Supervised machine learning to predict smoking lapses from Ecological Momentary Assessments and sensor data:...

Area of Science:

• Digital Health
• Machine Learning in Behavioral Science
• Smoking Cessation Interventions

Background:

• Smoking lapses often precede full relapse, necessitating timely interventions.
• Just-in-time adaptive interventions (JITAIs) offer a promising approach to target lapses proactively.
• Developing effective JITAIs requires understanding lapse triggers and predicting lapse incidence.

Purpose of the Study:

• To train and test supervised machine learning algorithms for predicting smoking lapses.
• To evaluate the feasibility and performance of algorithms using Ecological Momentary Assessments (EMAs) and wearable sensor data.
• To identify optimal algorithms for informing the decision points and tailoring variables of a lapse prevention JITAI.

Main Methods:

• Adult smokers attempting to quit completed hourly EMAs assessing cravings, mood, context, and lapse incidence over 10 days.
• Participants wore a Fitbit Charge 4 to collect passive data on steps and heart rate.
• Group-level, individual-level, and hybrid machine learning algorithms were trained and tested with and without sensor data.

Main Results:

• Group-level algorithms achieved high predictive performance (AUC up to 0.952 with sensor data) but showed variable individual performance.
• Individual-level and hybrid algorithms, though constructible for fewer participants, demonstrated improved performance, especially with sensor data (median AUC up to 0.983).
• Sensor data integration significantly enhanced the predictive accuracy of algorithms, particularly for individual-level models.

Conclusions:

• Machine learning algorithms, especially when incorporating wearable sensor data, can effectively predict smoking lapses.
• Individualized and hybrid algorithms show potential for personalized JITAIs, despite feasibility constraints for some users.
• Further development is needed to balance algorithm performance, feasibility, and implementation criteria for JITAIs.