A new approach to integrating patient-generated data with expert knowledge for personalized goal setting: A pilot...

Area of Science:

• Biomedical Informatics
• Computational Health
• Personalized Nutrition

Background:

• Patient-generated health data (PGHD) from self-monitoring offers potential for personalized nutrition.
• Existing computational methods struggle with sparse, irregular PGHD for individual-level personalization.
• Integrating PGHD with clinical expertise is crucial for effective nutritional goal setting.

Purpose of the Study:

• To develop and evaluate an informatics approach for generating personalized nutrition goals from PGHD.
• To model registered dietitians' decision-making processes for dietary self-management of diabetes.
• To create a knowledge model and inference engine for personalized nutrition recommendations.

Main Methods:

• Qualitative process coding and decision tree modeling to capture dietitian reasoning (knowledge model).
• Development of an inference engine to process diet and blood glucose data into recommendations.
• Validation of the inference engine against clinical narratives and gold standards from expert clinicians.

Main Results:

• The knowledge model represented dietitian reasoning for personalized nutrition.
• Inference engine recommendations achieved 63% consistency with the gold standard (42%-75%).
• Recommendations showed 74% consistency with narrative clinical observations (63%-83%).

Conclusions:

• Automating dietitian reasoning via informatics shows promise for leveraging PGHD.
• The developed knowledge model and inference engine can synthesize PGHD with clinical knowledge.
• Further research is needed to compare algorithmic and human expert performance in interpreting PGHD.