Randomized controlled trial

"A clinical trial is defined as a prospective scientific experiment that involves human subjects in whom treatment is initiated for the evaluation of a therapeutic intervention. In a randomized controlled clinical trial, each patient is assigned to receive a specific treatment intervention by a chance mechanism." The theory behind these trials is that the value of a treatment will be shown in an objective way, and, though usually unstated, there is an assumption that the results of the trial will be applicable to the care of patients who have the condition that was treated.

Trials of potential treatments, for ethical reasons, tend to involve multiple stages, starting with small safety tests of the drug or other therapy. Once there is evidence of safety, and preliminary clinical trials, the effort moves to a larger scale: large multicentre clinical trials that are randomised, controlled, and double-blind using a control group of patients (i.e., "arm" of the trial) and an experimental arm.

Trials should be large, so that serious adverse events might be detected even when they occur rarely. Multi-centre trials minimise problems that can arise when a single geographical locus has a population that is not fully representative of the global population, and they can minimise the effect of geographical variations in environment and health care delivery. Randomisation (if the study population is large enough) should mean that the study groups are unbiased. A double-blind trial is one in which neither the patient nor the deliverer of the treatment is aware of the nature of the treatment offered to any particular individual, and this avoids bias caused by the expectations of either the doctor or the patient.

Ethical trials exist in a framework. Before any work with human subjects begins, an appropriate independent review board evaluates the proposed experiment, the risks and benefits to subjects, and way in which informed consent will be given to participation. During the trial, an external safety board monitors progress; it has access to the actual, not double-blinded, data. If the experimental arm of the treatment, in the judgment of the board, is significantly more dangerous than existing therapy, the board can stop the trial. If the experimental treatment is strikingly better than the control arm,

Major developments in randomized controlled trials

In 1996, the Consort Statement was published which improved how trials are reported. The statement was revised in 2001. As of 2005, among high impact journals, "CONSORT was mentioned in the instructions of 36 (22%) journals (see bmj.com), more often in general and internal medicine journals (8/15; 53%) than in specialty journals (28/152; 18%)".

In 2004, the International Committee of Medical Journal Editors (ICMJE) announced that all trials starting enrollment after July 1, 2005 must be registered prior to consideration for publication in one of the 12 member journals of the Committee.

Again, clinical trials exist in a framework. When dealing with treatments, adverse effects may be discovered only after the end of the last phase, or in the larger populations to which a drug, approved for general release based on trial data, is used. New Drug Application is intended to detect hazards that the trials are not powerful enough to find.

Control group

"Control", according to the current Declaration of Helsinki ethical guides, may or may not involve a placebo. If there is no accepted treatment, or the disease is mild and self-limiting, placebo controls may be ethical. If there is accepted treatment, the best accepted treatment becomes the control arm. Trials that are controlled, but not by a placebo, still generate arguments about information value versus ethics.

Placebo controls are important, because the placebo effect can often be strong. The more value a subject believes an unknown drug has, but more placebo effect is has. The use of historical rather than concurrent controls may lead to exaggerated estimated of effect.

Placebo effect can be seen in controlled trials of surgical interventions with the control group received a sham procedure.

The Hawthorne effect is the improvements seen in control subjects simply from participating in research.

Variations in design

Cluster-randomized trials

In some settings, health care providers, or healthcare institutions should be randomized rather than randomizing the research subjects. This should occur when the intervention targets the provider or institutions and thus the results from each subject are not truly independent, but will cluster within the health care provider or healthcare institution. Guidelines exist for conducting cluster randomised trials. Cluster-randomized trials are not always correctly designed and executed.

Designing an adequately sized cluster-randomized trial is based on several factors. One factor is the intraclass (intracluster) correlation coefficient (ICC). The ICC between clusters in analogous to the variance between subject in a randomized controlled trial. Just as in Student's t-test for randomized controlled trial more variance between subjects means a larger study is needed, the less correlation between clusters means more clusters are needed.

Before-after studies

Uncontrolled before-after studies and controlled before-after studies probably should not be considered variations of a randomized controlled trial, yet if carefully done offer advantages to observational studies. As in a true cluster-randomized trial, the intervention group can be randomly assigned; however, unlike a cluster-randomized trial, the before-after study does not have enough clusters or groups. An interrupted time series analysis can try to improve plausibility of causation; however, interrupted time series are commonly performed incorrectly.

