Transcription of Handwritten Forms for Medical Study Documentation

Area of Science:

• Clinical data management
• Artificial intelligence in healthcare

Background:

• Manual data entry into electronic clinical Trial Master Files (eTMF) is a laborious and time-intensive process.
• Efficient management of clinical trial documentation is crucial for regulatory compliance and operational efficiency.

Purpose of the Study:

• To evaluate the effectiveness of automated transcription methods for populating eTMF databases.
• To compare the performance of optical document recognition (ODR) and multimodal large language models (LLMs) for document analysis.

Main Methods:

• Development and application of an optical document recognition (ODR) pipeline.
• Implementation and testing of both web-based (global) and local multimodal large language models (LLMs).
• Experimental validation of transcription accuracy and efficiency across different document types.

Main Results:

• The study demonstrated that different automated approaches are optimal for various column types within table-based documents.
• Optical document recognition and LLMs show promise in accurately transcribing information from clinical trial documents.
• Performance varied depending on the specific document structure and content.

Conclusions:

• Automated transcription using ODR and LLMs offers significant time savings compared to traditional manual data entry.
• The findings suggest a viable technological solution to streamline eTMF data management.
• Further optimization of LLM and ODR approaches can enhance efficiency in clinical trial operations.