Sentinel dosing: A proposed algorithm to guide decision making on which cohorts in early phase clinical pharmacology trials should use this approach

Abstract

Aim

The aim of this paper is to describe a proposal for an algorithm to guide decision making on which cohorts in early phase pharmacology trials should employ a sentinel approach, thereby standardising and harmonising practices, and improving decision making around sentinel dosing. Sentinel dosing is an approach described in EMA guidelines; however, the guidance does not provide detail on which cohorts or dose levels sentinel dosing applies.

Methods

This algorithm was designed using the expertise of senior scientific staff from the largest clinical research organisations in the Netherlands and after consultation with members of the ethics committee handling most early phase clinical drug trials in the Netherlands.

Results

The algorithm describes a decision tree considering different aspects of the design of the trial, the IMP, and the prior knowledge of the IMP based on (pre)clinical data, to be used by investigators and regulators. Hereby the decision‐making process on sentinel cohorts will be tailored to the specific IMP and information available. Starting in 2024, all three expert Phase 1 units that co‐authored this paper are using the algorithm as guidance for decision‐making on the implementation of sentinels in trials that are submitted to the EC Stichting BEBO.

Conclusion

The algorithm provides further guidance and specification on when to implement sentinel dosing in early phase clinical trials, thereby creating a standardised and harmonised approach, which will improve the protection of subject safety and trial efficiency.

What is already known about this subject

• Sentinel dosing is a measure in early phase clinical trials to minimize the risk and harm for participants that was implemented in the wake of the TeGenero and BIAL tragedies.

• Guidance such as the EMA first‐in‐human guideline describes the importance of sentinel dosing, but lacks details on the rationale for when to implement sentinel dosing.

• This leads to a lack of clarity for sponsors, investigators and ethics committees, and subsequently can result in inefficiencies, lengthy discussions and delays.

What this study adds

• This paper describes an algorithm to guide decision making on which cohorts in early phase pharmacology trials should employ the sentinel approach.

• Sponsors, investigators and ethics committees can use this algorithm to standardize and harmonize practices, and improve decision making around sentinel dosing.

1. INTRODUCTION

There is an inherent risk in administration of pharmacological agents to human subjects. Especially when uncertainty is relatively high, such as when a compound is administered for the first time in humans, specific attention should be paid to the risks, and mitigation strategies should be in place to minimize the safety risks for the participants. Guidelines and regulations for such strategies for first‐in‐human trials were originally described by the Council for International Organizations of Medical Sciences (CIOMS) and later adopted in international guidelines. ¹ Nevertheless, several tragedies with severe side effects in participants in early phase clinical trials have occurred. In 2006, the first cohort of subjects receiving a new investigational compound, a CD28 agonist called TGN1412, experienced life‐endangering side effects. ² All six subjects treated with TGN1412 suffered from a cytokine storm leading to severe side effects including tissue damage in the extremities. This tragedy in clinical pharmacology triggered heavy scrutiny of the guidelines and implementation of new safeguards to protect the human subjects in these trials. ³ , ⁴ , ⁵ , ⁶ Various lessons were learned, and guidelines were updated, which in Europe led to the creation of a guideline on strategies to identify and mitigate risks for first‐in‐human clinical trials with investigational medicinal products by the European Medicines Agency (EMA). ⁷ Some ten years later, another catastrophe in clinical trials shook the foundations of clinical pharmacology again. In 2017, five participants in a multiple dosing cohort with the fatty acid amide hydrolase (FAAH) inhibitor BIA‐102474 became severely ill and were hospitalized. One participant ultimately died, and all the five participants were shown to have deep brain haemorrhage and necrosis. ⁸ This led to scrutiny of the trial and its events, ⁹ , ¹⁰ , ¹¹ and another update of the EMA first‐in‐human guidelines in 2017. ¹²

Many aspects of the complex process of drug development were evaluated, and changes in guidelines and regulations were implemented across the board in the wake of these tragedies. One aspect identified carrying risk of an unnecessary amount of harm was related to the fact that subjects in trials were dosed at approximately the same time for a specific dose cohort. As there is always inevitable uncertainty in early phase clinical trials, and therefore a potential risk for participants in these trials, the potential harm could be reduced by dosing only a single subject on an active drug before dosing the rest of the cohort. This was termed sentinel dosing, and this approach aims to minimize the number of participants being exposed at the same time to a new dose level of an (investigational) pharmacological agent.

