How to Design a Phase I Trial of an Anticancer Botanical

Abstract

Phase I trials are an important part of traditional drug development in oncology. Such trials address two key issues: safety and dose. Currently, there is a dearth of phase I trials of anticancer botanicals. This may result from the apparently widespread view that a history of human use precludes the need for early-phase study. However, the safe use of a botanical by the population at large does not guarantee safety when the botanical is used in combination with other agents in the complex medical setting of oncology. Several cases of unpredictable adverse events have been recorded following the use of botanicals by cancer patients.

We propose a simple, robust design for phase I trials of anticancer botanicals. This design incorporates important characteristics of botanical medicines including low toxicity, prior data on a likely safe dose, a limit on the highest dose it is feasible to administer, and the unknown relationship between dose-toxicity and dose-response curves. Two principal design features are the use of predetermined dose levels and the direct measurement of a response endpoint such as survival or immunity. This response endpoint can be used to determine the optimal dose if toxicity is acceptable at all dose levels.

Increasing the use of phase I methodology would ensure a more systematic development of botanicals as anticancer agents. This would likely increase the chance that at least one such agent would be proven to extend lives.

Traditional oncology drug development occurs in several stages. The most well known, and important, is the phase III trial. This is often described as the “definitive” study as it provides an answer one way or another as to the clinical value of an anticancer agent. In a typical phase III trial, many hundreds of patients are randomly assigned to receive either what is currently considered the most effective regimen or an alternative regimen including the new agent. Patients are then followed up for many years. It is normally only after a treatment has been shown to improve survival compared with standard care in a phase III trial that an agent enters clinical practice.

There are a number of things we might wish to know about a drug before we would be ready to conduct a large and expensive phase III trial. Two key issues are safety and dose: Can a drug be given safely? What dose optimizes anticancer activity at an acceptable level of toxicity? These research questions are addressed in what is known as a phase I trial. Such trials are a common and important part of cancer drug development; a search of MEDLINE in November 2005, for example, retrieved 6,000 hits for “phase I” and “cancer.” However, we have previously shown that phase I trials of unconventional agents such as botanicals are extremely rare.¹ In a systematic review, we identified 198 clinical trials of unconventional anticancer agents. We classified 105 as phase III but only 23 as phase I trials. A mere 8 of these phase I trials were designed to establish an appropriate dose. This is in sharp distinction to conventional oncology, where there are approximately five times as many phase I trials as phase III trials.

One possible reason for the relative dearth of phase I studies of anticancer botanicals is the apparently pervasive view that a history of human use precludes the need for early-phase study. It has been argued that a typical phase I trial is conducted on a drug that has never been administered to a human. It is appropriate therefore to start slowly and carefully and give the drug only under the most controlled experimental conditions, starting with very low doses. In this view, tight experimental control is unnecessary for an agent that has been used for hundreds or thousands of years and is available in the local health store.

Such an argument would constitute an argument against phase I trials of botanicals only if their sole purpose was to establish that an agent was not overly toxic to a patient in relatively good health. Phase I trials in oncology may have additional purposes. First, with respect to safety, it is quite possible that a botanical used safely in the community has unexpected adverse effects in cancer patients, resulting from the complex medical history and concurrent treatment of these patients. One well-known example concerns interactions between St. John's wort and chemotherapy. Although St. John's wort appears to be taken safely by many hundreds of thousands of patients every year, it has been shown to reduce blood levels of the active metabolite of irinotecan.² Note that this is highly serious: if lowered chemotherapy levels were to lead to reduced tumor cell kill, a patient's life could be shortened by the use of St. John's wort. As a second example of a drug-botanical interaction, we have found that the addition of an immunostimulant botanical to a monoclonal antibody regimen reduces the effects of the antibody if a concurrent conventional immune adjuvant is used (unpublished data). This is apparently because the combination of conventional and botanical adjuvant potentiates an endogenous immune response against the antibody. We also found that use of the botanical in pediatric cancer patients immunocompromised after chemotherapy led, in rare cases, to thrombocytopenia (unpublished data). Such adverse effects related to reduced effectiveness of oncologic agents or special medical problems of cancer patients are completely unpredictable from the safety record of botanicals in general use.

