Randomization

Randomization is a method in which the test subjects (for example, participating patients) using a random mechanism different groups are assigned. This is intended to known and unknown persons borne disturbance variables are distributed evenly across the experimental and control groups. It should equal conditions and test conditions are created in the means to achieve a higher confidence level in spite of the interference. The method finds application in, among other psychological experiments ( field and laboratory experiments). By the method of alternative explanations are eliminated, and the probability can be reduced that the proven effectiveness in a test subject to effect of a systematic bias. Randomization is a prerequisite for blinding.

• 3.1 Assessment Factors
• 3.2 Quasi - randomization
• 3.3 Simple, non-restricted randomization
• 3.4 block randomization, Balanced randomization
• 3.5 Adaptive Randomization
• 3.6 Choice of procedure

• 5.1 Examples

Generally

RA Fisher developed the principle of " randomization" as a consequence of " ceteris paribus " clause. The experimental treatment conditions are assigned to experimental groups, these, in turn, the subjects randomly ( " randomized "). This pseudo-explanations are ruled out, according to which, for example, a behavior is referred to as an effect of the experimental treatment, which has actually already passed präexperimentell - not the new teaching method has led to better results, even before the investigation, the subjects of the experimental group had a learning advantage. The degree to which it is actually randomized, is a feature to differentiate the types of the experiment.

Evidence of effectiveness

Clinical trials are conducted to generalize considering the effectiveness of a treatment in the sample, a statement of the study results to the population. Need prior to the study for the test plan various claims are made, such as proof of efficacy.

Statistical aspect

Based on the data obtained from the study of efficacy test (eg significance test ) performed to determine an effect. Now, if an effect is found, this may have occurred by chance, with the reality match (ie, to find themselves in the population ), or have been affected by a systematic bias.

A systematic bias can be, for example, if a treating investigator copes particularly well with a treatment method. Thus, the effect will be biased because this relationship is not representative. The physician is thus a determinant of the effect. Another factor may be the age of a patient. If for treatment A randomly included in the study only patients who exceed a certain age and below for the B a certain age, this may influence the effect significantly. The following are the types of systematic bias:

• Selection bias: The next treatment allocation can be determined and influences the effect. Improvement: unpredictable randomization
• Accidential bias: It is not considered the Balanced awareness regarding the various influencing factors. Factors may be prognostic factors or risk factors. Improvement: Stratifizierendes method.
• Information bias: The knowledge of the mechanism used affects the outcome indirectly, as stakeholders (patient, investigator, statisticians ) tend by the information to which to prefer better treatment for her feelings. Improvement: blinding of all those involved.
• Concealment bias: The knowledge of the procedure to wendenen prior to treatment by the investigator can influence the result of the treatment itself. Improvement: blinding of the investigator.

Therefore, in order to validate that a test with the proven effect ( most likely) would also be found in the population, all factors that could cause a systematic bias to be considered. There are different randomization. In a randomization a randomization list may already be present before the study (block randomization, stratified randomization) or be based on properties of the already randomized sampling. The latter is thus dynamically calculated during the study (adaptive randomization, stratified randomization).

Organizational aspect

Even if the type of randomization would be found ideal statistical point of view, there may be a systematic bias. In a block randomization often Randomisierungsboxen for each participating center will be created. In such a box are then sequentially numbered envelopes, which must be opaque sealed so that they see not, and can thereby influence, for example, the patient choice.

There are two methods of randomization:

• Central randomization assignments via the Internet or by telephone to each center.
• Decentralized randomization allocation with the help of the mentioned Randomisierungsboxen.

Randomization

To perform a randomization there are different procedures, the patient's treatments are assigned as.

Evaluation factors

The evaluation of a method based on the following factors:

• Predictability: A treating investigator may not predict the next allocation of a type of treatment (otherwise it could affect the result ).
• Poised incident: Bzgl. various factors such as the treatment itself should be made an equally strong assignment. To already be able to perform an interim analysis during the study should be possible at any time the Balanced awareness.

Quasi - randomization

In a quasi- randomized allocation is determined by a mechanism that is not due to real randomness, for example, by an alternating allocation. The all- decisive arrival of a patient in the center is not a valid random mechanism. The Balanced awareness is guaranteed, however, easy to see the next allocation.

Simple, non-restricted randomization

The allocation of treatment to a new patient is random with no restrictions. In two treatments so this corresponds to a coin flip for each patient, without the awareness Poised to respect ( this is statistically guaranteed only from about 1000 samples). Accordingly, a low predictability is provided.

Block randomization, Balanced randomization

To be compared to the simple randomization can ensure better Poised awareness will ensure that is the case with N patients a certain predefined ratio between allocated treatment (eg, 1-1 at two types of treatment ). The N entrapped ( and unknown ), patients are divided into blocks prior to the study. In each block, then the ratio of the permuted types of treatment is considered. The randomization list then consists of the compositions of the individual blocks

The simplest version consists of only one block. The Balanced awareness is ensured only end of the study, but the predictability is low.

