Theory-Guided Randomized Neural Networks for Decoding Medication-Taking Behavior.

Area of Science:

• Computational modeling
• Behavioral economics
• Psychology

Background:

• Long-term endocrine therapy is vital for breast cancer recurrence prevention.
• Low medication adherence rates pose a significant challenge.
• Understanding behavioral mechanisms is crucial for intervention development.

Purpose of the Study:

• To develop an individual-level model predicting breast cancer survivors' daily medication-taking behavior.
• To evaluate the model's predictive accuracy using survey data over time.
• To inform the design of future adherence interventions.

Main Methods:

• Development of a randomized neural network model.
• Utilizing survey data from three time points: baseline, 4 months, and 8 months.
• Guiding neural network structure with random utility theory from psychology and behavioral economics.

Main Results:

• The proposed randomized neural network model accurately predicts daily medication-taking behavior.
• The model demonstrates superior prediction accuracy compared to existing computational models.
• Performance was evaluated under conditions of randomness.

Conclusions:

• Individual-level modeling, particularly with advanced techniques like randomized neural networks, can effectively predict medication adherence.
• This approach offers a promising tool for understanding and improving adherence in breast cancer survivors.
• The findings support the integration of psychological and economic theories into computational models for health behavior research.