Cardiometabolic diseases (CMDs) are a group of interrelated conditions, including heart failure and diabetes, that increase the risk of cardiovascular and metabolic complications [1]. The number of Australians managing CMDs is increasing, necessitating new approaches to better manage these conditions [2,3]. Approximately one-quarter of people globally are estimated to be living with metabolic syndrome and approximately 1 in 13 are living with cardiovascular disease [4,5]. The self-management of CMDs requires individuals to adhere to treatment regimens such as taking prescribed medications, maintaining a healthy diet, and performing regular physical activity to effectively manage their condition [6]. As these lifestyle modifications can be complex and ongoing, digital health interventions have emerged as an integral part of a scalable and accessible strategy to support individuals self-managing CMD [7,8].

In CMD contexts, the prolonged course of illness, often spanning decades, increases the difficulty of sustained engagement with self-management, where the risk of relapse remains a significant concern over time [9,10]. Therefore, it is essential that interventions aiming to improve self-management must be able to sustain user engagement over long time periods [10]. Indeed, low engagement rates undermine the effectiveness of digital health interventions; a recent review found an average participant dropout rate of 43% in studies assessing digital health interventions applied to the self-management of chronic diseases [11].

A recent systematic review identified improving personalization and interactivity as effective approaches to improving long-term engagement rates of digital health interventions [12,13]. This is particularly crucial in managing CMDs, where ongoing engagement and sustained behavior change are necessary for the effective attenuation of CMD complications [10]. As components within digital health interventions, conversational agents could greatly enhance the effectiveness of these self-management interventions by contextualizing information, offering constructive feedback, and fostering critical reflection [14-18]. A conversational agent is a technology that interacts with people using natural language, whether text-based or spoken, enabling accessibility to broader populations, including people with motor or cognitive disabilities [19]. Use of conversational agents could also reveal a linguistic dimension to a person’s health status, which could improve the personalization and effectiveness of digital health interventions [16,20].

Within the emerging research field of conversational agent–enabled interventions, the majority of existing tools are text-based, driven by machine learning algorithms, and delivered through mobile apps [21]. The literature on conversational agent–enabled interventions that support individuals with chronic diseases, including CMDs, to manage their condition is mainly composed of qualitative studies, quasiexperimental studies, pilot tests, and a limited number of randomized controlled trials, which all primarily examine prototypes of these interventions [22-25]. In clinical contexts, mental and physical wellness are the primary domains for the application of conversational agents, where they often provide emotional support [26,27]. In contrast, conversational agents applied to CMD self-management tend to provide a modality for monitoring symptoms and to deliver patient education [26]. However, recent trends indicate a shift from these basic functions to more complex and long-term end points in future applications, such as motivating behavior change [21]. This shift has occurred in parallel with technological advancements in cloud computing and transformer models, which have enabled a new generation of conversational agents termed large language models.

Large language models use deep neural networks to handle complex language tasks such as summarizing, generating, and translating natural language [28]. These tasks are achieved not by large language models understanding the prompts but rather by repeatedly predicting the word that is statistically the most likely to follow a sequence of words given as a prompt until a full response is offered [29,30]. However, this predictive approach means that large language models are potentially unsafe in clinical contexts, as they are liable to misinterpret prompts or generate unforeseen and inaccurate content [31-33]. Additionally, the opacity of this predictive approach hinders the ability to anticipate all possible outputs, making it challenging to establish safeguards that effectively prevent the dissemination of harmful or misleading information [34,35]. Furthermore, integrating these models into health care systems can involve transmitting sensitive, patient-identifiable data to third-party servers for processing, thereby imposing significant legal and ethical risks on health care providers [36,37].

Evidence indicates that the design features of conversational agent–enabled interventions impact the engagement of people with mental health issues in managing their condition [38,39]. For example, studies indicate that specific design features such as enhancing the anthropomorphic qualities of conversational agent–enabled interventions can improve engagement by reducing the monotony of repeated interactions [38,40]. Additionally, a recent review explored how design features pertaining to conversational architecture, such as delaying responses, proactive dialogues, and self-disclosures, affected user perceptions of conversational agents [41]. However, to date, there has not been a synthesis of evidence pertaining to conversational agent–enabled intervention design features (eg, personality) and their impacts on the engagement of people managing CMDs. In filling this gap, this review will contribute to the broader developer community making better informed choices when developing conversational agents, leading to more engaging, and therefore effective, conversational agent–enabled interventions for CMD self-management.

The review was initiated in May 2023 and, prior to title and abstract screening, was registered with PROSPERO in June 2023. Full-text screening of articles was completed in August 2023, followed by data extraction. Final searches were conducted in April 2024 (Figure 1) and the completed review was submitted to Digital Health in July 2024. Data pertaining to study design and sample size have been extracted and are summarized in Table 1.

The effectiveness of digital health interventions in supporting people with CMD is dependent on the extent to which users engage with the digital tool [17,18]. Augmenting digital health interventions with conversational agents, technologies that interact with people using natural language, may enhance engagement because of their human-like attributes [24,25,57]. This protocol outlines a systematic review that aims to synthesize evidence pertaining to conversational agent–enabled intervention design features and their impacts on the engagement of people managing CMDs. The anticipated outcomes of the analysis include the identification of specific design features or themes within various domains that improve engagement with digital health interventions. Additionally, the quality appraisal process is expected to uncover research gaps on conversational agent–enabled interventions, thereby providing a clearer direction for future studies applied to investigating the design of conversational agent–enabled digital health interventions.

This synthesis will provide guidance on how best to embed engagement strategies within these interventions, thereby facilitating more engaging and effective strategies for supporting people managing CMDs. Additionally, characterizing the literature, in terms of how engagement tends to be reported, will have the benefit of identifying research gaps and highlighting areas for future exploration. These outcomes can in turn support researchers in developing a greater understanding of user engagement with conversational agent–enabled interventions.