Logo: to the web site of Uppsala University

Smartphone apps offer new opportunities to study mental health in real-world settings through a combination of passive sensor data and active self-report. This thesis explores how digital mental health research tools can be designed to collect meaningful, ecologically valid data while respecting user experience, motivation, and autonomy. Across four interrelated studies, I examine two app-based cohort studies targeting perinatal women (Mom2B) and young people (UPIC) in Sweden.

The first study presents the technical and ethical foundations of the Mom2B platform, including its integration of digital phenotyping methods. The second study applies machine learning techniques to self-reported data to assess the potential for early prediction of antenatal depression. The third study investigates user attitudes toward data sharing and task engagement, revealing the nuanced balance between research goals and participant comfort. The fourth study follows a user-centered design and usability testing process in the development of the UPIC app, highlighting how early user involvement can improve design, trust, and engagement.

Together, the findings demonstrate the importance of aligning technological possibilities with thoughtful, user-informed design. The thesis contributes to the growing field of digital mental health research by offering practical and ethical insights into the design and evaluation of emotion- and experience-aware research tools.

Perinatal depression (PND), affecting up to 20% of individuals during the perinatal period, presents serious risks to both maternal and child health. Despite the availability of effective interventions, many PND cases remain undiagnosed, highlighting the need for more accessible screening methods. This thesis explores the potential of machine learning, mobile health (mHealth), and socially assistive robots (SARs) to enhance PND screening and diagnosis. It focuses on three objectives: (1) to develop predictive machine learning models for PND prediction using app data; (2) to explore stakeholder perspectives on SARs in PND care; and (3) to design an SAR prototype for PND screening and diagnosis and evaluate user acceptance and ethical concerns.

The first part of the research employed machine learning, which demonstrated that the occurrence of PND in the third trimester can be predicted using ecologically collected data. This highlights the potential of mHealth platforms for scalable early prediction of PND. The study also found that psychological health factors are the most predictive indicators for early PND.

The second part of the research explored the integration of SARs into PND screening and diagnosis through qualitative interviews with psychiatrists, PND experts, gender scholars, and mothers. The findings revealed diverse perspectives on the efficiency, acceptability, and design of SARs. Ethical concerns, like data privacy and informed consent, were also raised. While participants generally welcomed the potential of SARs to support PND screening, they consistently emphasized the need for human oversight. 

To translate high-level ethical guidelines into design practice, an exploratory case study developed SAR prototypes focusing on transparency and anthropomorphism. An online study examined how these traits influence the perception and efficacy of SARs. Results indicated that participants were more likely to agree with the transparent robot's recommendations. Findings from these studies, along with insights from subsequent participatory design sessions with mothers, informed the development of a real-world SAR prototype. User evaluations indicated that SARs are mostly welcomed in PND screening and can facilitate nonjudgmental interactions and emotional disclosure.

This thesis shows how machine learning, mHealth, and SARs can complement clinical care, offering effective, scalable, ethical tools to address gaps in perinatal mental healthcare.