Crossover trial

In crossover trials, patients start in intervention and controls, but later all patients switch groups.

Factorial design

		Intervention A
		Given	Not given
Intervention B	Given	Group 1	Group 2
	Not given	Group 3	Group 4

Variations on the standard AB, BA design have been proposed.

Factorial design

A factorial design allows two interventions to be be studied with ability to measure the treatment effect of each intervention in isolation and in combination.

n of 1 trial

In a "n of 1" trial, also called single-subject randomized trials, a single patient randomly proceeds through multiple blinded crossover comparisons. This address the concerns that traditional randomized controlled trials may not generalize to a specific patient.

Underlining the difficulty in extrapolating from large trials to individual patients, Sackett proposed the use of N of 1 randomized controlled trials. In these, the patient is both the treatment group and the placebo group, but at different times. Blinding must be done with the collaboration of the pharmacist, and treatment effects must appear and disappear quickly following introduction and cessation of the therapy. This type of trial can be performed for many chronic, stable conditions. The individualized nature of the single-subject randomized trial, and the fact that it often requires the active participation of the patient (questionnaires, diaries), appeals to the patient and promotes better insight and self-management as well as patient safety, in a cost-effective manner.

Noninferiority and equivalence randomized trials

Noninferiority and equivalency randomized controlled trials.

As stated in The Declaration of Helsinki by the World Medical Association it is unethical to give any patient a placebo treatment if an existing treatment option is known to be beneficial. Many scientists and ethicists consider that the U.S. Food and Drug Administration, by demanding placebo-controlled trials, encourages the systematic violation of the Declaration of Helsinki. In addition, the use of placebo controls remains a convenient way to avoid direct comparisons with a competing drug.

The appropriate use of placebo is being revised. When guidelines suggest a placebo is an unethical control, then an "active-control noninferiority trial" may be used. To establish non-inferiority, the following three conditions should be - but frequently are not - established:

1. "The treatment under consideration exhibits therapeutic noninferiority to the active control."
2. "The treatment would exhibit therapeutic efficacy in a placebo-controlled trial if such a trial were to be performed."
3. "The treatment offers ancillary advantages in safety, tolerability, cost, or convenience."

Noninferiority and equivalence randomized trial are difficult to execute well. Guidelines exists for noninferiority and equivalence randomized trials.

Add-on design

"Sometimes a new agent can be assessed by using an 'add-on' study design in which all patients are given standard therapy and are randomly assigned to also receive either new agent or placebo."

Pragmatic trials

Consort has described pragmatic trials and standards for their reporting. CONSORT describes a pragmatic trial as addressing "does the intervention work when used in normal practice" as opposed to an "explanatory trial" that addresses "can the intervention work." CONSORT states the there is not a dichotomy of trial designs, but that pragmatic reflects an attitude in the design. CONSORT offers an example of a pragmatic trial and an explanatory trial.

Ethical issues

The Declaration of Helsinki requires informed consent for participation in a trial. In the United States, there is an approval procedure for clinical trials in human subjects, whether for research only or for potential approval of a commercial drug. Most industrialized countries have such procedures; some permit reciprocal approvals.

Ethics of randomizing subjects

The appropriate use of placebo is being revised. One tension is the balance between using placebo to increase scientific rigor versus the unnessessary deprival of active treatment to patients. This is the conundrum faced by Martin Arrowsmith in the book by the same name.

Comparing a new intervention to a placebo control may not be ethical when an accepted, effective treatment exists. In this case, the new intervention should be compared to the active control to establish whether the standard of care should change. The observation that industry sponsored research may be more likely to conduct trials that have positive results suggest that industry is not picking the most appropriate comparison group. However, it is possible that industry is better at predicting which new innovations are likely to be successful and discontinuing research for less promising interventions before the trial stage.

There are times when placebo control is appropriate even when there is accepted, effective treatment.

There are ethical concerns in comparing a surgical intervention to sham surgery; however, this has been done. Guidelines by the American Medical Association address the use of placebo surgery.

Interim analysis - stopping trials early

Trials are increasingly stopped early; however, this may induce a bias. Data safety and monitoring boards that are independent of the trial are commissioned to conduct interim analyses and make decisions about stopping trials early.

Reasons to stop a trial early are efficacy, safety, and futility.

