11.2. Design Considerations
11.2.1. Types of Dose-Ranging Studies
To define dose-ranging studies, we may consider the entire class of studies that involve multiple doses of an active compound. This broad class includes dose-escalation or dose-titration studies (i.e., the dose could go up or down for the same trial subject), parallel group studies (subjects are randomized to distinct dose groups) or cross-over studies involving sequences of various dose administrations. When considering the design of a dose-ranging study, we must first consider patient safety, especially in the early stages of drug development. Regardless of how much in vitro or animal testing has been completed, human subjects must be exposed to a new chemical entity on a gradual basis both in terms of the amount of the drug to be administered as well as the duration of exposure.
11.2.2. Dose-Escalation Studies
The typical dose-ranging trials that are done initially in humans are dose-escalation trials in which there is a single administration of a very small dose of the drug to a small number of subjects—usually four–ten volunteers. Doses are escalated depending on the safety and pharmacokinetic responses of the patients in an effort to explore the boundary of tolerability to the drug. Such designs are often replicated using repeated administration of a fixed dose of the drug for one- or two-week intervals, again to gradually increase exposure and establish tolerable doses of the new compound. It is worth noting that the assignment of subjects to dose groups in dose-escalation trials is non-random. Subjects are randomized within each dose group but not across the groups as in parallel group designs.
We can consider having the same patients progress through the entire dose escalation or having independent groups of volunteers for each step of the way. The choice depends on the nature of the responses and the objectives of the trial:
If the response is one that persists, independent groups will be required to avoid prolonged washout periods.
If the objective is to collect safety information on as many subjects as safely possible, independent groups will be preferred. Of course, using the same patients throughout the entire escalation of doses allows more direct comparisons of responses across doses and thereby more precise estimates of parameters of interest.
Lastly, in dose-escalation studies, we needs to be prepared to deal with incomplete data caused by dropouts, especially when the dropout is due to intolerance.
11.2.3. Cross-Over Design
In the early stages of drug development, but after tolerability has been established as described above, larger studies with greater sophistication can be initiated. An example of a cross-over design that embodies elements of randomization and dose escalation is shown in Table 11.1. A very useful aspect of the design is that, in the first period, two-thirds of the patients receive the lower dose of the drug. As the periods of the study progress, more patients are exposed to the higher dose, and all patients are exposed to the lower dose before the final period.
| Sequence | Period 1 | Period 2 | Period 3 |
|---|---|---|---|
| 1 | Placebo | Lower dose | Higher dose |
| 2 | Lower dose | Placebo | Higher dose |
| 3 | Lower dose | Higher dose | Placebo |
This design is usually appealing to clinicians because of the safety of patients inherent in the dose escalation embedded in the cross-over periods. In addition, the use of placebo in each period helps maintain the double-blind nature of the trial, which is usually desirable, and allows for the separation of treatment and period effects. All of the usual considerations need to be made when considering a cross-over versus parallel design (e.g., within- and between- subject variability, stability of the disease state).
11.2.4. Parallel Group Design
The most common and simple design is the controlled, randomized, parallel dose response study. In this study design, patients are randomly allocated to one of several active dose groups or control, most often placebo. While a placebo control leads to the most clear interpretation of results, active controls can also be used. The parallel design is most popular in Phase II development when larger studies are done in order to explore safety and effectiveness of a new drug. Since the only difference between treatment groups is the dose of the drug, the design leads to more straightforward analysis and interpretation as will be described in subsequent sections.
11.2.5. Factorial Design
In a growing number of situations, researchers want to study the effect of more than one aspect of the dose of a new compound on a disease state. Several examples come to mind: the amount of the dose and the frequency of dosing (i.e., once a day or twice a day) or the size of a bolus injection and the subsequent infusion rate of a compound. Furthermore, in some disease states combination therapy is the norm, or at least a common phenomenon, such as chemotherapy and antihypertensive therapy. It may be of interest to study multiple doses of each drug in combination with each other. Obviously, such explorations of dose response lead to factorial studies. Statisticians have long recognized the value and economy of conducting factorial experiments but their use in clinical trials is less common. This may be due to the complexity (e.g., packaging and blinding clinical medications) of many clinical trials. Such designs may be most relevant if one is interested in finding the optimal dosing regimen for a drug or combination of drugs.
