JMIR Publications

ABSTRACT

Background:

Unhealthy lifestyle behaviors have been identified as a major cause of numerous health issues with a steady global increase. Addressing this challenge requires comprehensive behavioral changes to promote sustainable healthier lifestyle adoption. However, despite the prevalent need, cost effective and successful health-related behavior change virtual coaching remain scarce.

Objective:

This paper presents a holistic framework for designing, modeling and executing behavioral change strategies, through a multi-agent reasoning system that selects optimal virtual coaching techniques based on individual assessments and integrates data-driven decision making.

Methods:

Behavioral change theories have been explored to design a multi-agent system that aims to achieve sustainable lifestyle changes. This system selects behavior change techniques (BCTs) based on individual user assessments, prioritizing those with the strongest impact on key behavioral components. The framework incorporates evidence-based practices stemming from behavioral change science and integrates them into Healthentia Behavioral Change Coaching (BCC) scheme. Healthentia, a certified software as medical device (SaMD), implements this framework specifically for chronic disease management, serving as an eHealth solution advancing decentralized care by enabling remote monitoring, data-driven content selection and personalized virtual coaching that adjusts to patient progress and engagement patterns.

Results:

This paper explores the application of the Healthentia BCC scheme in Type 2 diabetes patients. Behavioral attributes have been evaluated on 9 patients, yielding notable results in terms of fasting glucose dropping by an average of -17.3 mg/dL (Cohen’s d = 1.5, P = 0.002), further underscored by a narrow 95% CI [-26.1, -8.43], and in terms of weight and BMI, with mean reductions of -2.89 kg and -1.05 kg/m² respectively. These changes yielded large effect sizes (Cohen’s d ≈ 1.05) and were statistically significant (P = 0.01). The positive outcomes are at least partly attributed to the personalized delivery of content, 75% of which is well-received by the patients.

Conclusions:

Our study of this multi-agent system, tested through simulated patient behavior and a preliminary, limited Type 2 diabetes patient behavior observations, promises improved health outcomes using personalized virtual coaching strategies. Future directions include optimization of the multi-agent selection process, a further exploration of the Type 2 diabetes program and an in-depth evaluation of its results, including HbA1c measurements, and an applications’ expansion to other chronic conditions. Clinical Trial: Not applicable