Background
The contribution of the statistician to the design and analysis of a clinical trial is acknowledged as essential. Ability to reconstruct the statistical contribution to a trial requires rigorous and transparent documentation as evidenced by the reproducibility of results. The process of validating statistical programmes is a key requirement. While guidance relating to software development and life cycle methodologies details steps for validation by information systems developers, there is no guidance applicable to programmes written by statisticians. We aimed to develop a risk-based approach to the validation of statistical programming that would support scientific integrity and efficient resource use within clinical trials units.
Methods
The project was embedded within the Information Systems Operational Group and the Statistics Operational Group of the UK Clinical Research Collaboration Registered Clinical Trials Unit network. Members were asked to share materials relevant to validation of statistical programming. A review of the published literature, regulatory guidance and knowledge of relevant working groups was undertaken. Surveys targeting the Information Systems Operational Group and Statistics Operational Group were developed to determine current practices across the Registered Clinical Trials Unit network. A risk-based approach was drafted and used as a basis for a workshop with representation from statisticians, information systems developers and quality assurance managers (n = 15). The approach was subsequently modified and presented at a second, larger scale workshop (n = 47) to gain a wider perspective, with discussion of content and implications for delivery. The approach was revised based on the discussions and suggestions made. The workshop was attended by a member of the Medicines for Healthcare products Regulatory Agency Inspectorate who also provided comments on the revised draft.
Results
Types of statistical programming were identified and categorised into six areas: generation of randomisation lists; programmes to explore/understand the data; data cleaning, including complex checks; derivations including data transformations; data monitoring; or interim and final analysis. The risk-based approach considers each category of statistical programme against the impact of an error and its likelihood, whether the programming can be fully prespecified, the need for repeated use and the need for reproducibility. Approaches to the validation of programming within each category are proposed.
Conclusion
We have developed a risk-based approach to the validation of statistical programming. It endeavours to facilitate the implementation of targeted quality assurance measures while making efficient use of limited resources.
CORE (COnnecting REpositories)
CORE (COnnecting REpositories)