Incorporating end-user perspectives into the development of a machine learning algorithm for first time perinatal...

Area of Science:

• Artificial Intelligence in Healthcare
• Mental Health Technology
• Clinical Decision Support Systems

Background:

• Machine learning (ML) algorithms offer potential for advancing clinical care, particularly in identifying mental health conditions like perinatal depression.
• Current ML algorithm development often overlooks the crucial perspectives of the populations they are intended to serve.
• Integrating end-user insights is vital for the successful development and implementation of healthcare technologies.

Purpose of the Study:

• To describe the process of incorporating end-user perspectives into the development and implementation planning of a prediction algorithm for new perinatal depression onset.
• To ensure the developed algorithm is interpretable, complete, and acceptable to both healthcare providers and patients.
• To inform the clinical implementation of a patient-facing screener derived from the prediction algorithm.

Main Methods:

• Conducted a focus group with 12 providers and four virtual community engagement studios with 21 patients.
• Presented the initial development of a novel prediction algorithm for detecting first-time perinatal depression.
• Utilized rapid qualitative analysis to code algorithm completeness, interpretability, and stakeholder acceptability.

Main Results:

• Providers and patients reached consensus on algorithm interpretability and suggested additional predictive variables.
• Patients desired discussing screening results with providers, who raised concerns about limited bandwidth.
• Both groups emphasized the need for post-screening resource connection, noting concerns about resource availability.

Conclusions:

• Qualitative findings from end-users were integrated into iterative algorithm development.
• The study informed the implementation pilot plan for a perinatal depression prediction algorithm.
• Incorporating end-user expertise can enhance the clinical adoption of risk prediction algorithms.