The second reason to conduct phase I trials of anticancer botanicals is to determine dosing. There is no particular reason to believe that the doses used in the community, or those recommended by practitioners or manufacturers, are optimal. An interesting comparison can be drawn with the botanical treatment of conditions other than cancer, of which eczema would be an illustrative example.³ A practitioner of herbal medicine treating eczema can tell within a few weeks whether the treatment appears successful and adjust the dose accordingly. Over the course of treating many patients, a practitioner will titrate the dose toward an optimum. The problem for the botanical treatment of cancer is that the effects of treatment are not easy to discern. Unlike with eczema, treatment results cannot be judged visually within a week or two; indeed, it is unclear what information a medicine practitioner could use to determine success: if a patient survives a certain period of time, would they have survived as long (or longer) without treatment?

Phase I trials of anticancer botanicals can be recommended as a systematic means to determine safety when used in the complex medical setting of cancer. They may also establish optimal dosing. In this article we describe traditional approaches to phase I trials, discuss why these may not be applicable to botanical agents, and outline a simple alternative methodology of wide applicability.

TRADITIONAL PHASE I DESIGN

Cytotoxic chemotherapy is the most common type of intervention studied in phase I oncology trials. Such agents have typically never been used by humans before initiation of the phase I trial and are relatively toxic. Chemotherapy is normally administered by infusion; accordingly, and unlike drugs given orally, there are no important practical constraints on dose. Moreover, the dose-response curve is known to be “monotonic”; that is, cancer cell kill increases with ever higher doses (Figure 1). This means that the optimal therapeutic dose is the highest that the patient can tolerate. A related point is that the dose-toxicity and dose-response curves are thought to be relatively close together (see Figure 1). Indeed, many of the toxicities associated with chemotherapy provide evidence of anticancer activity. For example, immune suppression occurs because chemotherapy kills rapidly dividing immune cells and can therefore be taken as an indication that a particular dose of an agent will also kill tumor cells.

Figure 1.

Dose-response (solid line) and dose-toxicity (dotted line) curves for a typical cytotoxic drug. The arrows show the maximum tolerated dose (MTD); small reductions in the dose below the MTD lead to large decreases in effectiveness (response).

Each of these characteristics of cytotoxic drugs influences traditional phase I design (Table 1). The general outline of a phase I trial is that patients are treated in cohorts of three or six at a particular dose. Any toxicities resulting from treatment are graded. The important distinction is between grade 1 and 2 toxicities, and the more serious grade 3 and 4 toxicities, which may require drastic treatment or be life threatening. For example, grade 1 or 2 diarrhea is an increase in the number of stools per day; grade 3 diarrhea might involve need for IV fluids, and grade 4 might be used to indicate hemodynamic collapse. Dose is increased for new cohorts until 33% of patients experience a grade 3 or 4 toxicity. The preceding dose is classified as the maximum tolerated dose (MTD) and is the dose chosen for further study.

Table 1.

Design Features of Phase I Trials Relating to Characteristics of Chemotherapy Agents

Characteristic of Chemotherapy	Design Feature
Drug never been used in humans	Start with a very low dose
Drug is relatively toxic
No important practical constraints on dose	Increase dose by a fixed proportion between dose levels
Dose-response curve is monotonic	Chose highest dose that patients can tolerate
Dose-response and dose-toxicity curves are close

To illustrate this design, a typical trial is shown in Figure 2. Note that patients are treated initially in groups of three per dose level, with an additional three patients treated if one of the initial three experiences a toxicity. Note also that the rate of dose escalation is initially rapid (50%) but decreases as toxicities are documented. For a typical phase I trial published in the Journal of Clinical Oncology, see an article by Faivre and colleagues.⁴ For more on phase I design, see articles by Storer and by Geller.^5,6

Figure 2.

Results of a typical phase I dose-escalation trial for a chemotherapy agent.

PROBLEMS APPLYING THE TRADITIONAL DESIGN TO BOTANICALS

Although the traditional phase I design has worked well for cytotoxic drugs, there are various reasons why it appears inappropriate for botanicals. First, toxicity tends to be very low. An agent such as maitake mushroom, for example, does not ever seem to have been associated with side effects, let alone serious toxicities requiring hospitalization. It appears plausible that a phase I trial of maitake could continue indefinitely, with ever higher doses. This relates to a second drawback of the conventional design: there are practical constraints on dose for many botanicals. Increase a dose of a chemotherapy agent 100-fold and, toxicity notwithstanding, it could probably still be infused from a single IV bag; increase the dose of a botanical 100-fold and a patient might have to take, say, 300 capsules a day.

