Implementation Of Randomization

In the pharmaceutical industry, for good clinical practice a set of standard operating procedures (SOP) for generation, implementation, and administration of randomization is usually established to ensure the integrity of clinical trials. In this section we will introduce an implementation procedure for the method of nonadaptive randomizations which is adopted by most of pharmaceutical companies for clinical research and development. In the pharmaceutical industry the department of Biostatistics and Data Management (or Biometrics) is usually responsible for the activities of statistics, programming, and clinical data management. Within the department, a drug-specific (or project-specific) team (or unit) is usually formed to oversee the development of statistics, programming, and data management during the process. This team usually consists of biostatisticians, programmers, and data coordinators. Note that unlike a clinical research associate or monitor, this team does not involve itself with the day-to-day activities of clinical projects. However, this team is responsible for the selection of randomization methods, case report forms design and review, clinical data management, statistical analysis, and report writing. Because generation of randomization codes is the key to the success of the intended trials, a group within the department of Bio-statistics and Data Management, not involved with clinical trials, is designated to be responsible for generation and management of randomization codes. Since there are many different methods for generating randomization codes as discussed above, the group is responsible for the implementation of a system that incorporates the various methods of randomization by developing computer programs. Since the system, which may contain a number of computer programs, is not designed for commercial but for internal use, it is recommended that the methods of randomization employed and the corresponding computer programs be adequately documented. Also it is desirable for a user-friendly User's Reference Manual to be developed. The User's Reference Manual should contain detailed instructions for the use of the system, references to the pseudonumber generator, methods of randomization, programs for listings of the pseudonumber generator and for the production of a listing of the randomization codes. In addition a prospective validation of the design programs should be performed according to a validation protocol before the implementation of the system. Note that the FDA requires that the results of the validation test be documented and that the system be validated periodically.

Generation, Labeling, and Packaging

During the development of the clinical protocol, the project clinician and biostatistician usually discuss the selection of an appropriate method of randomization and some related logistic issues for the implementation of randomization according to study objectives, primary endpoints, stratified covariates (if any), and sample size of the trial. The randomization method employed for the study should be described in detail in the study protocol without disclosure of the block size, if the permuted-block randomization is used. The study protocol and the investigators's brochure should also describe in detail a standard procedure for treatment assignment and drug dispensing. In general, patients should not receive any medication unless they have met all eligibility criteria and have signed the informed consent forms as defined in the study protocol. A formal request for randomization codes cannot be sent to the project statistician unless the study protocol has obtained an approval from an internal protocol review committee. The project statistician can then check whether the request is adequate with respect to the study protocol and design. The randomization codes will be generated according to the selected method of randomization by the randomization group if no concerns are raised by the project statistician. The group is not only responsible for the generation of randomization codes but also for performing quality assurance (QA) procedures of the generated randomization codes. The QA procedures are to (1) check every generated randomization codes, (2) document the program logs for generation of the randomization codes, and (3) maintain information for the generation of the randomization codes including the seed and the first and last random numbers generated from the seed. If the randomization codes meet the requirements of the QA procedures, then a list of randomization codes is sent to the drug packaging department or some contracted laboratory for packaging the study drugs. If the trial is a triple-blind study, the project statistician and clinician or other project team members should be informed only of the generation of the randomization codes by a cover memo. If, however, the trial is a double-blind study, then the project statistician and clinician might get a copy of randomization codes upon request. The information of the randomization codes will then be locked in the database until the time at which an interim analysis or final analysis is performed. For a triple-blind study the clinical data coordinator and clinical research associate identify the patients through a sequentially assigned patient (subject) number to maintain the blindness. A patient or subject number usually contains three parts, which include the project number, the study center number, and a sequentially assigned patient number within the individual study center. Let us take, for example, a number 01401015 (i.e., 014-01-015) used to identify a patient in a clinical trial. The first three digits 014 are an identifier of the study drug XXX, the next two digits 01 represent the first study site, and the last three digits 015 indicate that the patient is the 15 th patient to enroll in the study. The project team will also generate a set of dummy randomization codes for the project statistician to perform necessary programming for patient listings or case report tabulations as required by the FDA to shorten the statistical analysis after the study is completed and the database is locked.

When the drug packaging department receives the randomization codes, the study drugs are packed according to the method and instruction as stated in the protocol. The most secure method for maintaining blindness is to use identical blister packs or drug kits with identically appearing contents. Usually the drug kits have a three-part double-blind tear-off label affixed to the cover of the kit. This label has the protocol number and the preprinted patient number. Patients' initials and the time and date the drug dispensed will be recorded on each label. The time and date are important for establishing an audit trail of treatment assignment. The double-blinded tear-off portion, which will be attached to the appropriate page in the case repart form, contains the actual treatment group information to which the patient is assigned. These sealed labels will not be opened unless it is required in a medical emergency when knowledge of the respective treatment may influence medical care. At the conclusion of the study, the investigators should return all used and unused study drugs to the sponsor. Usually there is a broilerplate paragraph included in the study protocol for drug accountability. Figures 4.4.1 and 4.4.2 provide flow charts of the randomization procedure discussed above.

