From Reflection to Action: Combining Machine Learning with Expert Knowledge for Nutrition Goal Recommendations.

Area of Science:

• Digital Health
• Machine Learning in Healthcare
• Behavioral Science

Background:

• Self-tracking is crucial for managing chronic conditions like type 2 diabetes (T2D).
• Personalized interventions require users to interpret and act on their health data, which demands motivation and health literacy.
• Existing machine learning (ML) methods can identify patterns but often struggle to generate actionable insights for users.

Purpose of the Study:

• To introduce GlucoGoalie, a system combining ML and expert systems for personalized nutrition goal suggestions in T2D.
• To evaluate the understandability and actionability of ML-generated goal suggestions.
• To explore the impact of personalized goal suggestions on self-discovery, goal setting, and adherence in a real-world setting.

Main Methods:

• Development of GlucoGoalie, integrating ML pattern identification with an expert system for actionable recommendations.
• A controlled experiment to assess user perception of goal suggestion understandability and actionability.
• A 4-week in-the-wild study to observe the effects of goal suggestions on user behavior and experience.

Main Results:

• Participants in the controlled experiment found T2D nutrition goal suggestions understandable and actionable.
• The in-the-wild deployment indicated that goal suggestions enhanced self-discovery and highlighted individual preferences.
• User experiences revealed the importance of feedback and context, alongside challenges with abstract goals and ambiguous static text.

Conclusions:

• ML-driven interventions can be enhanced by expert systems to provide personalized, actionable health goals.
• Future systems should incorporate greater interactivity, feedback mechanisms, and negotiation capabilities to bridge the gap between abstract goals and concrete behaviors.
• Addressing user needs for clarity and context is vital for the success of digital health tools in chronic disease self-management.