11.2.6. Choice of a Control and Research Hypothesis
Negative and positive controls play a key role in dose-ranging clinical trials. As pointed out in many publications, a significant dose-response trend in the absence of a control group cannot serve as evidence of a drug effect. In studies with a negative control, a significant dose-response relationship can be observed even if the lowest dose is less efficacious than the control and the highest dose is generally comparable to the control.
Most commonly, dose-ranging clinical trials are designed to investigate the effect of multiple doses of an experimental drug compared to a negative control (placebo). Placebo is a "dummy" treatment that appears as identical as possible to the test treatment but does not contain the test drug. As stated in the (amended) Helsinki Declaration and ICH E10 guidance document, the use of a placebo may be justified in studies where no proven therapeutic method exists. Using a placebo as the control may also be justified if the disease does not cause serious harm (e.g., seasonal allergy) or when treatment may be optional (e.g., treatment for pain or depression). However, if the disease causes mortality, the inclusion of a placebo-controlled group may not be ethical. Even if the disease does not cause mortality but causes irreversible morbidity, that is, the disease invariably progresses unless treated (e.g., Type II diabetes), the inclusion of a placebo group may be unethical as well.
11.2.7. Superiority and Non-Inferiority Testing
In comparing an experimental drug against a negative control, we hopes to prove that the new drug is superior to the negative control by a clinically meaningful amount. If μnegative is the mean response of the negative control, and μdrug is the mean of the experimental drug, then the drug is considered efficacious if
where δsup is a prespecified non-negative quantity representing a clinically important treatment difference for establishing superiority.
An active control is a drug which has been proven to be efficacious, typically already approved and in use. In comparing a new (experimental) drug against an active control, clinical researchers may hope to prove that the new drug is superior to the active control. If μactive is the mean of the active control, then the new drug is considered efficacious if
where δsup again represents a clinically important treatment difference.
This is done primarily in the European Union. The rationale for demanding proven superiority before approving a drug is that, by limiting the number of drugs and thus allocating each a large share of market, it might bepossible to better negotiate a cost-effective managed care plan.
Alternatively, clinical researchers may hope to demonstrate that the new drug is non-inferior to the active control, i.e., show that
where δnon-inf is the so-called non-inferiority margin that defines a clinically meaningful difference in non-inferiority trials.
This approach is often taken in the United States. In such cases, δnon-inf may be a fraction (e.g., 50%) of the presumably known improvement the active control provides over the negative control. For example, in the U.S., a new drug for some mild digestive disorder might be approved if it is shown to maintain at least half the treatment effect of another drug on the market. The rationale for approving such a drug is to let the consumers make their own decisions, based on cost, efficacy and side effect considerations.
In dose-ranging trial with a non-inferiority objective, in addition to the new treatment and the active control groups, it is desirable to include a negative control group as well. This is done to ensure assay sensitivity. It is best to proceed to compare the new drug with the active control after establishing that the non-inferiority trial is sensitive enough to detect the known difference between the active and negative controls. Otherwise, clinical researchers may fail to detect a difference between the new treatment and active control (and conclude that the new treatment is non-inferior to the active control) simply due to lack of power. Obviously, if there is a statistically significant difference between the new treatment and the active control, the difference stands whether the active control is found to be different from the placebo or not.
11.2.8. One-Sided and Two-Sided Testing
Of particular relevance when designing a study to assess a dose-response relationship is the matter of one-sided versus two-sided testing. This issue is debated passionately by some, but it is clear that in the vast majority of situations, the researcher has a known interest in the direction of the desired response. This would naturally imply a one-sided alternative with a suitable sample size to carry out the appropriate statistical test. The size of the test also needs to be considered carefully in the design of the trial.
11.2.9. Control of Type I and Type II Error Rates
When taking a hypothesis testing approach, we must consider the importance of Type I and Type II errors. If the response of interest is an efficacy response, then the Type I error is of greatest interest (i.e., we do not want to conclude a drug is effective when it truly is not). However, if the response of interest is a safety variable (e.g., QTc prolongation), the Type II error plays a more important role (i.e., we do not want to conclude there is no effect on safety when in fact there is). Often, this is under-appreciated and in the case of a safety study it may be perfectly acceptable to have a larger than usual Type I error rate such as 0.10 or 0.15 to shrink the Type II error for a limited or fixed sample size that may be needed for a practical clinical trial.