Testing a Novel Design Framework for Patient-Facing Machine Learning-Based Predictions of Heart Failure...

Area of Science:

• Biomedical informatics
• Digital health
• Patient-facing technology

Background:

• Growing patient access to personal health data necessitates safe methods for presenting machine learning predictions.
• Cardiac implantable electronic devices (CIEDs) generate data suitable for predictive algorithms.

Purpose of the Study:

• To design and evaluate a patient-facing smartphone application prototype for displaying cardiac decompensation risk predictions.
• To assess patient comprehension, risk perception, and behavioral intentions regarding algorithm output.

Main Methods:

• Developed a design framework for algorithm output presentation in a smartphone app prototype.
• Conducted a mixed-methods usability evaluation with adult patients having CIEDs.
• Assessed primary endpoint of patient objective comprehension and secondary endpoints of risk perception and behavioral intention.

Main Results:

• High patient comprehension (80-85%) of algorithm output was observed across different risk change conditions.
• Comprehension of algorithm thresholds and contributing CIED sensors varied (60-93%).
• A significant risk change prompted worry in 70% of participants, with 80% intending to contact their doctor.

Conclusions:

• The prototype demonstrated high patient comprehension, appropriate risk perception, and intended actions.
• The findings suggest the viability of patient-facing applications for delivering actionable health insights from predictive algorithms.
• This approach can empower patients with CIEDs to better manage their cardiac health.