Samenvatting

Artificial Intelligence (AI) is transforming workflows across various industries, including life sciences. When deployed securely, generative AI tools can automate structured programming tasks. In clinical research, a significant challenge is converting raw trial data into the formats required for regulatory purposes. The Clinical Data Interchange Standards Consortium (CDISC) establishes global data standards for submissions to regulatory authorities such as the FDA or EDA. Its Study Data Tabulation Model (SDTM) ensures uniformity in clinical trial datasets. Producing SDTM-compliant datasets is mandatory for regulatory authorities such as the FDA. It's obligatory to use SAS (Statistical Analysis System) as a programming language. Although powerful, SAS requires manual coding by expert statistical programmers. This task is labour-intensive, time-consuming, and susceptible to human error. Contract Research Organisations (CROs) often assist sponsors in managing this workload. Large CROs can utilise proprietary macros and automated pipelines, while small and medium-sized enterprises (SMEs) typically lack such infrastructure. As a result, they frequently encounter heterogeneous data from multiple vendors, complicating the SDTM conversion process. This study evaluates the use of generative AI as a supportive programmer for dataset conversion. We deployed ChatGPT-4 within a secure Azure AI environment.Iterative prompting, standards SDTM documents, and an increasingly curated Knowledgebase Document guided the AI. Our case study focused on the Demographics (DM) domain, which is a key SDTM dataset. Step by step, the AI generated structured and increasingly accurate SAS code. Through feedback loops, omissions and errors were systematically corrected. After four iterations, we produced 95 clean lines of SAS code. The resulting DM dataset met the compliance requirements for SDTM version 3.4. The Knowledgebase Document created reusable, explicit instructions for future conversions. This approach demonstrated reproducibility, scalability, and alignment with regulatory standards. We conclude that generative AI can reduce time, costs, and resource barriers, and its adoption can accelerate submission readiness for both CROs and SMEs.