Identifying Patterns of Smoking Cessation App Feature Use That Predict Successful Quitting: Secondary Analysis of...

Area of Science:

• Digital Health
• Behavioral Science
• Machine Learning

Background:

• Smartphone apps offer accessible, evidence-based interventions for smoking cessation.
• Research is needed to understand how specific app features influence cessation success.

Purpose of the Study:

• To develop supervised machine learning (SML) algorithms to identify smoking cessation app features that promote successful quitting.
• To assess if app feature usage explains cessation variance beyond established predictors like tobacco use behaviors.

Main Methods:

• Utilized observational data from 133 participants in a quitSTART app experiment.
• Employed logistic regression SML modeling to predict cessation probability based on 28 app usage variables, experimental conditions, and phone type.
• Validated the SML model's accuracy in a held-aside test set and assessed its contribution to explaining cessation variance.

Main Results:

• The SML model demonstrated reasonable accuracy in predicting cessation based on app feature usage patterns (sensitivity=0.67, specificity=0.67).
• Including SML-predicted cessation probabilities in a logistic regression model did not significantly improve prediction compared to a model with only demographic and tobacco use variables (P=.16).

Conclusions:

• Supervised machine learning (SML) can analyze user data to identify effective features within smoking cessation apps.
• This methodological approach can guide the development and enhancement of digital tools for tobacco cessation.