Regarding efficacy, various rules exist that adjust alpha to decide when to stop a trial early. A commonly recommended rules are the O'Brien-Fleming (the O'Brien-Fleming rule requires a varying p-value depending on the number of interim analyses) and the Haybittle-Peto (the Haybittle-Peto which requires p<0.001 to stop a trial early) rule.

Using a more conservative stopping rule reduces the chance of a statistical alpha (Type I) error; however, these rules do not alter that the effect size may be exaggerated. According to Bassler, "the more stringent the P-value threshold results must cross to justify stopping the trial, the more likely it is that a trial stopped early will overestimate the treatment effect." A review of trials stopped early found that the earlier a trial was stopped the larger was its reported treatment effect. Accordingly, examples exists of trials whose interim analyses were significant, but the trial was continued and the final analysis was less significant or was insignificant. Methods to correct for exaggeration exists.

As an alternative to the alpha rules, conditional power can help decide when to stop trials early.

Seeding trials

Seeding trials are studies sponsored by industry whose "apparent purpose is to test a hypothesis. The true purpose is to get physicians in the habit of prescribing a new drug."

Measuring outcomes

Subjectively assessed outcomes are more susceptible to bias in trials with inadequate allocation concealment.

Missing data

Missing data created decisions regarding what outcome to assign to that patient and what experimental group to assign the patient to.

• Regarding assigning an outcome to the patient, using a 'last observation carried forward' (LOCF) analysis may introduce biases.
• Regarding group assignment, a 'per protocol' analysis may introduce bias compared to an 'intention to treat' analysis.

Surrogate outcomes

The costs and efforts required to measure primary endpoints such as morbidity and mortality make using surrogate outcomes an option. An example is in the treatment of osteoporosis, the primary outcomes are fractures and mortality whereas the surrogate outcome is changes in bone mineral density. Other examples of surrogate outcomes are tumor shrinkage or changes in cholesterol level, blood pressure, HbA1c, CD4 cell count. Surrogate markers might be acceptable when "the surrogate must be a correlate of the true clinical outcome and fully capture the net effect of treatment on the clinical outcome".

Composite outcomes

Composite outcomes are problematic if the components of the composite vary in their magnitude and contribution to the composite.

Overdiagnosis

In trials of mass screening, overdiagnosis is the diagnosis of non-harmful disease. Overdiagnosis inflates the importance of the screening problem.

Analysis

When protocols for trials are publicly accessible, discrepancies between planned and actual analyses may be found.

Dichotomous outcomes may be summarized with relative risk reduction, absolute risk reduction, or the number needed to treat.

Subgroup analyses

Subgroup analyses can be misleading due to failure to prespecify hypotheses and to account for multiple comparisons.

Assessing the quality of a trial

Cochrane bias scale

The Cochrane Collaboration uses a six item tool.

Jadad score

The Jadad score may be used to assess quality and contains three items:

1. Was the study described as randomized (this includes the use of words such as randomly, random, and randomization)?
2. Was the study described as double blind?
3. Was there a description of withdrawals and dropouts?

Each question is scored one point for a yes answer. In addition, for questions and 2, a point is added if the method was appropriate and a point is deducted if the method is not appropriate (e.g. not effectively randomized or not effectively double-blinded).

Publication bias

Publication bias refers to the tendency that trials that show a positive significant effect are more likely to be published than those that show no effect or are inconclusive.

Trial registration

At the same time, in September 2004, the International Committee of Medical Journal Editors (ICMJE) announced that all trials starting enrollment after July 1, 2005 must be registered prior to consideration for publication in one of the 12 member journals of the Committee. This move was to reduce the risk of publication bias as negative trials that are unpublished would be more easily discoverable.

Available trial registries include:

• http://clinicaltrials.gov
• World Health Organization's International Clinical Trial Registry Platform (ICTRP)

External validation

Judging external validity is more difficult than judging internal validity. "External validity refers to the question whether results are generalizable to persons other than the population in the original study." A framework for assessing external validity proposes:

1. "The study population might not be representative for the eligibility criteria that were intended. It should be addressed whether the study population differs from the intended source population with respect to characteristics that influence outcome."
2. "The target population will, by definition, differ from the study population with respect to geographical, temporal and ethnical conditions. Pondering external validity means asking the question whether these differences may influence study results."
3. "It should be assessed whether the study's conclusions can be generalized to target populations that do not meet all the eligibility criteria."

External links

• U.S. National Institutes of Health: Research Involving Human Subjects