Phase I Designs

The clinical trial paradigm was developed in the setting of evaluating cytotoxic agents or combinations of cytotoxic agents. The first step in assessing a cytotoxic agent in humans, a phase I clinical trial, is to determine its safety profile, pharma-cokinetic characteristics, and a schedule for its administration. Trial designs were developed under the assumption that as the amount of the agent administered increases, the number of tumor cells destroyed increases, and the patient's ability to tolerate the agent decreases. The traditional goal of a phase I trial is to find the maximum amount of the agent that can be administered without inducing intolerable toxicity [maximum tolerated dose (MTD)]. The intent is one of estimation, not hypothesis testing. As the focus is tolerability not clinical benefit, the size of phase I trials is kept small (three to six patients per dose level), intrapatient dose escalation is not usually seen, and enrollment is open to patients of any tumor type, who have exhausted standard treatment options.

The traditional phase I clinical trial design begins enrollment at a dose based on findings from animal studies—for example, one-tenth of the murine equivalent LD10 (the dose that is lethal in approximately 10% of mice it is administered to). The dose escalation scheme is based upon the modified Fibonacci series (e.g., increase by 100%, 65%, 50%, 40%, and then 33% for all subsequent dose levels) as well as a threshold for declaring when excessive toxicity has been encountered (e.g., 33% of the patients treated at the dose will develop an unacceptable degree of toxicity). The toxicities associated with an experimental agent may be anticipated from the toxicities observed in animal or other human studies or the toxicity profile of agents in the same class of drugs. A set of toxicities and the degree of their severity

[called dose-limiting toxicities (DLTs)] is defined, which halts dose escalation at the present dose level if an unacceptable proportion of patients treated at that dose level develop a DLT. This event leads to the conclusion that the dose level currently being examined induces unacceptable levels of toxicity and the next lower dose level is declared to be the MTD. See Geller (2) for more details on traditional phase I clinical trial design.

Alternatives to the traditional design have been proposed to reduce the number of patients, who are exposed to dose levels that may not be therapeutic or too toxic and to shorten trial duration. Storer (3) considered a dose escalation scheme in which one patient per dose level is enrolled until a DLT is observed, and then a traditional dose escalation scheme is initiated starting at the dose just prior to the dose producing the DLT. This approach was regarded by some to be too aggressive in the pace at which doses were escalated. Korn et al. (4) offered an alternative to counter this concern in the setting in which the DLT is a severe [National Cancer Institute Common Terminology Criteria for Adverse Events (NCI-CTCAE) Grade 3] or life-threatening toxicity (NCI-CTCAE Grade 4). They suggested that two patients per dose level be enrolled until a DLT is observed or until the second instance of a moderate (NCI-CTC Grade 2) toxicity is reported, then enrollment continues using a traditional dose escalation scheme (4,5).

Ivanova et al. (6) examined the properties of several up-and-down rules for sequentially assigning patients to a dose level. One rule proposed by Gezmu (7) suggests entering as many as k patients per dose level, where p is the target probability of developing a DLT at the MTD and k is the value that satisfies the equation p = 1 — (0.5)1//k. The rules for dose escalation are as follows: If the last patient entered on trial was enrolled onto the dth dose level and developed a DLT, then the next patient enters onto the (d — 1)th dose level. If the last patient entered was enrolled onto the dth dose level and did not develop a DLT and the previous k— 1 patients entered were enrolled on the dth dose level and did not develop a DLT, then the next patient enrolled is entered onto the (d + 1)th dose level. If the last patient entered was enrolled onto the dth dose level and did not develop a DLT and the previous j patients entered were enrolled onto the dth dose level and did not develop a DLT but j < k — 1, then the next patient enrolled is entered onto the dth dose level.

Another variation of an up-and-down rule was proposed by Narayana (8) in which the criteria for assigning the (s + 1)th patient to a dose level is based on the proportion of patients, who develop a DLT among those enrolled on the jth dose level up to that time point, pjs. Ivanova et al. (6) found that these designs perform well in terms of assigning patients to doses close to the MTD.

The up-and-down designs of Narayana (8) and Ivanova et al. (6) use dose escalation schemes in which the toxicity data from patients treated at the current dose level are used to determine the dose level to assign to the next patient enrolled. The Bayesian approach does not restrict its dose escalation criteria to the toxicity data from the current dose level, but uses all the toxicity data up to that point. The Bayesian approach not only requires specifying the set of toxicities to be considered DLTs, the target probability for developing a DLT at the MTD, and the dose levels of interest, but requires the investigators to specify a functional form for the dose-toxicity relationship (e.g., a logit or probit model) and a probability distribution function for the unknown parameters of the dose-toxicity relationship (e.g., the uniform or beta distributions). Based on these initial specifications, O'Quigley et al. (9) proposed that the dose with the estimated probability of inducing a DLT nearest to the target probability be given to the first patient enrolled. After each patient is entered onto the trial and her toxicity results are ascertained, the probability distribution function for the unknown parameters of the dose-toxicity relationship is updated using Bayes theorem. The expected values of the unknown parameters are used to update the dose-toxicity function and the dose with the estimated probability of inducing a DLT nearest to the target probability is assigned to the next patient enrolled. This approach is referred to as the continual reassessment method (CRM) (9,10). Several recommendations have been made as to when to end enrollment such as after a prespecified number of patients have entered, when a prespecified number of patients have been treated at the nominal MTD, or when the estimate of the MTD is within 10% of its previous estimate.

