Proof-of-concept for using machine learning to facilitate data extraction for human health chemical assessments: a...

Area of Science:

• Environmental Health Sciences
• Toxicology
• Computational Biology

Background:

• Systematic review (SR) methods are crucial for transparent chemical assessments but often rely on manual, unscalable data extraction.
• Existing SR tools offer semi-automation for data discovery but struggle with efficient, interoperable data extraction.
• Manual data extraction remains a bottleneck in developing comprehensive chemical assessments.

Purpose of the Study:

• To explore the integration of a semi-automated data extraction tool, Dextr, into chemical assessment workflows.
• To assess whether Dextr improves the overall user experience in the chemical assessment process.
• To evaluate the efficiency and performance of Dextr compared to manual data extraction methods.

Main Methods:

• Utilizing template systematic evidence map (SEM) methods for study identification.
• Comparing fully manual data extraction with semi-automated (human-in-the-loop) extraction within Dextr.
• Evaluating Dextr's user experience and the precision, recall, and F1 scores of its automated data extraction.

Main Results:

• The study protocol outlines a comparative evaluation of manual versus semi-automated data extraction using Dextr.
• Performance metrics (precision, recall, F1 score) will be used to quantify the accuracy of Dextr's automated extraction.
• User experience feedback will be collected to assess the impact of Dextr on workflow efficiency.

Conclusions:

• Semi-automated tools like Dextr show promise for enhancing efficiency in chemical assessment workflows.
• Further research is needed to determine if Dextr improves operational efficiencies and user experience.
• Successful integration of AI/ML tools is key to advancing scalable and sustainable chemical assessment processes.