mCrave: Continuous Estimation of Craving During Smoking Cessation

Area of Science:

• Computational psychiatry
• Behavioral science
• Digital health

Background:

• Cigarette craving often precedes relapse in smoking cessation.
• Passive estimation of craving using sensor data could aid in relapse prevention.
• Current methods for craving assessment are often reactive and intrusive.

Purpose of the Study:

• To develop a computational model for minute-level estimation of cigarette craving during smoking abstinence.
• To explore the relationship between sensor-derived stress measures and self-reported craving.
• To lay the groundwork for passive, in-situ craving detection.

Main Methods:

• Utilized 2,012 hours of mobile sensor data and 1,812 craving self-reports from 61 participants in a smoking cessation study.
• Derived a continuous measure of stress from sensor data.
• Developed feature functions based on stress and other insights.
• Employed a Conditional Random Field (CRF) model to infer craving probabilities.

Main Results:

• Identified a correlation between higher stress levels and self-reported high craving during specific times of day.
• Successfully developed a model capable of inferring craving probabilities using sensor data.
• Demonstrated the feasibility of minute-level craving estimation.

Conclusions:

• Passive estimation of cigarette craving is feasible using mobile sensor data and computational modeling.
• Stress levels derived from sensor data can serve as an indicator of craving.
• This approach holds potential for real-time support in smoking cessation interventions.