GUIDELINES FOR GOOD PRACTICE IN THE CONDUCT OF CLINICAL TRIALS IN HUMAN PARTICIPANTS IN SOUTH AFRICA
6. Data Management and Statistics - Contents
6.1 Protocol
6.2 Randomisation
6.3 Data Management6.3.1 Data Integrity and Transfer
6.3.2 Case Report Form
6.3.3 Data Quality Control
6.3.4 Code Breaking for Data Analysis6.4 The Final Study Report
6.5 Preservation of Records
6.6 Archiving by the Principal Investigator
6.7 Archiving by the Sponsor
Statistical considerations must be an integral part of study design and the description and analysis of the final results. The aim should be to integrate the clinical and statistical aspects as fully as possible through all phases of a clinical study.
Statisticians and data managers should have advisory and operative functions in the design of clinical studies, in reviewing study protocols, CRFs and manuscripts, and in the design of computer systems for data processing. The operative functions consist of writing or advising on statistical sections in protocols and reports, adapting or developing appropriate methods for the statistical analysis, performing the analysis and making the statistical interpretation of the results.
The protocol and the final study report should be reviewed and commented upon by a statistician.
6.1 PROTOCOL
Statistical considerations in a study protocol should be an integral part of the document and special attention should be paid to the following aspects:
- that the experimental design of the study is appropriate to the purpose of the study and the nature of the study condition;
- that the sample size is large enough to allow detection of clinically meaningful differences or to document clinical equivalence as appropriate, in variables specifically related to the hypotheses stated in the protocol;
- that the calculation of the sample size is based on sufficient statistical power and adequate levels of significance specified before commencement of the study;
- that the calculation of sample size takes into account any planned interim or sub-group analyses;
- that the randomisation procedures must be scientifically justified in the protocol;
- that for comparative, open-label studies measures taken to avoid bias are specified (preferably assigned by central randomisation);
- that the reasons and procedures for any planned interim analyses, whether used in sequential design for possible early termination or for any other purpose, are clearly specified in writing. It is advisable to have an independent supervisory committee to perform interim analyses to maintain study blindness among investigators and others involved;
- that the general approach to the statistical analysis is described, including methods of adjusting for possible imbalance in prognostic factors at baseline;
- that criteria which can be foreseen for including and/or removing participants from certain analyses are defined;
- that the use of intention-to-treat analysis and/or per protocol analysis is specified. (This will usually only apply to analyses of efficacy variables.);
- that methods for dealing with/analysing incomplete data (missing observations, premature withdrawals, etc.) are specified; and
- that specific statistical models and methods which can be foreseen are used even if some flexibility must be allowed at the time of analysis.
6.2 RANDOMISATION
In the case of randomisation of participants, the procedure must be documented. In a blinded, randomised study it is usually necessary to supply and keep the treatment code for each individual participant at both the study site and with the sponsor.
The protocol must state when the treatment code for an individual participant may or may not be broken. Breaking the treatment code must be possible 24 hours a day in case of an emergency and the procedures and persons involved clearly designated in the protocol and related documents. The date, reasons for, and name of the individual breaking the study code must be documented.
Before the treatment code is broken for statistical analysis, the code for each participant must be returned to the sponsor with a documented explanation for each episode where the code was broken. Any master code supplied (e.g. to the pharmacy) must be returned to the sponsor. No copies of the code should be taken by any person involved in the study. Copies of the treatment code will be available to the investigator at the end of the study after the database is 'locked'.
6.3 DATA MANAGEMENT
The aim of data management is to turn the information from the participant, efficiently and without errors, into data in the statistical database.
All steps involved in data management should be part of a standard operating procedure and should be documented to allow for a step-by-step retrospective assessment of data quality and study performance; i.e. an audit trail. Documentation is facilitated by the use of such means as checklists and forms giving details of action taken, dates and the individuals responsible.
6.3.1 Data integrity and transfer
A basic aspect of the integrity of data is the safeguarding of "blindness" with regard to treatment assignment. It starts with the randomisation of participants into treatment groups and it is maintained through all steps of data processing up until the decision to break the code is formally taken.
The confidentiality of the database must be secured by appropriate standard operating procedures including passwords for all staff involved in the case of a computer database. The use of a computer system which logs who has had access to the information, and logs and dates all changes to the information is recommended. Satisfactory maintenance and back-up procedures for computer databases must be provided.