Random Assignment

In the pharmaceutical industry, the randomization procedure is not limited to the generation of randomization codes for treatment assignments. It can also be applied to laboratory evaluations. For example, routine hematology, blood chemistry, urinalysis, or some other

Figure 4.4.1 Randomization codes at the planning stage.

special compounds such a serum hormone levels of the patients from a clinical trial are usually assayed at a centralized contracted laboratory. However, the assay of active ingredients of the study drugs are often performed by the method developed by the laboratory. Since samples from a clinical trial may be enormous due to multiple visits, assays may be required to perform at different times of a day over a period of several days due to the capacity of the laboratory. As a result, a proper design for the drug assay is necessary to eliminate variability due to analyst, time, and day. In addition the assay should be performed in a blinded fashion to avoid possible bias caused by the knowledge of the study drugs. Therefore randomization codes may also be generated by the same randomization group according to the design and method of randomization as deemed appropriate by the project statistician. The randomization codes for treatment assignments and drug assays are to be stored in the central file and cannot be released until the database is locked. The generation and implementation of randomization codes for drug assays can be logistically complicated when assays for different active ingredients or their metabolites are required to perform individually with subdivisions of the blood samples.

In some cases, randomization is also used for poststudy evaluation. For example, for the evaluation of contrast agents in the enhancement of images obtained with magnetic

Figure 4.4.2 Generation of randomization codes.

resonance imaging (MRI), two sets of films with and without contrast agent are usually obtained. The set of films without the contrast agent is obtained before that with the contrast agent. In general, this type of trial is conducted in an open-label fashion without a concurrent control because it is almost impossible to maintain the blindness during the trial. As a result the FDA requests that the sponsors perform a blinded reader study after the trial is completed. A separate protocol for the blinded reader study is also prepared after all films are obtained from the clinical trial. The blinded reader studies are considered as adequate well-controlled studies for approval. The package insert is usually derived from the blinded reader studies rather than the actual clinical trials. For blinded reader studies, several qualified readers, who are not engaged the clinical part of the trial and are not associated with any investigators of the trial, are asked in a random but blinded fashion to evaluate the films. For this purpose, randomization codes for a blinded reader study can also be generated according to the design and randomization method specified in the protocol under the supervision of the project team.

Note that the randomization procedure described above is probably the most frequently employed procedure for conducting clinical trials in the pharmaceutical industry with sample sizes smaller than a few thousands. For most clinical trials sponsored by the NIH or other cooperative groups, the sample size can be quite large. For example, the ISIS-2 (1988) study randomized 17,187 patients with suspected acute myocardial infarction to four treatment groups and 22,071 male physicians were enrolled to receive one of the four treatments in the U.S. Physician's Health Study (1989). An even larger study is the GUSTO (1993) study which enrolled a total of 41,021 patients with evolving myocardial infarction. As a result, it may not be feasible to adopt the randomization procedure described above. As an alternative, a centralized randomization center may be established for random assignment of treatments either by mail or by telephone. If the time between the screening and request for random assignment is long, say a month, then the mailing system may be possible. For example, see the study conducted by the Coronary Drug Project Research Group (1973) which is described in detail in Meinert (1986). It should be noted that it is not an easy task to handle treatment assignments of more than tens of thousands of patients, especially when the time from the onset of symptom to the treatment is also considered as a crucial factor such as rt-PA for acute ischemic stroke (National Institute of Neurological Disorder and Stroke rt-PA Stroke Study Group, 1995). Hence a central administrated telephone-based assignment system such as Interactive Voice Response System (IVRS, Chen, 2003), should be employed. For example, ISIS-2 (1988) used a 24hour telephone service, based in Gent and Brussels for Belgium, Berlin for Germany, Valencia for Spain, Bellinzona for Austria and Switzerland, Lyon for France, and Oxford for England and all other countries. The information of patient identifiers such as age, systolic blood pressure, hours from onset of the episode of pain that led to admission, aspirin use during the week before, and the planned treatment in hospital must be completed before a patient is randomized to receive treatments. In addition the method of minimization randomization was also adopted at Oxford for balancing the prognostic factors recorded at entry. However, on January 24,1986, a programming error was discovered that led more patients randomized at Oxford to being allocated to the placebo infusion and placebo tablets over a period of two months (see Chapter 2 for more information on the treatments of ISIS-2). This programming error was corrected and the exact balance restored in August 1986. Similarly the GUSTO trial used a 24-hour a day, seven-day-per-week randomization center to verify patient eligibility, informed consent, and to assign treatments to more than forty thousand patients. Note that randomization can be performed through a computer networking system such as internet, or web-based networking system. A computerized standard form of eligibility information and informed consent must be sent with the request to the randomization center. Then a validated computer program at the randomization center can immediately enter the data of eligibility for a patient interactively on line through web-interface and verify the patient's eligibility. If inclusion criteria are met and none of the exclusion criteria are observed, then a random assignment of the patient to a particular treatment can be issued in a blinded fashion and sent to the study center. Otherwise, a message of reasons for refusal to issue randomization codes should be sent to the study center. This randomization process is not only accomplished in seconds but also eliminates the human errors that often occur during the randomization process. It should be noted that a computer system should be validated if it is to be employed for the generation of randomization codes. In addition all personnel should have appropriate training. It is suggested that several dry runs with simulated cases be done before the actual implementation of the system takes place.

Your Heart and Nutrition

Your Heart and Nutrition

Prevention is better than a cure. Learn how to cherish your heart by taking the necessary means to keep it pumping healthily and steadily through your life.

Get My Free Ebook


Post a comment