Using an AI + Human in the Loop (HITL) approach can be utilized for Study Data Tabulation Model (SDTM) transformation, potentially alleviating current challenges
The Study Data Tabulation Model (SDTM) is a set of standards developed by the Clinical Data Interchange Standards Consortium (CDISC) to exchange clinical data. SDTM is an essential standard for organizing and formatting clinical trial data. It is a regulatory mandate for many countries, including the United States and the European Union, to submit clinical datasets for drug approval. SDTM helps to ensure the quality and accuracy of data and the consistency of the data across multiple studies. It also helps to improve the efficiency of data review and analysis and the speed of the regulatory review process. The SDTM standards are made up of several different components, including Domains, Elements & Tables. However, the clinical data is not collected in the native SDTM format. The clinical data needs to be converted into SDTM format, a time-consuming, tedious, and expensive process with challenges.
The challenges of SDTM transformation
SDTM transformation is a complex and challenging process that requires a deep understanding of the standards, the data, and the regulatory environment. The standards constantly evolve, making it challenging to keep up with the latest changes. Additionally, the data can vary widely, depending on the study design, indication, population, and other factors. This variability makes it challenging to develop a one-size-fits-all approach for SDTM transformation. The regulatory environment constantly changes, with new requirements and guidance being released. This can make it difficult to stay compliant with the latest regulations. Finally, the standards are only sometimes compatible with existing clinical data management systems (CDMS) and other data acquisition systems, which can lead to further challenges in the SDTM transformation process. To address these challenges, an AI + Human in the Loop (HITL) approach can be utilized for SDTM transformation, potentially alleviating these challenges.
Traditional approach to SDTM transformation
The traditional approach to SDTM transformation involves manual and rule-based processes that are time-consuming, error-prone, and require specialized technical expertise. This approach often involves creating custom programs to extract data from different sources, manually mapping the data to SDTM domains and elements, and validating the data against the standards. Many sponsors have also attempted Metadata Repository (MDR) approach to manage SDTM transformation. MDR SDTM conversion is an approach to converting clinical trial data to SDTM format that involves metadata and a centralized repository. The metadata describes the structure and content of the source data, and the repository contains information about the SDTM domains, variables, and their relationships. Though MDR-driven, SDTM conversion can improve the efficiency and accuracy of the SDTM. However, the study-specificity, scale, and maintenance are still the most significant issues with this approach. Many sponsors are turning towards advanced technologies and solutions that can automate SDTM transformations through AI + HITL approaches to tackle these challenges.
AI + HITL approaches for SDTM transformation
AI + HITL approaches for SDTM transformation are emerging as promising solutions to address the challenges of traditional manual processes. These approaches leverage machine learning algorithms to learn from data patterns and generate rules for automated SDTM transformation. Human oversight and interaction are still required with these approaches to ensure the accuracy and validity of the data. Additionally, these approaches can handle various data formats and sources, making them more flexible and adaptable to the variability in clinical trial data. The AI + HITL approach involves data preparation, data conversion and validation, and final review by human experts, and are listed below.
Data preparation: The first step is to prepare the data for conversion. This includes cleaning the data, identifying missing values, and standardizing the data formats.
Data conversion and validation: In this step, the AI algorithms analyze the data patterns and generate rules for mapping the source data to SDTM domains and variables. The HITL component occurs during the validation stage, where human experts ensure the AI-generated rules are accurate and complete.
Finally, the human experts review and validate the converted data to ensure its accuracy and completeness. AI + HITL approaches for SDTM transformation offer a more efficient, accurate, and adaptable means of converting clinical trial data to SDTM format. These approaches have the potential to significantly reduce timelines and costs associated with SDTM conversion while ensuring the quality and accuracy of data.
Lessons learnt from AbbVie implementation
AbbVie deployed a state-of-the-art clinical data repository and implemented a Machine Learning (ML) module for SDTM transformation. The ML models were trained using legacy SDTM datasets and the CDISC SDTM Implementation Guide (IG). The ML models were validated with actual study datasets. Once the module was validated, several new studies were transformed into SDTM and sent to experts for their validation. The team learned some valuable lessons for scaling this to the broader portfolio, such as:
Selecting a diverse set of studies to train the ML model. This will help to ensure that the model can generalize to new data.
Expecting an increase in auto transformation as more studies are deployed. This is because the ML model will learn from the data that is being transformed and will become more accurate over time.
Automating human feedback into the system for model enhancement. This will help to improve the accuracy of the ML model over time.
Evaluating the performance of the ML model before it is used to map data. This can be done using a holdout dataset not used to train the model.
Concluding thoughts
AI can automate many tasks in SDTM transformation, such as data cleaning, mapping, and transformation. This can free up human resources for more complex tasks like data analysis and interpretation. AI can also help improve accuracy by identifying and correcting errors. However, human oversight is critical to ensure data is converted accurately and by regulatory requirements. High-quality training data and monitoring of AI SDTM models are also essential.
Overall, AI and humans can work together to improve the accuracy and efficiency of SDTM transformation. AI can automate many tasks, while humans can provide oversight and insights. Using high-quality training data and monitoring and evaluating AI SDTM models’ performance is essential.
Disclaimer
“This publication was neither originated nor managed by AbbVie, and it does not communicate results of AbbVie-sponsored Scientific Research. Thus, it is not in the scope of the AbbVie Publication Procedure (PUB-100).”
