4570. Development of Web Toolkit for Automatic Data Management of RECIST Datasets from Conformity Check to Statistical Analysis in Clinical Trials
Authors* Denotes Presenting Author
Jimi Huh;
Ajou University Hospital
Amy Lee;
Trial Informatics
Kyung Won Kim *;
Asan Medical Center
Objective:
Major issues in data management of tumor response assessment using RECIST in clinical trials encompass substantial human errors during data entry and the allocation of significant human resources for statistical analysis. These issues underscore the need for an innovative IT solution to achieve a breakthrough in this field. Therefore, we developed a web toolkit for automatic conformity check and real-time statistical analysis.
Materials and Methods:
In accordance with the Clinical Data Interchange Standards Consortium (CDISC) standards, which is the only standard data format permitted by the US-FDA for clinical trials, the web-toolkit architecture is designed with several modules. The first module is an electronic case report form (eCRF) function for data entry based on CDISC standards. The second module is an automatic conformity check function that checks for any errors in data entry. The third module is an automatic statistical analysis function developed using a R-shiny techniques. Validation of the web-toolkit was performed by two experts (a radiologist and a statistician) in the clinical trial imaging field based on a real-world data composed of 54 cases, which had generated RECIST datasets.
Results:
The web toolkit's modules are showcased on a website, each serving distinct functions. The initial module enforces adherence to CDISC standards during data input. The second module automatically validates input data against RECIST conformity criteria. Subsequently, validated data are automatically converted into CDISC SDTM datasets. The third module offers real-time access to diverse statistical analyses like outlier detection, inter-reader variability analysis, waterfall plots, and survival curve efficacy assessment. Validation studies exposed 49 human errors in real-world RECIST datasets. Upon using our web toolkit, the radiologist achieved error-free data entry and CDISC-compliant datasets. Our statistician validated the ensuing statistical analyses.
Conclusion:
To ensure the high quality of tumor response assessment data, our innovative web toolkit employs a CDISC-compliant eCRF equipped with an automated conformity check module. Additionally, it delivers real-time access to a range of statistical analysis outcomes. This combination effectively minimizes human input errors and enhances data accuracy.