CRM has been criticized for many reasons including the time required to complete such a trial since the jth patient's outcome must be known before the (j + 1)th patient can be enrolled, the appropriateness of the initial guess as to the dose-toxicity relationship, the possibility of a sizeable increases in the dose, and the possibility of enrolling patients at too toxic a dose level. Goodman et al. (11) demonstrated that restricting dose escalation to one dose level (de-escalations can be any number of dose levels), starting at the lowest dose level, assigning two to three patients per dose level, and terminating enrollment at a prespecified number of patients can shorten trial duration and incidence of DLTs.

Moller (12) suggested a blending of the standard and CRM designs to estimate the MTD. Patients are enrolled onto the trial using the standard trial design until the first DLT is observed. The data from this first stage of the trial are used to find an initial estimate for the dose-response relationship. The second stage of the trial is then carried out using the CRM dose escalation.

The growing body of knowledge in the area of tumor biology and cancer genetics has led to the development of agents that target errant signaling molecules, cell cycle checkpoint control, tumor differentiation, tumor suppressor oncogenes, angiogenesis, apoptosis, and immune intolerance. These agents seek to alter a critical mechanism of action driving the malignant process. Their effect may not result in the tumor shrinking, but in arresting its progression. In addition, the targets of these agents may be expressed differently in tumor cells compared to normal cells, and the mechanism by which a biological effect is gained may be different than the mechanism that induces severe or life-threatening toxicities. Thus, the relationship between the dose administered and the biological effect of the targeting agent may not follow the behavior associated with a dose-toxicity relationship. The dose where maximum modulation of the target occurs may not correspond to the MTD. As such, the goal of a phase I trial of a targeted agent (that does not shrink the tumor in a dose-dependent manner) becomes one of finding the dose that corresponds to the maximum modulation of the target while maintaining the tox-icity rate within an acceptance range. Korn et al. (13) pointed out that finding the dose that provides maximum modulation of the target requires more patients than the number traditionally enrolled in phase I trials.

Enrollment to these trials is often limited to patients of a single tumor type whose tumors have the target under investigation. The starting dose is based on the results of animal models. Difficulties arise in defining what is considered a biologically meaningful effect, the minimum proportion of patients for whom the effect should be observed, and how aggressively doses can be escalated. Requisite to designing trials to evaluate biological parameter is the ability to obtain a measure of biological effect, such as through serial biopsies or noninvasive techniques that yield a surrogate for the biological effect of the agent (e.g., serum levels of the target, imaging parameters from PET or MRI scans). Presence of such a marker is not sufficient; the marker must also be able to be measured in a valid, reliable, and repeata-ble manner, and supported by evidence that modulation of the target correlates with clinical benefit (14-18).

Hunsberger et al. (19) suggested that a biologically adequate dose be sought where the dose yields a prespecified level of effect or lies in the plateau of the dose-response curve where neighboring dose levels yield effect levels within 10% of each other. They examined dose finding trial designs where the targeted agent does not induce excessive toxicity so that the dose escalation scheme is solely based on the dose-response curve. When the response rate is thought to increase with increasing doses of the agent, a modification of the standard dose escalation scheme is suggested based on the maximum response rate considered too low to be of clinical benefit and the minimum response rate considered to be of potential clinical benefit. When the dose-response curve is thought to plateau, the dose escalation scheme is based on the slope of the regression line through the last highest three to four dose levels. When the estimated slope of this regression line is near or below zero and at least one response rate is seen among these three to four highest dose levels, dose escalation stops. The recommended dose is the one with the highest proportion of responses.

Thall et al. (20) proposed a Bayesian dose-finding strategy where the "best" dose for the next cohort of patients depends upon the prespecified values for the minimum probability of a biological effect and the maximum probability of a DLT, as well as the toxicity and biological effects seen among all of the patients previously enrolled into the trial. This approach assumes that at a given dose level, the patient either develops a DLT, has documented tumor response, or neither. The number of patients to be treated per dose level is small, usually one, two, or three patients. At any step in the dose escalation process, if the estimated tumor response is similar for two or more doses (within say 5% of each other), they suggest that the dose with the lowest estimated probability of developing a DLT be chosen for the next cohort of patients to be enrolled. In terms of determining the maximum number of patients to be enrolled, Thall et al. (20) recommend that this be determined from the results of the simulations with clinically plausible dose-response scenarios. Thall and Cook (21) and Zhang et al. (22) presented alternative methods based on a similar approach of modeling both toxicity and efficacy in the phase I trial, extending the methodology to allow for a non-monotone dose-efficacy relationship.

0 0

Post a comment