David M. Vail, DVM, DACVIM (Oncology)
Clinical trials represent a special kind of cohort study in which interventions are specifically introduced by the investigators in ways to improve the possibility of observing effects that are free of bias. The basic structure and the specific type of clinical trial is defined by modifications of the basic component parts, namely patient selection, treatment allocation, intervention, and outcome measurement. The ultimate goal of any clinical trial is to improve upon the currently available standard-of-care. While this review will only deal with prospective clinical trials, unfortunately a significant proportion of the standard-of-care in veterinary oncology is still based on retrospective data. It is important to state that retrospective studies should only be used to create questions that can be answered in prospective trials and very rarely should the standard-of-care be changed based on the results of a retrospective analysis. Thankfully, more and more pharmaceuticals are being developed specifically for companion animals licensure and several pharmaceutical companies are seeking companion animal licensure of currently available off-label drugs; therefore, our reliance on retrospective data and the anthropomorphic translation of data from human trials is become less and less.
In veterinary medicine, an additional goal of clinical trials is to contribute to and inform human clinical trials; that is, utilize companion animal species with naturally occurring cancers for proof-of-principle or proof-of-concept trials that advance novel therapeutics or novel drug delivery techniques. My discussion in session 83 of these proceedings attest to the model potential of companion animals. Ultimately, it is hoped that advancements made using companion animal species will in the end advance the practice of veterinary oncology.
Careful planning and implementation are critical to the success of any trial and this review seeks to take the reader through the various stages of anticancer drug development, discussing the traditional trial Phases (I-IV) as well as exposing the reader to some alternative trial design modifications that are currently being evaluated or implemented. Several examples from the veterinary literature will be used to illustrate trial design.
Several reviews in the veterinary and human cancer literature have addressed the difficulties in advancing new drugs or therapies in the oncology realm.1-6 It is estimated that only 5-10% of drugs entering Phase I clinical trials ultimately get to market with a cost of between 0.8 and 1.7 billion dollars per drug through development.7,8 There is tremendous need to improve the efficiency and speed of drug development as too many patients and resources are used in one trial at the expense of other treatments, ultimately slowing medical process. The problem is further complicated by insisting on investigating newer molecular-targeted cytostatic agents using trial designs developed for traditional cytotoxic drugs. Indeed it is estimated that more than 40% of drugs currently under development are targeting novel targets. This represents a switch from a primary focus on toxicity to one of identifying a dose that optimally inhibits a specific target;that is, the biologically optimal dose (BOD) may not relate to the maximally tolerated dose (MTD), a dose that is the more traditional starting point for efficacy trials. This also means that availability of validated assays of target modulation are becoming more and more important in the successful implementation of clinical trials for static targeted agents.
Importantly, this review concerns itself with clinical trial design and implementation, and not statistical analysis of generated data it cannot be stressed enough that all trial designs be thoroughly reviewed by a knowledgeable biostatistician to ensure statistical design and power are appropriate prior to implementation.
The Traditional Drug Development Flow
Traditionally, first-in-species trials start with a Phase I dose-finding trial, followed by a Phase II efficacy/activity trial and concluding with a Phase III "comparative" trial that pits the novel agent against or with the current standard of care.
Phase I Trials (Dose-Finding)
Phase I trial design and statistical considerations have been reviewed.2,9,10 The primary goal of Phase I trials is to determine the maximally tolerated dose (MTD) to be used in future Phase II studies, by evaluating safety, tolerance and dose-limiting toxicities (DLT) in treatment cohorts of increasing dose. Activity/efficacy is not a primary goal of Phase I trials. In fact, response rates in Phase I trials are seldom more than 10%. This is particularly important with respect to informed consent as even though clients are informed that they may be receiving a drug with nonexistent activity or at a suboptimal dose in early dosing groups, a full 50% of humans entering Phase I trials believe they will experience a response. Secondary goals of Phase I trials may include scheduling issues, response rate, pharmacokinetic information (ADME = absorption, distribution, metabolism and elimination), and effects on molecular targets or pathways.
Who Enters Phase I Trials?
In human oncology the type of individual who enters a Phase I trial are those that are refractory to standard-of-care. As such, subjects are generally heavily pretreated, have advanced disease and poor performance status (i.e., significantly ill due to significant tumor burden or prior treatment effects). In veterinary medicine the Phase I patient may have failed standard-of-care, or no effective standard-of-care exists, or the standard-of-care is beyond the financial wherewithal of the client. For example, several veterinary trials currently ongoing by the clinical trials group at the University of Wisconsin offer treatment at reduced cost and/or also have financial assets in place that can be used by the client for standard-of-care should the novel test-therapy prove ineffective in their companion animal. This is truly a win-win situation in many respects for our clients.
