Ethical aspects and user preferences in applying machine learning to adjust eHealth addressing substance use: A mixed-methods study

Background: Digital health interventions targeting substance use disorders are being increasingly implemented. Data science methodology has the potential to enhance involvement and efficacy of these interventions, though application may raise ethical considerations. This study aimed to explore possible ethical aspects and preferences among users of an online digital intervention for substance use and gambling disorder regarding the application of supervised machine learning (ML) methodology. Methods: We recruited participants from a widely used, evidence-based online substance use and gambling intervention from the Netherlands (Jellinek Digital Self-help). Initially, we conducted two online focus groups (n = 5 each) to explore topics related to ethical considerations and user preferences regarding the application of ML for adapting unguided digital interventions. Subsequently, the findings from these focus groups informed the development of an online, quantitative, self-reported questionnaire study regarding this topic (n = 157). Data collection and analyses were guided by the principles of biomedical ethics by Beauchamp and Childress. Results: Our qualitative and quantitative results revealed that digital intervention users found the application of machine learning analyses to be ethically acceptable, although they had difficulties conceptualizing ML applications. Participants believed that it could benefit the intervention and subsequently their well-being. Both qualitative and quantitative results emphasized the importance of preserving user autonomy in applying supervised ML to adjust digital interventions. In addition, based on both data sources we found that digital intervention users trusted Jellinek's integrity to apply ML. Ethical concerns identified in the qualitative data (e.g., data security, human control), were not confirmed in our quantitative findings. Conclusions: This mixed-methods study revealed that users of digital intervention demonstrated limited concern for ethical aspects regarding applying ML to adapt digital interventions. Ethical aspects were primarily pertained to their needs for autonomy and privacy.