Perhaps most critically from a methodologic point of view, there is no particular reason why a traditional design should provide the optimal dose for a botanical. In brief, this is because, in contrast to cytotoxic drugs, the relationships between dose, toxicity, and therapeutic response have not been well studied. Figure 1 shows the dose-toxicity and dose-response curve for most chemotherapy agents. The dose at which 33% of patients experience toxicity, the usual threshold for an acceptable level of toxicity, must be considered optimal: raising the dose leads to a significant increase in toxicity; lowering the dose leads to a large decrease in response. Hence, choosing the dose at which one in three patients experience toxicity is a rational strategy.

The comparable curves for a botanical are usually unknown. However, Figure 3 shows one plausible scenario for a low-toxicity agent. Note that the dose for a 33% toxicity rate is far from optimal: the dose can be substantially decreased without an important effect on response. Indeed, it seems that the activity of this agent can be maximized with almost no toxicity. It is also possible that some agents have what is called a “non-monotone” dose-response function, with some high doses actually being less effective than intermediate doses. We have seen such an effect in preliminary laboratory studies of immunostimulant agents in our laboratories at Memorial Sloan-Kettering Cancer Center. As can be seen in Figure 4, the optimal dose is not close to the highest tolerable dose. Indeed, this dose is considerably less effective than the optimum.

Figure 3.

Theoretic dose-response (solid line) and dose-toxicity (dotted line) curves for a low-toxicity agent. The arrows show the maximum tolerated dose (MTD): reductions in the dose below the MTD lead do not have a marked impact on therapeutic effectiveness (response).

Figure 4.

Theoretic dose-response (solid line) and dose-toxicity (dotted line) curves for an agent with a “non-monotone” dose response. The arrows show maximum tolerated dose (MTD): the agent is less effective at the MTD than at a lower dose.

SIMPLE, ROBUST DESIGN FOR PHASE I TRIALS OF BOTANICALS

The two key features of the proposed design are a predetermined dose-escalation schedule and measurement of a response endpoint related to agent effectiveness. With respect to dose escalation, the researchers pre-specify particular dose levels at the start of the trial and patients are only treated at one of these doses. The use of a response endpoint relates to the point illustrated by Figures 3 and 4: if toxicity is not a good surrogate for effectiveness, effectiveness must be directly measured in some way. Some possible response endpoints are given in Table 2. The choice of endpoint depends upon either the agent's mechanism of action or the patient group to be studied, or both. For example, in a study of an immunostimulant in early-stage breast cancer after surgery, there is no tumor, so tumor response cannot be an end point; very few patients recur during the study, so survival cannot be measured; and there is no biomarker comparable to prostate-specific antigen. Hence, in an ongoing study of this type at Memorial Sloan-Kettering Cancer Center, we are taking blood from patients before and after a 3-week period of maitake therapy. We conduct a battery of immune assays—including immunophenotyping (ie, counts of different types of immune cells), respiratory oxygen burst, and levels of cytokines such as interleukin-6, interleukin-2R, and tumor necrosis factor α—and then compare changes from baseline across dose levels.

Table 2.

Possible Response Endpoints for Phase I Trials

Survival

Time to death (d)

Time to progression (d)

Time to recurrence (d)

Death or progression at 6 mo (yes/no)

Tumor response

Partial or complete response (yes/no)

Stable disease (yes/no)

Pharmacokinetics

Area under the curve

Peak concentration

Biomarker

Biomarker response (eg, 50% reduction in PSA)

Biomarker change (eg, absolute reduction in PSA)

Ex vivo assay

Cytotoxicity of plasma

Target-based assay

Immune assay

PSA = prostate-specific antigen.

Each of these two design aspects addresses a specific feature of botanical research. The predetermined dose-escalation schedule reflects that there is knowledge of dosing from prior human use, particularly with respect to the highest feasible dose. The use of a response endpoint addresses the possibility that dose-response is non-monotone (as in Figure 4) or that there is strong separation of dose-response and dose-toxicity curves (as in Figure 3).

The design described below is based on ongoing statistical research that we hope to publish in the near future. The actual design steps are as follows:

1. Choose the first dose level. This should be a dose that is highly likely to be safe, based on current human use, and has a reasonable possibility of being effective.

2. Choose the final dose level. This should be the highest dose that it is feasible to take. Likely considerations include the number of pills, capsules or vials, or volume of an oral agent.

