Insights from Implementing an AI-Enabled Peripheral Edema Monitor in UK-based Community Settings for Therapeutic Decision-Making

Purpose

This study evaluates the implications of deploying a telemonitoring device in the homes of patients with heart failure through family doctors.

Background

In recent years, a decline in in-person primary healthcare consultations has posed significant challenges, especially for elderly patients with heart failure, who often contend with multiple comorbidities. A notable concern is their lack of awareness regarding symptom changes, such as weight gain, breathlessness, or swelling of the feet.

Methods

A cohort of 122 patients from 11 primary care practices across the UK were equipped with connected weighing scales and an AI monitoring device, including an internet dongle when necessary. These patients primarily resided in socioeconomically deprived areas. The study focused on device acceptability, utilization rates, and the impact of generated alerts on clinical teams. Initially planned for a six-month duration, a majority of participants opted to extend their use of the AI-device.

Results

The deployment targeted high-risk patients with heart failure, characterized by complex health conditions and histories of non-adherence with self-checks.

Data availability was high, with usage statistics indicating:

Weighing scale usage (at least once = 60.7%; at least 16days/mth = 5.7%)

AI-device usage (at least once = 100%; at least 16days/mth = 71.3%)

This indicates the AI device's passive, automated nature may enhance data collection from traditionally disengaged patient groups. Acceptability was high, with few rejections and of the 40% of participants responding to usability surveys, 92% would likely recommend the device to a friend. Alert management did not significantly burden GP teams, with most alerts being relevant to heart failure or other critical conditions needing attention (COPD exacerbation, COVID-19 infection, lymphoedema etc.). GP feedback was notably positive, especially from practices with a dozen or more participating patients.

Conclusion

Despite the challenging chosen patient group (representative of the high-risk heart failure population in the area), data acquisition was excellent, allowing clinicians to get an overview of the patient's health status remotely. While the investigation was not a randomized controlled trial (RCT), anecdotal evidence suggests early problem identification without significantly increasing GP workload. Further research is essential before broader implementation. The pivotal trial for the AI-device is expected to take place in the US in 2025.