The sentinel dosing approach is a design where, in a randomized, double‐blind fashion, two subjects are dosed hours or days (e.g. 24 h) prior to dosing the remaining subjects in the cohort. In that sense, these two subjects serve as a sentinel for the rest of the cohort. Of these two subjects, one is on active and one on placebo treatment, to maintain the blind. The adequate time window between the sentinel and the remainder of the cohort is determined based on the anticipated pharmacokinetic (PK) and pharmacodynamic (PD) characteristics of the compound under investigation. This time window may differ between a single ascending dose (SAD) and a multiple ascending dose (MAD) study. After review of relevant safety information and performing the decision process as laid down in the clinical study protocol, the remainder of the cohort can be dosed.

The sentinel approach was adopted in the EMA first‐in‐human Guideline of 2007, ⁷ and maintained with some updates in the revision of 2017. ¹² The specific wording in the current 2017 EMA guideline reads: “It is considered appropriate to design the administration of the first dose in any cohort so that a single subject receives a single dose of the active IMP (often known as sentinel dosing). Flexibility in this approach is allowed but should be on a risk‐proportionate basis with a clear scientific rationale for any proposals not to use this strategy. … This approach is expected for all single and multiple dosing cohorts, in order to reduce the risks associated with exposing all subjects in a cohort simultaneously.” In other words, according to the EMA guideline, every cohort in SAD and MAD studies receiving a new dose should have a sentinel approach, except if there is a clear scientific rationale showing that the risk (or uncertainty) is sufficiently low for that dose. The guidance, however, does not provide much detail on the decision‐making process to determine which cohorts or dose levels in a first‐in‐human trial should implement sentinel dosing.

The aim of this paper is to describe a proposal for an algorithm designed to guide decision making on which cohorts in early phase pharmacology trials should employ this approach, based on specific aspects of the trial, the investigational medicinal product (IMP) and the (pre)clinical package. Thereby, this algorithm could aid to standardize and harmonize practices, and improve decision making around sentinel dosing, both for sponsors, investigators, governing bodies and ethics committees.

2. METHODS

This algorithm was designed based on the knowledge and experience of senior scientific staff of the largest clinical research organizations of The Netherlands, PRA/ICON, QPS and CTC Netherlands and CHDR, in collaboration and after consultation with members of the ethics committee Stichting Beoordeling Ethiek Biomedisch Onderzoek (BEBO), the committee that handles most early phase clinical drug trials in the Netherlands. A first version of the algorithm was drafted by the first author, after which review and input was collected from other senior scientific staff of the different institutes. An informal meeting was held to discuss discrepancies and disagreements, resulting in this final version of the algorithm that reached unanimous accord.

2.1. Nomenclature of targets and ligands

Key protein targets and ligands in this article are hyperlinked to corresponding entries in http://www.guidetopharmacology.org, and are permanently archived in the Concise Guide to PHARMACOLOGY 2021/22. ¹³ , ¹⁴

3. RESULTS

The algorithm is depicted in Figure 1, and provides a decision tree considering different aspects of the design of the trial, the IMP, and the prior knowledge on the IMP based on (pre)clinical data. Both investigators and regulators can run through the algorithm and answer the questions based on the available information. Below we will discuss each decision point or question in the algorithm separately and describe why that question is important for determining the necessity of a sentinel dosing and in which cohorts, and which factors should be taken into account when answering the questions.

FIGURE 1.

Sentinel dosing algorithm decision tree.

In general terms, when both the uncertainty around the potential (side) effects of the proposed dose(s) of the IMP and the risk of harm are considered low, no sentinel is required in a study. With increasing uncertainty, sentinels may be appropriate, and unless an important part of that uncertainty is addressed by a previous human dose cohort, subsequent cohorts may also require a sentinel, or sentinels may have to be re‐introduced at later cohorts when the uncertainty or risk of harm increases again based on emerging data, or increased uncertainty in how to extrapolate information from preceding cohorts to future cohorts. Therefore, the decision on sentinel cohorts should be tailored to the specific IMP and information available. In case of doubt, it is advised to select the conservative answer to a question.