Setting the Starting Dose
Generally, some preclinical data exists (in other than the target species) and that data is used to inform a starting dose for Phase I. If other species (e.g., rodent) toxicity data exists, 1/3rd the "no observable adverse event level" (NOAEL) or 1/10th the severe toxicity dose in the most sensitive species is used to start with. If normal laboratory dog (usually beagle dogs) data is available, we have found over the years that it is prudent to start at 50% the MTD in beagles, as they appear to be less sensitive to toxicity than tumor-bearing patient dogs. If the starting dose is too low the length of the trials are longer and there is poor utilization of resources and the number of patients exposed to sub-optimal doses is increased. Some patient advocate groups are allowing patients to pick a starting dose based on varying degrees of risk; that is, some patients are willing to risk more toxicity for a higher likelihood of activity.
Dose Escalation Strategies
As with starting dose, escalation strategies greatly affect the number of patients treated at a potential ineffective dose, the length of the trial and the risk of toxicity. The traditional method of escalation uses a 3 + 3 cohort design where dose escalations are made with 3 dogs per dose level and the MTD is set based on the number of patients experiencing a dose-limiting toxicity (DLT). A DLT is defined as > grade III toxicity in any category (except hematologic) according to predefined adverse event categories such as those in the Veterinary Cooperative Oncology Group Common Terminology Criteria for Adverse Events (VCOG-CTCAE v1.0).11 Grade IV is the cut-off most preferred for the DLT for bone marrow suppression in human trials as these events are usually considered manageable and transient. The MTD is defined as the highest dose level in which no more than 1/6 dogs develops a DLT. Traditionally, a fixed-dose modified Fibonacci method of dose-escalation is used where the dose is escalated 100, 67, 50, 40 and then 33% of the previous dose as the cohorts increase. Similar to starting at too low of a dose, If the escalations are too conservative, more patients receive a sub-optimal dose; however, if the escalations are too rapid, more patients are at risk for significant toxicity and the accuracy of the MTD is poor. Alternative "accelerated titration" dose-escalation strategies have been suggested and in the end, it is always a trade-off of risk versus benefit; however, rapid accelerations is less likely to deny efficacious dosing to someone with a fatal disease.
Phase II Trials (Activity/Efficacy Trials)
Several good reviews have outlined phase II trial design.2,12-15 The primary goal of Phase II trials is to, utilizing the MTD established in phase I, identify the clinical or biologic activity in defined patient populations (e.g., tumors of a particular histology or tumors with a particular molecular target) and inform the decision to embark on a larger, pivotal Phase III trial. The traditional Phase II design (phase IIA), the single-arm open label phase II trial is a non-randomized, non-blinded activity assessment of a novel drug/therapeutic modality that lacks a control group, or uses historical controls which are prone to bias (selection, population drift, and stage migration bias). Simplistically, a minimum of 9 patients of the same histology or molecular target are treated with the investigational drug in order to test the null hypothesis of insufficient efficacy. Assuming the likelihood of spontaneous regression is less than 5% and expecting at least a 25% response rate for the agent to be clinically useful, with a p < 0.05 (type I [α] error; false positive) and a power of 0.8 (type II [β] error; false negative), if no responses are observed after 9 cases the study ends. However, if a response is noted in 1 of the cases, then the accrual is increased to 31 to get an accurate response rate. If you expect a less sizable response rate, (e.g., 5-20%), then the initial accrual number must be increased. Several modifications of Phase II trial design will be discussed.
Endpoints of Activity/Efficacy
Since the primary goal of Phase II trials is assessment of activity/efficacy, the endpoints used to evaluate response are critical to the design. With traditional cytotoxic chemotherapeutics, response criteria are fairly straight forward as size and/or volume are used to assess response according to several published methodologies (e.g., RECIST or WHO;).16,17 However, it is readily evident that such criteria may not be appropriate for the newer molecular targeting agents that are more likely to be cytostatic than cytotoxic and result in stabilization of disease rather than measurable regression. In such cases, temporal measures such as progression-free survival (PFS) or time to progression (TTP) would seem more appropriate end-points; however, these often take too long to mature for timely Phase II trials. Alternatively, an adequate compromise could be progression-free rate (PFR) at predetermined time-points. Other endpoints for targeted agents could be a validated surrogate biomarker or measure of a molecular effect such as dephosphorylation of a growth factor receptor, changes in microvascular density, or specific target modulation that is linked to clinical outcome. Secondary endpoints that can be evaluated in Phase II trials are quality of life assessments, comparative cost of therapy, days of hospitalization, toxicity, etc.
Importantly, Phase II trials also serve to expand our knowledge of the cumulative or long-term toxicities of new agents that may not be observed in short-term phase I trials designed only to elucidate acute toxicity. An example of this in the veterinary literature involved a combined phase I/II trial simultaneously investigating the safety of liposome-encapsulated doxorubicin (LED) in cats while comparing its activity to native doxorubicin in cats with vaccine-associated sarcomas.18 Unexpectedly, the MTD established for LED in the acute Phase I component of the trial was found to result in delayed and dose-limiting nephrotoxicity following long-term follow-up in the Phase II component of the trial.
