Version 1.3 Effective date: 21 May 2012 Author: Approved by: Dr Ranjit Lall, Research Fellow Statistician Dr Sarah Duggan, CTU Manager Revision Chronology: Effective Date Version 1.3 21 May 2012 Version 1.2 1 April 2010 Version 1.1 31st January 2008 Reason for change Bi-annual review: Web site link updated. Format change Bi-annual review: Web site link updated Bi-annual review: Format change. Version 1.0 March 2006 Page 1 of 5
1. Purpose The purpose of this Standard Operating Procedure (SOP) is to detail the statistical input into the conduct of a clinical trial. 2. Background Access to statistical expertise is essential before and during the entire trial procedure. This should commence with designing the protocol and end with completion of the final report. 3. Procedure 3.1 Who? This SOP has been written primarily for the statistician who will be assisting in the conduct of a trial. 3.2 When? Statistical input should be provided throughout the entire trial. 3.3 How? 3.3.1 Planning and preparation of a new trial After agreement on the objectives of the trial, a statistician must critically review the entire first draft of the protocol with special attention being made to: objective of the trial (endpoints) and an a priori rationale for the target differences between treatments which the trial is being designed to detect; trial design e.g. parallel groups, crossover design; Criteria for evaluation and definition of end-points; e.g. response rate, quality of life assessment and methods of computation and calculation; Registration and randomisation of patients (stratification factors), procedures and practical arrangements. In the case of a blinded trial the protocol must state the conditions for which the code may/must be broken. A system is required enabling access to the treatment of individual subjects in case of emergency. The system must only permit access to treatment code of one subject at a time. If the code is broken this must be justified. Measures to avoid bias. 3.3.2 Statistical considerations for the protocol Sample size, taking into account clinical and scientific information and professional judgement on the clinical significance of differences that could be detected. Power Significance Statistical methods Page 2 of 5
Interim analysis: the possibility and circumstances of these, along with their frequency, must be specified. Early stopping rules (generalised) The statistician must review the final version of the protocol. After the approval of the protocol, the statistician should review the case report forms (CRFs) and the implemented registration/randomisation procedures. 3.3.3 Statistical Analysis Plan See SOP 21 Statistical Analysis Plan The Statistical Analysis Plan (SAP) should be produced following the guidelines detailed in the Statistical Analysis Plan SOP. Any deviations from the final SAP should be described and justified in the final report of the trial. 3.3.4 Statistical Analysis The analysis of the trial must be carried out or confirmed by an identified, appropriately qualified and experienced Statistician. Prior to any analysis the trial statistician should perform an initial data freeze of the trial database and place it on a secure specific directory (i.e. create a non-dynamic dataset). The trial statistician should be responsible for importing data into their preferred validated statistical package (e.g. SAS, Stata, S-Plus, SPSS, BMDP), SAS being the current standard. Prior to analysing the data, the statistician must carry out validation checks on the data quality and integrity (e.g. range checks, outliers, missing observations) as detailed in the SAP. The trial statistician should refer any data queries arising during the analysis to the trial co-ordinator and/or data manager (as appropriate) for investigation or resolution. The trial/data manager should amend the current (dynamic) database and the statistician should delete their initial data freeze and perform a re-freeze of the current database. Steps 1-4 may be repeated until data queries are satisfied before the final data lock. The trial statistician is responsible for the statistical programmes for analyses, which include appropriate exploratory annotation throughout. The final version will be placed in the Statistical Analysis Plan Master file. The trial statistician is responsible for running the statistical programmes to carry out analyses. Results should be discussed with the Chief Investigator as appropriate. Statistical results should be reported according to the CONSORT guidelines. (A link to these is available via the clinical trials unit website at http://www2.warwick.ac.uk/fac/med/research/hscience/ctu/conducting) Page 3 of 5
The results of the analyses should be presented in a manner likely to facilitate the interpretation of their clinical importance. Estimates of the magnitude of the treatment effects or differences and confidence intervals should be quoted, rather than placing sole reliance on significance testing. The statistician is responsible for producing open and closed (i.e. confidential) statistical reports as required. The open report will be endorsed by the inhouse Trial Management Group for typographical or analytical errors. The final version is signed and dated by the statistician(s) involved and placed in the Trial Results Master file. All closed reports and patient data should be kept confidential (e.g. locked in filing cabinet). Statistical Analysis Master File: This should include the statistical analysis plan and specify the location of all electronic files contained in the SAP master file. 3.3.4 Interim analyses 3.3.4.1 Objectives An interim analysis includes any examination of the data during the course of a trial for which results are presented for one or more treatment groups. The interim analysis CTAP should provide a comprehensive and detailed description of the methods of analysis and presentation of the data. Interim analyses should follow the same procedure as a final analysis. 3.3.4.2 Procedures Any changes, e.g. unplanned interim analysis, must be justified and fully documented in the final statistical report. If appropriate, the interim analysis should also state the method of unblinding. This must ensure that the conduct of the trial is not comprised. The SAP for the interim analysis/analyses should clearly state the reasons for the interim analysis (e.g. ethical (including safety or efficacy) or to provide information for trial management). The plan should also consider whether there is likely to be sufficient power to satisfy the trial objectives. The plan should also state which committees review full or summarised (e.g. blinded) reports of the interim analysis results. Any restriction on circulation should be defined. 3.3.5 Trial results The Trial Results Master File is a file which includes all documents produced: Independent Data Monitoring Committee reports, abstracts, final paper(s) and summary reports from other analyses such as planned interim analyses. Both files (Statistical Analysis Plan Master file and Trial Results Master file) are the responsibility of the trial statistician. 3.3.6 Missing data The handling of any missing data within clinical trials is an important consideration, as failure to identify properly the influences of the missing data Page 4 of 5
may cause bias and possibly nullify the value of the obtained results, as their validity will be questionable. The number of incomplete variables, the patterns and the frequency of any missing data in all outcome variables and covariates should be investigated. Any retrievable data should be collected prior to analyses (see SOP 15 Data Management). If possible, record any potential reasons for the missing data that will help to determine the type of missing data; that is, whether it is Missing Completely at Random (MCAR), Missing at Random (MAR) or Missing Not at Random (MNAR). The mechanism that resulted in the missing data should be investigated as far as possible, by comparing the characteristics of cases with complete data with those with incomplete data. A decision is required regarding the most appropriate method for handling the missing data. Justifications for this choice should be provided. Complete case analysis uses only the complete cases, and, therefore, is always inefficient, even when the estimates are unbiased as the data are MCAR. Imputation fills in the incomplete data with plausible values, based on the strengths of the associations with the observed data, thus allowing all cases to be analysed and reducing the non-response bias. When imputation is considered appropriate, multiple imputation will generally be preferred over single imputation as it properly accounts for the uncertainty in the imputed values, especially when missing data exists in more than one variable. The method and model used for imputation, together with the statistical programme used should be clearly specified. All missing data methods have underlying and often untestable assumptions. Therefore, the sensitivity of the results to these assumptions and also other model specifications should be investigated. Some data may require their own specific algorithms for imputing missing data (e.g. the SF-12 scale). These algorithms should be stated and referenced in the SAP. List of abbreviations CRF Case Report Form MAR Missing at Random MCAR Missing Completely at Random MNAR Missing Not a Random SAP Statistical Analysis Plan SOP Standard Operating Procedure Page 5 of 5