6.3.2 Case Report Form (CRF)
The design of the case report form should meet the specific data requirements set out in the study protocol. In addition, basic form design concepts should be adopted where feasible. Such concepts relate to consistency in the use of reference codes, terminology and format. Standardisation in the design of forms and computer programmes used in the data processing and statistical analysis will save time and prevent errors.
6.3.3 Data quality control
The aim of quality control procedures is to minimise the effects of missing and inaccurate data. The data editing process should be part of the standard operating procedures documentation which describe the process for confirmation and correction of data. The standard operating procedure for data editing should guarantee that any queries about data validation are brought rapidly to the attention of the investigators.
An audit trail should be available to trace the nature of any changes to data, the dates of changes and the person responsible for the changes.
Data collection and entry should be performed continuously during the course of the study. It should be checked either by double-entry (preferable) or by proof-reading for the primary variables and on a random basis for other parameters.
Checks for validity and consistency of the database should be on separate items as well as on predetermined combinations of items in the CRFs.
Checks should be manual as well as computerised. In the latter case it should be combined with data entry (e.g. immediate automatic checks or batch checking) in order to speed up feed-back on data requiring clarification.
To supplement the continuous checking of each individual’s data during the study, descriptive statistics on each important variable in the database (performed without breaking the code) are useful in the detection of doubtful and/or unusual data.
6.3.4 Code breaking for data analysis
When the validation and editing process is concluded the formal 'locking' of the database should be documented.
Data for each individual participant should be classified and coded with respect to its inclusion in the various statistical analyses planned in the study and the code entered into the data base.
After the above actions have been documented the treatment code can be broken and included in the data base for each individual participant.
6.4 THE FINAL STUDY REPORT
The protocol, statistical and clinical aspects should be integrated in order to obtain a final study report that is entirely consistent with the study data generated. Essential elements in the presentation of the results of a study of an investigational product are:
- baseline comparisons between the treatment groups;
- the number of participants actually randomised into the study by treatment group and the number of participants excluded from any of the analyses, by reason and by treatment group;
- major efficacy and safety results by treatment group in the form of tables, graphs, test variables and statistical parameters (e.g. p-values) as appropriate;
- an assessment of between-group differences with confidence intervals; and
- in multicentre studies, an evaluation of centre effect may be a valuable addition and should always be conducted where significant inter-centre variation is suspected.
An account must be made of missing, unused or spurious data during statistical analyses. All omissions of this type must be documented to enable review to be performed.
The final study report must be recorded in the database of the Director General of Health.
6.5 PRESERVATION OF RECORDS
Both the principal investigator and the sponsor are obliged to retain records and data from the study for safety reasons and for audit and inspection subsequent to study completion. The time frames depend on the nature and duration of the study and must conform to the requirements of the relevant privacy legislation and the South African Constitution. Acceptable archiving practices are 5 years for biological samples and 10 years for written materials. Samples, documents and any computer records should be retained in a secure place to prevent undue access, loss or tampering.
6.6 ARCHIVING BY THE PRINCIPAL INVESTIGATOR
The principal investigator is responsible for maintaining copies of all documentation which contains identifying source data and other essential documentation including, the study protocol and amendments, applications to the ethics committee, serious adverse event reports and all other correspondence relating to the study.
All correspondence between the principal investigator and the sponsor must be preserved and be available on the request of the study sponsor, the regulatory authority, independent auditors or ethics committee. Investigational product accountability records detailing the storage and use of the product should also be retained.
Adequate steps must be taken to ensure that the hospital case records of all participants in clinical research are retained for this period, which is longer than the time to destruction interval in some hospitals and institutions.
If the principal investigator is unable to maintain custody of the study documents and samples, the sponsor should be informed in writing about the location of the records and the name of the person responsible for their retention. If necessary the sponsor may inventory and retain, in a sealed container, the investigator’s documents. The means by which prompt access can be assured should also be stated.
6.7 ARCHIVING BY THE SPONSOR
The sponsor should retain copies of all essential documentation relating to the study which do not contain participant identifying information. These include reports to the regulatory authority, records of monitor-investigator contacts and investigational product supplies.
The files should also include information on the person(s) at the study site maintaining custody of the participant lists and responsibility for the archiving of the investigator’s documents.
The period and conditions under which the documents should be saved is no different than those imposed on the investigator; i.e. at least 10 years after termination of the study and preferably for the commercial lifetime of the product.
Computer records must be reproduced in hard copy, which are to be signed and dated as a verified accurate copy of the original data. The verified hard copy should then be stored with other paper-based records, to overcome the possibility of loss or inability to read the information due to technological redundancy.