3. Choose two intermediate dose levels. Intermediate dose levels are normally chosen so that they are equally spaced on a logarithmic scale; that is, the percent increase is the same from dose level to dose level. For example, if the first dose level is three 250 mg capsules/d and the final dose 25 capsules, intermediate doses of 6 and 12 capsules would mean that dose is approximately doubled between dose levels.

4. Accrue six patients on the first dose level. Document the toxicity and measure the response endpoint.

5. Decide on the dose escalation. If two or more patients experience grade 3 or higher toxicities, stop the study; otherwise, continue to the next dose.

6. Accrue six patients on the next dose level. Document the toxicity, measure the response endpoint, and decide on dose escalation.

7. Repeat steps 5 and 6 until the final dose level is reached or the trial is stopped for toxicity.

8. Determine the optimal dose. In some cases, it may be possible to choose a dose on a simple eyeballing of the data, for example, if response is far superior in the highest dose compared with that in the lower doses. However, in many cases, a more formal statistical analysis may be required.

STATISTICAL ANALYSIS OF PHASE I: DETERMINING THE OPTIMAL DOSE

The purpose of any statistical analysis of phase I data is to help determine the most effective dose. As such, analysis should focus on estimation (a “best guess” of the optimal dose) rather than inference (p values). Statistical evaluation of dose response is not straightforward, and a professional statistician should be involved. The first step is to fit a statistical model where the response endpoint is the dependent variable (on the y axis) and dose is the predictor (on the x axis). Initially, both a linear and a quadratic term (dose and the square of dose) should be included. This allows the model to incorporate a non-monotonic dose-response curve (such as that in Figure 4). However, in our experience, such a dose-response relationship is rather unusual, and it is possible that chance variation may lead investigators to conclude that response worsens with increasing dose. This would result in the selection of an inappropriately low dose. We therefore suggest that both linear and quadratic terms should be kept in the model only if the joint hypothesis test for both terms is statistically significant; if the joint test is not significant, only the linear term should be retained.

After the model is chosen, it should be used to predict response for all possible dose levels. For example, if the minimum dose in the trial is 750 mg, and the maximum dose is 6,250 mg, and one capsule contains 250 mg, the response should be predicted at 750, 1,000, 1,250, 1,500, and so on up to 5,750, 6,000, and 6,250 mg.

The actual choice of dose for further study is a matter of investigator judgment; we do not believe that dose can be determined by algorithm. For example, imagine that doses between 5,000 and 6,250 mg have similar levels of response, although response does increase slightly with the higher dose. Whether the investigators choose to select 4,000 or 5,000 mg might depend on whether toxicity is reported, tolerability of the medicine and compliance, and the strength of evidence that there is indeed a dose-response relationship.

TESTING FOR METABOLIC INTERACTIONS BETWEEN BOTANICALS AND CONVENTIONAL AGENTS

Botanicals may interact with conventional agents via a large number of different mechanisms.⁷ If a regimen is proposed that combines a botanical with a conventional agent, special steps may be necessary to test for interaction. One important type of interaction is metabolic interaction, where the botanical induces changes in the metabolism that alter the pharmacokinetics of chemotherapy. The interaction between St. John's wort and irinotecan, described above, is perhaps the best example.²

A design to test for metabolic interactions is shown in Figure 5. In brief, consent is obtained from the patient before he or she starts a new chemotherapy regimen. Blood is sampled regularly immediately after the first infusion to determine drug pharmacokinetics. A few days later, the patient starts taking the botanical. Pharmacokinetics is repeated for the second infusion: blood levels of the chemotherapy agent before and after botanical use are then compared to determine whether the botanical affects pharmacokinetics.

Figure 5.

Study design for determining whether a botanical affects pharmacokinetics (PK) of a chemotherapy agent.

Note that this design does not test for all types of interaction. For example, it has been postulated that botanicals such as garlic, which contain antioxidant constituents, may protect tumor cells from chemotherapy agents acting through oxidative damage.⁸ Such interactions can only realistically be tested for in a large, randomized trial.

CONCLUSIONS

There appears to be a consensus that the lack of phase I trials of anticancer botanicals is because such trials are not necessary; after all, these agents are “known to be safe.” It has also been argued that such trials cannot be done because the maximum tolerated dose is irrelevant for botanicals as they generally have a low toxicity. In this article we provide both a rationale and a methodology for phase I trials of anticancer botanicals. Increasing the use of such a methodology would ensure a more systematic development of botanicals as anticancer agents and would likely increase the chance that at least one such agent would be proven to extend lives.