3.1. Is this the first time the IMP is administered in humans (first‐in‐human trial)?

When an IMP is administered to human subjects for the first time (first‐in‐human study, FIH), the knowledge on safety, pharmacokinetics and pharmacodynamics is based solely on in vitro and preclinical data in animals. Usually there is also some knowledge and animal or human data on the targeted mechanism or (patho)‐physiological process involved, and possibly there is information or data available on similar compounds, e.g. if the IMP is not a first‐in‐class. However, regardless of how much information is available, all information requires a translation to the human situation for this specific, new IMP. For each such IMP, there will be inherent uncertainty (including the risk of unexpected [off‐target] effects), because of this translation. The more information available, the smaller the uncertainty could be, but nonetheless a certain amount of uncertainty will always remain that cannot be mitigated. If the translation is wrong, this can have major implications for the human subjects being dosed (see the TGN1412 example ² ). Therefore, current consensus is that one should always apply sentinel dosing in (at least) the first dose of the first cohort for a new IMP. If this is not the first time the IMP is administered in humans, a sentinel may not be required, depending on whether the dose and predicted exposures of the IMP for this new trial have already been covered previously and were determined safe. This is discussed in question 2. Note that the answer to question 1 should also be no for the special cases of biosimilars and of new formulations of a known active substance, as the active substance for these products has been previously tested in humans.

3.2. Are the doses and predicted systemic exposures being considered in this trial already covered by previous clinical trials with the IMP and were these safe?

If the answer to this question is yes, no sentinel is required in the current trial. If the answer is no, we approach another area of uncertainty, namely exposures that have not previously been attained in humans. Whether a sentinel is required then depends on whether these uncertainties and accompanying risks are considered major, which is discussed in question 5.

3.3. Is the IMP a first‐in‐class drug?

While dosing a new IMP in an FIH trial comes with inherent uncertainty, this uncertainty is larger if the IMP is also a first‐in‐class drug. A first‐in‐class drug targets either a new pathway or process, or does so in a new way, and thus has a mechanism of action not previously tested in humans. Because there is no data available from previous compounds with the same (or similar) mechanism of action, there is more uncertainty than when the mechanism has been targeted and modulated in humans before. This includes a risk of unexpected (off‐target) effects. Therefore, the uncertainty at each dose level and cohort is increased, as new ground for that class of drug is being explored, and so when we are dealing with a first‐in‐class drug, the balance will shift to sentinel implementation. However, depending on how much is known about this new mechanism of action, and how predictable or monitorable the effect of targeting this mechanism is, the uncertainty may vary, which is addressed in question 4.

3.4. Is the effect that can be expected from the mechanism of action predictable or is the pathway well known?

Although all IMPs considered under this question are first‐in‐class compounds, there may still be substantial differences in the uncertainties involved with the new mechanism of action or targeted pathway. If the pathway is well known, and the effect that can be expected from the mechanism of action appears predictable, the uncertainty is limited, and so sentinels may not be required in each cohort. However, if the anticipated effect related to the mechanism of action is difficult to predict as little is known about the pathway(s) involved, or multiple pathways or systems are involved, or the target is involved in different essential processes, it is prudent to include sentinels in all SAD cohorts, and possibly in MAD cohorts too (see question 7).

3.5. Are there major risks or uncertainties at the intended doses or predicted exposures based on the available (pre)clinical data?

Apart from the factors identified in all the other questions, there may still be specific major risks or uncertainties related to the doses or predicted exposures of the IMP that cannot be addressed within those factors. These major risks may include:

• Unpredictable translation to humans

• Poorly predictable exposure or response

• Narrow therapeutic window, steep dose‐(adverse) effect curve

• Potentially major effect on immune system, on‐ or off‐target

This should be evaluated based on the available (pre)clinical data, and if this data indicates any of these major uncertainties or risks, this warrants sentinels.

The question whether there are major risks or uncertainties at the intended doses or predicted exposures based on the available (pre)clinical data may be relevant both for IMPs administered for the first time in humans, and for IMPs that have been dosed in humans previously. The considerations for answering this question are the same for both types of IMPs. If the answer to this question is no, the final outcome on implementation of sentinels is different, as depicted in Figure 1; the FIH IMP will require a sentinel only in the first cohort, and the IMP that is beyond FIH does not require sentinels at all. If the answer to this question is yes, sentinel dosing will be required for the IMP, whether it is an FIH or not. The decision on which cohorts require a sentinel is driven by the answer to questions 6 and 7.

3.6. Can these uncertainties be linked to specific exposures or dosing regimen?

This question applies if the answer to question 5 was yes and there are major risks or uncertainties based on the preclinical data for the IMP. If this occurs at specific exposures that can be translated to humans with relative certainty, then sentinels are only required in the cohorts that are expected to reach these exposures. However, if not, all SAD cohorts, and possibly some or all of the MAD cohorts, require sentinels too (see question 7).