• Permuted blocks of equal length: When using multiple blocks generated it is placed within the randomization list at random. By the use of a plurality of blocks of equal length greater Balanced awareness over the entire recruitment period is given at the expense of predictability. All blocks have a length of 4, it may be determined by an observer. With knowledge of the block length, it is then easy to determine in the use of the next block which treatment assignment is made. For example, if the first two assignments already laid down the type of treatment A, then the allocation of B must be done twice.
• Permuted blocks of variable length: thus the length of the blocks can not be detected and thus be no light treatment predictability allocation permuted blocks of variable length may be used. That is, it generates blocks of different lengths whose allocation provisions contained are determined randomly and happen to be ranked within the randomization list together.
• Stratified randomization: The methods described so far consider only those Poised awareness regarding the type of treatment. But in order to make a Balanced awareness regarding the influence factors, a randomization list for each combination of factors can be created ( at baseline ). That is, with two factors ( each with two forms ) there were four lists for each type of treatment and thus eight lists with two types of treatment. The influence factors and the treatment itself are referred to as strata. Each allocation sequence may be calculated by any of the above methods. The Stratified randomization should (under 50 samples) and multicenter studies are used in small studies. For the implementation of a central randomization is necessary.

Adaptive randomization

In the adaptive allocating randomization of the treatment with a probability that depends on the distribution of the previous treatments carried out in the individual strata. The data of the already randomized sampling so will be considered. Randomization is thus dynamically during the study.

• Baseline adaptive randomization: This adaptive randomization, the number of previous allocations into account, and thus affects the probability of the next allocation. The treatment was equally often assigned to the probability of the method is the same. If a method is less often allocated, the allocation to the current patient is higher. Both for the calculation of the probability and the decision variable (instead of the difference of the numbers of allocation ), there are more complex methods. An example of this is the biased coin method. This allows the Balanced awareness can be improved. The predictability is limited by the probability of assignment. This method can be applied only if no further Strata must be included. Very important for open trials ( with no blinding ) so that predictability is minimized.
• Minimization method: The minimization method is a method for Stratifying the baseline adaptive randomization. For a patient with fixed characteristics of each stratum, the sum of the respective characteristics for the types of treatment is calculated. If the patient is in the center 1, male and over 50 years old, so A is used for treatment to determine how many treatments were carried out in the center of A 1, how many were male and how many over 50 years. Analog is performed this calculation for Treatment B. The patient then receives that type of treatment ( with a higher probability ) is assigned, having the smaller sum. So therefore a Balanced awareness is based on all strata considered. The good Poised awareness contrasts with the also good predictability.
• Response Adaptive Randomization: in contrast to the baseline adaptive randomization, allocation by the already measured Strata is uniquely determined. No chance, depending on the strata so it is defined.
• Play -the- winner: Successful treatment with a method causes them is also applied the next patient. A change thus takes place only when the application was not successful. The bad Poised awareness and good predictability is compared to that the better method was probably used more often ( Patientenbenefit ).

Choice of procedure

In large studies, a block randomization can be selected, if it is not a multi-center study. Then a stratified randomization is certainly necessary. If several Strata present an adaptive method should be selected, otherwise a lot of randomisation should be managed (product of Strata - forms ). Overall, however, the number of strata should be kept low.

Blinding

A distinction is made blind, double-blind, triple blind studies. In a blind study, only the patient does not know which treatment option he gets. In addition, it is important that the physician does not know which patient is being treated with said drug. This method is referred to as a double -blind study. To preserve the objective and in the data analysis, and this can be done without regard to the each treatment were carried out, then there is a triple -blind study.

Emergencies - for example, severe side effects - may, however, in individual subjects, the premature disclosure of the assignment to the study groups make it necessary (so-called unblinding ).

Treatment diffusion

As Treatment diffusion blurring the boundary between the control and the experimental group is referred to in the experimental design of a psychological experiment. The Treatment - the treatment, their effectiveness will be shown - is applied not only to the experimental group, but also partly unintentionally to the control group. Thus, in the analysis, the resultant effect will not be clearly attributed to the treatment and are thus attributed to a defective randomization.

Examples

• A control group is treated by the usual method of psychotherapy, the experimental group, however, by a new method. Both groups are managed by the same psychotherapist. Since he is convinced of the new (yet to be tested ) method, he turns unconsciously to some procedure of the new method in the control group.
• In a department of orthopedics clinic, patients are divided into a control and an experimental group. The control group receives the usual therapy, the experimental group is instructed to independently additional exercises (eg muscle groups to tense and consciously loosen again) several times to make a day. Patients talk to each other and exchange information on the forms of therapy, after which some patients in the control group to start doing these exercises because they expect a better treatment outcome.