Clinical trials are experimental studies performed mainly on humans, but sometimes on animals, tissues, and cultures as well, to assess the effectiveness and safety of an intervention under investigation such as a new drug, diagnostic procedure, surgical procedure, prophylactic procedure, or screening program. There are many types of clinical trials. This entry focuses on randomized controlled clinical trials and randomized crossover designs for studying treatment effects. Rigorously designed clinical trials such as randomized concurrently controlled clinical trials conducted with human subjects, usually patients, have become well established as the scientific method based on empirical evidence that investigators must use to assess new treatments if their claims are to find widespread acceptance. Clinical trials have become indispensable in discovering new techniques to prevent and treat diseases, and their applications have been largely responsible for the compression of morbidity and decline in mortality rates at advanced ages in recent years.
In experimental studies, the investigator manipulates the study factor (exposure groups) and randomly allocates experimental units to different exposure groups. An experimental study is also called a trial. A trial must progress longitudinally in time from exposure to outcome. This together with the ability to manipulate the study factor and to randomize the experimental units makes a stronger causal inference using experimental studies than using quasi-experimental studies, which involve manipulation of the study factor but not random allocation of experimental units, or using observational studies, which involve neither manipulation of the study factor nor randomization of experimental units. Experimental studies without control or comparison groups (e.g., Phase I/II clinical trials) are called uncontrolled trials, while experimental studies with control or comparison groups are called controlled trials. A controlled trial in which the allocation of experimental units to different exposure groups is done randomly so that all experimental units have an equal chance of being allocated to each of the exposure groups is known as a randomized controlled trial (RCT), which can be a randomized controlled clinical trial (when the unit of randomization is a patient such as the Phase III clinical trial); a randomized controlled field trial (when the unit of randomization is a normal individual, rather than a patient, such as the vitamin C trial of Karlowski et al., 1975); or a randomized controlled community (cluster—a group of people in a community) trial (when the unit of randomization is a community or cluster such as Project Burn Prevention trial of Mackay and Rothman and water fluoridation trial of Ast and Schlesinger (1956). The less common field trials and community (cluster) trials are aimed at the evaluation of primary preventives, while the more common clinical trials are used to evaluate treatment effectiveness of a disease or to find a preventive of disease recurrence or death. All randomized trials are controlled. In nonrandomized controlled trials, the groups may not be comparable due to selection and confounding biases and so estimates of effects of the intervention may not be valid without further statistical adjustments.
Researchers for each clinical trial follow a protocol reviewed and approved by an institutional review board (IRB), a separate board of scientists, statisticians, physicians, and nurses who are not associated with the clinical trial. A Clinical Trial Protocol contains a study plan that describes the organization of a clinical trial, the background, rationale, objectives, and hypotheses; how the subjects are to be selected and how data are to be collected; primary exposures and methods of their measurements (an exposure is a factor that either causes, prevents, or treats an outcome), outcomes of interest and methods of their measurements; as well as intervening variables and methods of their measurements, type of study design, method of randomization, methods to control confounding bias prior to data analysis if randomization is not used, measures of association to be used, statistical methodology and analysis, including methods to control confounding bias during data analysis, and power calculations. It may also address issues of noncompliance, dropout as well as selection and information biases, and how nondifferential misclassification may affect the interpretation of results. The protocol also states the number of participants, eligibility requirements, agents that will be used, dosages, schedule of tests, the length of the study, and the larger population to which the results can be generalized. Note that both issues of internal validity and external validity have been covered in the description of clinical trial protocol given above. This protocol will also serve as the basis for writing the final report.
Essentially, a randomized controlled clinical trial is conducted by recruiting a group of patients from a target population. With adequate allocation concealment to protect against selection bias, the consenting eligible patients from the group of recruited patients are then randomly allocated to the treatment and control arms, which are then followed to the end of the trial with outcomes between the different arms compared. The patients recruited from the target population must meet the required eligibility criteria (inclusion and exclusion criteria) and, for ethical reasons, must give their informed consent before their randomization into different treatment groups to avoid selection and confounding biases. This ensures that the difference in treatment groups is caused by the difference in treatments alone. A concurrent control arm is needed so that outcomes with and without treatment(s) can be compared.
The choice of the control group will have an impact on research question and sample size. Use of a placebo in the control arm would help achieve blind treatment allocation and exclude placebo or Hawthorne effects, but ethical considerations demand that the control group should be the established treatment for the disease under study provided that its therapeutic effects have been well documented. Blinding is also used to exclude detection and performance biases, which occur when the investigator/patients know the treatment being given and which could affect assignment, assessment, or compliance. These biases occur particularly when a subjective outcome variable such as pain or quality of life is measured.
Noncompliance—the failure to follow protocol requirements—can (1) result in a smaller difference between the treatment and control arms than truly exists, thereby diluting the real impact of a treatment, and (2) reduce study power, making it harder to detect an effect when it exists. Compliance measures can be used to improve estimates of treatment effects. The real strength of a randomized controlled clinical trial lies in the randomization. With sufficient sample size, randomization in which each treatment group is equally likely to be allocated to each patient would produce close similarity across the groups in all respects other than the intervention, including the unmeasured and unknown factors. That is to say, by randomization, both known and unknown confounders are controlled at the outset. This strengthens the validity of the causal inference. To preserve the baseline comparability, to maintain the statistical power of the original study population, and to ensure unbiasedness against noncompliance, intent-to-treat analysis should be conducted. Such analysis, also known as treatment assignment analysis, gives information on the effectiveness of the treatment under real-life conditions, unlike efficacy analysis, which determines the treatment effects under ideal conditions. In an intent-to-treat analysis, all participants who are randomized to a treatment are analyzed, regardless of whether they complete or even receive the treatment. Given the covariate measures at baseline and during follow-up monitoring, the statistical methods for analyzing the treatment effects of randomized controlled clinical trials depend on what endpoints are chosen to measure outcomes. For example, if disease incidence or death or occurrence of some event is the endpoint, logistic or Poisson regression models may be used to analyze the data; such analysis would also serve to control the residual confounding during the analysis. If survival time is the endpoint, then the Cox regression models as well as counting processes martingale methods may be used. If the endpoint consists of continuous response measures, then simple two-sample t tests or, when several treatments are involved, analysis of factorial experiments may be used. If repeated measures are involved, then methods for longitudinal data analysis such as random coefficient analysis may be used.