3.7. Are exposure or response/toxicity likely to change on continued exposure, or time‐dependent?

Most uncertainties discussed so far do not differentiate between single or multiple dosing. However, some side effects or toxicity may specifically occur after longer exposure, such as with multiple dosing. This may be because the toxicity builds up when exposure remains above a certain threshold for a longer period of time (see, for example, the BIAL case ⁸ ). If there are no signs that the response or toxicity to the IMP would change on continued exposure, there is no need to include sentinels in the MAD cohorts, as long as the predicted exposure for such a MAD cohort is covered by exposures obtained in the SAD cohorts. However, if there are indications, for example in the preclinical data, that with similar exposure at single and multiple dosing, toxicity is observed only in the multiple dosing setting, sentinels are required also in the MAD cohorts, as the observations in the SAD cohorts may not predict the safety of similar exposures in MAD cohorts. The actual duration of the sentinels during repeated dosing of the sentinel group will need to be tailored to the IMP, based on the observed PK characteristics such as time to reach steady state, and including the interpretation of the preclinical safety observations and how these are to be extrapolated to humans.

4. DISCUSSION

Here we describe a tool to guide investigators, sponsors and regulators on how to implement sentinel dosing and in which cohorts, in early phase clinical pharmacology trials. The EMA first‐in‐human guidelines specify that such a sentinel approach is warranted in every cohort, except if there is good scientific rationale to deviate from this, but there is no further specification. The aim of the proposed algorithm (Figure 1) is to provide further guidance and specification, and thereby create a standardized and harmonized approach. This should not only improve practices on sentinel dosing and therefore protection of subject safety, but also help sponsors, investigators and regulators to set the standards. This creates clarity, leading to reduction of risk to study participants, and to more efficiency. It will reduce the risk of questions and discussions during the review of the Ethics Committee (EC) and help sponsors and investigators to create an adequate planning for the study in a clinical research unit, and prevent having to amend that planning due to requested changes in the sentinel design.

For making optimal use of the potential of the algorithm, it is recommended that sponsors/investigators include the underlying rationale for the answers/decisions in the algorithm in the submission documentation. This way, the EC can review these decisions, and an explicit evaluation of the design in terms of sentinels is then made by all stakeholders.

We designed this algorithm to be as broadly applicable as possible. Therefore, it encompasses all administration routes, although it focuses on systemic dosing (e.g. oral, intravenous, subcutaneous, intrathecal). When using the algorithm for topical dosing, one should consider the local dose and exposure, as well as the resulting systemic exposure. Furthermore, it can be applied to both small molecules and biologicals. And it can be used in the special case of biosimilars or new formulations of known active substances, although for these products, note the remark described under question 1.

Obviously, the algorithm should not be used by sponsors or investigators for manipulating answers towards an outcome favouring fewer sentinel cohorts. We believe that if the questions are addressed truthfully and to the best of the user's knowledge, the outcome will be in line with current thoughts and guidelines on protecting subject safety by means of sentinel dosing. Thereby, this approach will expedite clinical drug development by mitigating risk of delays and failure due to unexpected safety findings. Of course, an algorithm will never cover all scenarios that may occur; when in doubt on answering a certain question, we therefore recommend selecting the more conservative answer, to minimize risk to the human subjects participating in these trials, and maximize the likelihood of cautious, successful early clinical drug evaluation.

Starting in 2024, all three expert Phase 1 units that co‐authored this paper are using the algorithm as a guidance for decision‐making on implementation of sentinels in trials that are submitted to the EC Stichting BEBO. So far, this has not led to new major discussions on this subject, but the implementation time has been too short to reach final conclusions.

AUTHOR CONTRIBUTIONS

JAACH, PdJ, EJvH, WT, PdG and GJG made substantial contributions to conception and design. JAACH drafted the manuscript, and PdJ, EJvH, WT, PdG and GJG were involved in revising the manuscript critically.

CONFLICT OF INTEREST STATEMENT

The authors have no conflicts of interest to declare.

Heuberger JAAC, de Jongh P, van Hoogdalem E‐J, Tamminga W, de Graeff P, Groeneveld GJ. Sentinel dosing: A proposed algorithm to guide decision making on which cohorts in early phase clinical pharmacology trials should use this approach. Br J Clin Pharmacol. 2025;91(9):2705‐2710. doi: 10.1002/bcp.70090

DATA AVAILABILITY STATEMENT

Data sharing is not applicable to this article as no new data were created or analysed in this study.