Semi-automated entry of clinical temporal-abstraction knowledge

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Knowledge Engineering

Background:

• Raw time-oriented clinical data requires sophisticated methods for extracting meaningful patterns.
• Previous work developed the RESUME system for high-level concept formation from clinical data.
• A need exists for efficient methods for clinical experts to input knowledge into such systems.

Purpose of the Study:

• To evaluate the usability and effectiveness of a graphical knowledge acquisition (KA) tool designed for clinical experts.
• To assess the tool's ability to facilitate the entry of knowledge for the RESUME system.
• To compare the validity of knowledge entered via the KA tool versus manual entry.

Main Methods:

• Designed a graphical knowledge acquisition (KA) tool using Protégé, integrating with the RESUME system.
• Evaluated usability with expert physicians and knowledge engineers across three clinical domains.
• Measured time for understanding, learning, and knowledge entry; analyzed qualitative feedback.
• Compared system-generated abstractions with benchmark data and manually entered knowledge.

Main Results:

• Physicians required 2-6 hours (median 3-4) to learn the KA tool; knowledge entry times varied by domain (median 3-10 hours).
• Significant speed increases (up to 25x, median 3x) observed upon repeated use of the KA tool.
• Knowledge engineers' initial entry times were comparable to physicians' second attempts.
• RESUME generated nearly identical abstractions using KA tool-entered knowledge versus manually entered knowledge.

Conclusions:

• The graphical knowledge acquisition tool is usable and effective for expert physicians and knowledge engineers.
• The tool facilitates efficient entry of clinical temporal-abstraction knowledge.
• Knowledge bases created with the KA tool are as valid as those produced by manual entry.