Medical devices vary greatly with respect to their type, their population for intended use, and the significance of risk posed by their use. FDA's device regulations recognize these differences by classifying devices into three categories--Class I, II, and III--with Class III representing the highest level of risk of injury to the patient. For the same reasons, FDA does not automatically require clinical trials as part of the approval process for all devices. While the agency is more likely now than it was in the past to request clinical data in conjunction with a 510(k) premarket notification, requirements for clinical trials apply mostly to the technologically innovative devices that reach the market through the premarket approval (PMA) process.
When FDA does require clinical data as part of a product approval submission, it expects that the trials conducted by the manufacturer will make use of the best scientific methods available to answer the specific questions posed for the device. The sections that follow describe some of the issues that commonly arise in the process of designing, conducting, or evaluating clinical research. Subsequent installments of this series will discuss these issues in greater detail and offer specific guidance for manufacturers seeking to navigate these sometimes treacherous waters.
End Points. Every clinical trial should be designed to accomplish a clear study objective in the form of a medical claim for the product. The objective should be as specific as the state of clinical knowledge for that product allows. Hence, for a device undergoing a first feasibility trial, the objective may be related to demonstrating the plausibility of the device concept. For a device more advanced in the development process, the objective may be to compare patient survival times with that device to patient survival times with another currently approved device or treatment regimen. To ensure that the medical claim is entirely accurate, the manufacturer should also identify the intended patient population for the device, including any subpopulations of particular interest. These last considerations lead naturally to the inclusion and exclusion criteria for the study.
Variables. For every clinical trial, the manufacturer should identify all variables that will be monitored as part of the study. Two broad classes of variables are common. Prognostic or baseline variables are collected prior to treatment intervention, and are used as possible covariants in explaining or interpreting a patient's response following intervention. Outcome variables are the primary end points on which assessment of the safety and efficacy of the device will be based. The measurement scale for outcome variables can be continuous (e.g., height, weight, red blood cell count, survival time) or categorical (where the response is classified into one of a finite number of categories). Whenever feasible, continuous-scale responses are preferable because they can provide more information per patient, and thereby reduce the total sample size required to answer the question of interest. Even with continuous-scale responses, appropriate steps should be taken to minimize the potential for measurement bias and to maximize the precision and accuracy of the measurement methods.
Trial Design. The experimental design used for a clinical trial should be that which is most appropriate for the comparisons being made. The two most common types of design for comparing devices or treatment regimens are parallel group studies in which patients are assigned to only one device or treatment regimen, and crossover studies in which each patient receives more than one device or treatment regimen sequentially during his or her time in the clinical trial. Practical considerations such as the time required to complete the trial or the possibility of carryover effects will often determine which design is most appropriate.
The type of control used for the study may also have a determinant effect on its experimental design. Since the control group usually forms the primary basis for a study's comparisons and inferences about the safety and efficacy of the device, selection of the type of control to be used requires careful consideration. The control most commonly used in device clinical trials is an active control that corresponds to a currently accepted device or treatment regimen; placebo controls are relatively rare in device trials. Another alternative is historical controls, which are sometimes used in situations where the overall treatment regimen has not changed markedly over time. From a scientific standpoint, active or placebo controls where patients are randomly assigned to either the study group or the control group provide the most straightforward interpretation of results. In device trials, however, use of placebo controls is often impossible or unethical; even active controls may be difficult to identify. When there is no proven and acceptable treatment available for a life-threatening condition, for instance, the only acceptable control may be historical knowledge of the condition's progress when untreated.
Sample Size. The written protocol of any clinical trial should include a thorough discussion of the sample size planned for the study. An investigation's target sample size should include sufficient overage to accommodate patients who withdraw from the trial or are lost to follow-up. For studies comparing alternative treatment regimens, the manufacturer should derive the sample size required to detect specified differences between the treatments, and should specify the alpha level (risk of concluding that the treatments are different when they are actually equivalent) and power level (probability of detecting a specified difference between the treatments as a statistically significant difference) used for the study.
Statistical power is of particular importance in clinical trials aimed at demonstrating equivalence to an existing treatment regimen. A recent study in the Journal of the American Medical Association found that only 36% of the published randomized clinical trials reported as equivalent had at least an 80% power for detecting a 50% relative difference in response.4 Sample size has a direct effect on the validity of a study: if the sample size is too small to detect even large differences between the test and control groups, one cannot legitimately claim equivalence. The scientific need for large sample sizes, however, needs to be tempered by the realities of available patient populations.