the basis of the adverse events in this cohort at this dose? Should a cohort be repeated at the same dose, perhaps altering some other aspect of the protocol – a high fat breakfast, perhaps – that may have caused or contributed to the adverse events? Should new specific measures be introduced to minimize risk, such as increasing the dose interval in subsequent subjects? Inevitably, as with deciding on the first dose, subjective judgement enters into decisions on dose escalation. A strength of the CPI proposals [2] is that they would increase consistency in the way data from early phase studies are recorded and analyzed, without introducing a spuriously quantitative set of rigid rules. The CPI working group indicates that it is open to comment, and the editors would like to encourage debate on this important matter.

Other practical issues: sentinel groups and presentation of clinical data

As described in our earlier editors' view [1], the TGN1412 experience led to advice from an expert advisory group set up by the MHRA and chaired by Sir Gordon Duff [10] as to non-standard situations, in contrast to situations such as a new generic version of a drug whose pharmacology is well established, in which special caution should be exercised. Such situations include early human investigations of prototypical drugs [8], for example drugs that are first in class, particularly if their anticipated physiological effects are potentially profound. Special care is also needed when species specificity is plausible and when the drug is an agonist at cell surface receptors, especially when there is the potential for cross-linking of receptors [10]. One practical measure in such non-standard cases is the use of a ‘sentinel’ group, comprising one active- and one placebo-treated subject at the start of the study and at each dose increment. The intention is to identify any highly repeatable serious AE in a single subject rather than being faced with an entire cohort in trouble, as with the volunteers at Northwick Park Hospital [9]. Timing the dosing in such a sentinel group vs. the remaining members of the cohort is of great importance. It is essential to have adequate time to analyze and review chemical pathology samples and other clinical and laboratory data if the purpose of having such a group is to be achieved, and in practice this will usually require at least 48 rather than 24 h. Even when adequate time is allowed for review of laboratory data, it must be appreciated that some adverse effects will take much longer than this to manifest (consider the effect of a toxic dose of warfarin or of a cytotoxic drug); protocols are necessary but must not be allowed to substitute for individual thought, action, reflection and responsibility!

Phase I units rightly stress the importance of data integrity and accountability, and information technology systems for handling phase I trial data reflect these concerns. A disadvantage is that such systems often do not facilitate bedside access to raw data that can alert clinicians to worrying trends, such as progressive tachycardia, a fall in blood pressure, a rise in temperature, or a change in respiratory rate, any of which can presage cardiorespiratory arrest or another clinical emergency. Since phase I studies have for the most part been very safe, this would have proved problematic only in uncommon circumstances, such as those that occurred in relation to TGN1412 [11]. It would be paradoxical if attention to good clinical practice in data management had the unintended consequence of making human volunteers less safe, by impeding bedside evaluation of trends in vital signs and other clinical data during the prodromal phase of an unfolding severe adverse event.