Most clinical trials belong to the so-called parallel group designs in which subjects in different arms are followed in parallel. Parallel group designs have four phases. Phase I tests a new drug or treatment in a small group of people (usually less than 10 normal volunteers), using an uncontrolled trial, the purpose being to learn how to administer a treatment safely and to determine optimal dosage or the so-called maximally tolerated dose (MTD) based on dose escalation investigations. Phase II expands the study to a larger group of people (often 30 to 40 patients or normal volunteers), the purpose being to test patient responses, to monitor side effects, and to determine the minimum effective dose (MED). These may be either uncontrolled trials or RCTs with patients who are expected to benefit from the treatment as experimental units. Phase III expands the study to an even larger group of people (usually running from hundreds to thousands of patients) for a full-scale evaluation of treatment using a randomized controlled clinical trial, as described above, with control group being either placebo or standard treatment. Patients are also closely monitored for severe adverse side effects for possible cancellation of the trial. Phase IV takes place after the drug or treatment has been licensed and marketed—postmarketing surveillance. It typically compares two treatments that are approved for similar uses to determine which one is more effective. It may also be conducted to study long-term safety and efficacy and to study new uses or costeffectiveness of FDA-approved treatment.
In the design of RCTs, the opposite of parallel group designs are the crossover designs, which involve a switch of study treatments for each participant in a clinical trial. Thus, the participants serve as their own controls, which allow more precise estimates of treatment effects by making comparison within subjects rather than between subjects. Most crossover studies are planned crossover studies in which the predetermined period of treatment before switching to another treatment is specified in advance in the protocol. Noncurable medical conditions such as asthma and diabetes are suitable candidates for planned crossover studies as they present the possibility of giving more than one treatment to each patient. Other crossovers are unplanned such as when patients in medical care are switched to surgical treatment because of deterioration in their condition. In planned crossover studies, there are as many groups as there are permutations of treatment sequences so that each group is uniquely determined by its first treatment. Each participant is randomized to receive his or her first treatment so as to produce roughly equal numbers of participants in each group. They are then followed for a predetermined period of time and then switched to a different treatment for another predetermined period of time and so on. Each period is of the same length for all participants, with the crossover point being blinded where possible. At the end of the study, outcomes are compared across treatments by combining the responses to each treatment from different groups. Thus, in parallel group designs, each group receives just one treatment and the treatments are administered concurrently, while in crossover designs, each group receives all treatments one after another and the treatment order differs across the groups. To avoid the so-called carryover effects in planned crossover studies, a washout period may intervene between treatments to allow the human body time to metabolize and excrete the previous treatment. The double-blind, two-treatment, two-period crossover designs are particularly popular in clinical pharmacology. These can be analyzed as special cases of repeated measure analysis of variance. The hypothesis of no treatment difference is tested simply by using a twosample t test to compare the two sets of within-patient difference summed over all individuals. The main advantage of this design is the gain in statistical power, namely, that it can achieve statistically significant results with fewer subjects than would be required with a parallel design because the sample size required for this design depends on the variability within subjects, not between subjects as in the case of parallel design. (For most response variables, the within-subject variance is smaller than the between-subject variance.) Note that in the statistical analysis of treatment differences of crossover studies we assume no treatmentperiod interaction (which implies no carryover effects).
Clinical trials are costly and time-consuming. The idea of stopping such trials early when the treatments are being found to be unsafe or ineffective has been pursued on ethical and economic grounds. Wald’s sequential procedure has been used to achieve this aim.
Sequential trials in which the data are analyzed after each patient’s results become available, and the trial continues until a clear benefit is seen in one of the comparison groups, or it becomes clear that no significant difference between the groups will emerge serve to achieve this aim. Thus, unlike parallel group and crossover designs, the number of subjects studied is not fixed in advance in sequential trials. Flexible sequential trials based on the discretization of Wiener processes and the group sequential designs implemented with a boundary for the null hypothesis (futility boundary) and a boundary for the alternative hypothesis (efficacy boundary) allow early stopping favor in either of the null or of the alternative as soon as conclusive interim evidence of being efficacious or not became available. Sequential trials and development of monitoring strategy that allows for interim looks at the accumulating data so as to shorten the average length of the trial are discussed in the articles by Jennison and Turnbull, Lan and DeMets, Pampallona and Tsiatis, Pocock, O’Brien and Fleming, and Whitehead listed in the further readings below.
Analytic and Descriptive Epidemiology; Bias; Community Trial; Randomization; Study Design
Related Credo Articles
A clinical trial is a prospective controlled experiment with patients. There are many types of clinical trials, ranging from studies to prevent,...
(See EVIDENCE-BASED MEDICINE .) Clinical trials aim to evaluate the relative effects of alternative health-care interventions. They are...
Clinical trials are a form of medical research involving human participants. What follows is a brief overview about what clinical trials are and...