A knowledge-based clinical toxicology consultant for diagnosing multiple exposures.

Area of Science:

• Clinical Toxicology
• Medical Informatics
• Artificial Intelligence in Medicine

Background:

• Multiple toxin exposures are rare but account for over half of toxin-related fatalities.
• Accurate diagnosis of complex toxic exposures is challenging.
• A decision support system can aid clinicians in identifying unknown substances.

Purpose of the Study:

• To develop a knowledge-based system for diagnosing human toxic exposures, with a focus on multiple toxin cases.
• To create a practical decision support tool using medical mathematics and data mining.
• To improve the diagnosis of complex cases involving multiple unknown toxins.

Main Methods:

• A system was trained using data mining on Florida Poison Information Center (FPIC) data.
• Prior probabilities and likelihood ratios were extracted from the database.
• The system was tested on its ability to diagnose toxins at substance, major/minor category, and major category levels.

Main Results:

• Initial testing on multiple exposures showed low accuracy (18.5%-50.1%).
• Training with single exposures improved accuracy, reaching 83.5% for substance identification in multiple exposure cases.
• The system successfully identified primary contributors in complex toxic exposure scenarios.

Conclusions:

• The system, while not fully diagnosing multiple toxins, can identify primary contributors, aiding medical professionals.
• Continued data accumulation in the FPIC database is expected to enhance system accuracy.
• The developed system shows potential as a viable decision support tool for clinical toxicology.