Controlled Phase II Trials
Sometimes referred to as Phase IIB trials, these tend to be controlled, blinded and randomized investigations of 2 or more novel regimens that identify promising agents to send to Phase III for additional evaluation. Regarding the ethics of control groups, placebo controls are generally not used in human clinical trials in the USA. Placebo controls or historical controls have been used in veterinary clinical trials if no standard-of-care exists or if the historical outcome for a particular tumor type is very well documented and consistent. Examples of historical control trials in the veterinary literature often include cancer histologies with rapid and likely terminal progression (e.g., advanced state oral melanoma, stage II hemangiosarcoma). Randomized Phase II trials can be as simple as randomizing standard-of-care plus or minus the addition of a new drug. More complicated trials can randomize subjects into multiple treatment arms or schedules with only enough power to make "inferences" as to which is the best drug to take forward into Phase III; so called "pick-the-winner" trials. While they often don't have enough power for direct comparison like a Phase III trial would have, they may use a less rigorous statistical assessment, such as setting the p-value at 10% and using 1-tailed analyses. An example of a randomized Phase II trial in veterinary medicine was mentioned above and involved a randomized comparison of activity between liposome-encapsulated doxorubicin and doxorubicin; neither agent had previously been the subject of a Phase II activity trial in cats with vaccine-associated sarcoma.18 A "winner" was not picked with respect to response activity in the macroscopic setting (44% response rate with LED versus 33% with doxorubicin; p = 0.722), nor based on the temporal measurement of disease-free interval in the microscopic disease setting. A clear winner (doxorubicin) was picked, however, when the secondary end-points of cost and safety were factored in. Further examples of modified Phase II trials including seamless Phase II/III trials, randomized discontinuation trials, Bayesian continuous reassessment designs and combinations will be discussed in a subsequent section.
Phase III Trials (Comparative/Confirming Trials)
It has been suggested that if Phase II trials are "learning" trials, Phase III trials are "confirming" trials.5,15,19 These larger, randomized, blinded controlled trials have the goal of comparing a new drug or combination to therapy regarded as the standard-of-care. They are often performed by large cooperative groups that ensure greater case accrual. They are not common in veterinary medicine due to their size and expense. One example involving the multicenter approach would be the randomized comparison of liposome-encapsulated cisplatin (SPI-77) versus standard-of-care carboplatin in dogs with appendicular osteosarcoma.20 No difference was observed between treatment groups and SPI-77 did not show an activity advantage, despite allowing 5 times the MTD of native cisplatin to be delivered in a liposome-encapsulated form.
Phase IV Trials (Post-Market Trials)
Once a drug has been granted a license for a specific label use by the appropriate regulatory body (e.g., federal drug administration [FDA]) post-market Phase IV trials may be performed to gain more information on adverse events, safety, long-term risks and benefits. Essentially, Phase IV's investigate the drug more widely than in the clinical trials used for licensure. They often involve treatment of special populations (e.g., old and the very young, or individuals with renal or hepatic dysfunction). The body of data on PK generated from licensure trials is used to inform decisions on dose in these special populations.
Modifications/Alternatives to Standard Clinical Trial Designs
Comments on Randomization
Randomization is the assignment of subjects into treatment groups based on chance-governed mechanisms such as the flip of a coin, roll of the dice, randomization tables or computer programs. That is, each subject has an equal chance of being assigned to one treatment or the other. This is done to equally distribute known, and in particular, unknown factors that may affect outcome and to limit conscious and unconscious bias. If the sample size is large enough, randomization is usually successful.
Regarding factors that are well known to effect outcome, a more specific form of randomization, stratification, can be used to ensure an equal distribution of patients with these factors within the treatment groups. The classic example in veterinary cancer trials would be stratification by immunophenotype in dogs with NHL to ensure equal numbers of the poorly responding T cell immunophenotype in the various treatment arms.
Unbalanced randomization schemes have also been proposed that allocate more subjects into one treatment group in an attempt to enhance the ethical palatability of some trials or to decrease the cost of trials; however, unbalanced randomization remains controversial.
The timing of randomization is also important. It is usually best to randomize as late as possible, because once randomization has occurred, subsequent analysis should be reported on an Intention-to-treat basis rather than on a treatment-received basis.21,22 That is, all subjects should be included in the analysis regardless of whether they received all the prescribed treatments or not. This will minimize bias based on temporal issues or treatment toxicity. An example to illustrate this in veterinary oncology involves a presentation some years ago at an annual meeting that compared dogs with osteosarcoma receiving 4 doses of cisplatin following amputation to dogs receiving 2 doses of cisplatin. The conclusion, following treatment-received analysis was that dogs receiving 4 doses lived longer. However, because dogs that were scheduled to receive 4 treatments but did not due to early metastasis were excluded in the analysis, the 4-treatment group data was biased to a positive result as those dogs destined to metastasize early were removed, no matter how many treatments they would have received. That being said, It is acceptable to present both treatment-received and intention-to-treat analysis in trial reports in order to discuss how loss to follow-up or treatment withdrawal may effect the conclusions. There are few situations where randomized patients can be removed from analysis; such as when ineligible patients are mistakenly randomized or those that were prematurely randomized but did not receive any intervention.
Note: Continuation of clinical trial design and implementation follows in "Clinical Trial Implementation in Private Practice: A Continuation ".
All references are grouped at the end of the continuation session in "Clinical Trial Implementation in Private Practice: A Continuation".