Introduction

Healthcare has seen significant advancements in the use of wearable biosensors for real-time monitoring of specific biological analytes [1]. Such technology opens the door to delivering more personalised and timely interventions, which are pillars of the precision health movement [2]. Precision health offers a plausibly more efficacious approach to traditional ‘one-size-fits-all’ public health interventions by delivering the right support, to the right individual, based on their biological, behavioural, psychological, and social determinants of health [3, 4]. While some limitations of precision health still need to be addressed, such as inequities in social, environmental and economic influences [5], providing timely feedback that is based on one’s biological state (“biological feedback”) has great potential to support changes in behaviours that meaningfully impact health-related outcomes [6].

Biological feedback is defined as “providing individuals with their biological data through direct communication (via an unblinded body-worn assessment device such as a heart rate monitor or a continuous glucose monitor [CGM]); or indirect communication (via health coaches, patient educators, or messaging systems) about biological data to support health behaviour change explicitly or implicitly for improving health-related outcomes” [7]. This form of feedback is distinct from the traditional mind–body technique of “biofeedback,” which provides feedback on one’s autonomic nervous system to treat health conditions [8, 9]. In our recent scoping review, we found over 750 randomised controlled trials (RCTs) that used biological feedback to support health behaviour change [6]. Results from our scoping review indicated that many of these interventions aimed to modify diet and physical activity behaviours based on data from glucose monitors, particularly among people with diabetes. Given the prevalence of interventions focusing on glucose monitoring, it is crucial to delve deeper into the role of biological feedback from CGMs, which are reshaping the way we understand and manage metabolic dysfunction.

In the rapidly evolving field of healthcare technologies, CGM stands out as particularly pivotal. In contrast to the intermittent data provided by traditional methods of self-monitoring of blood glucose with a glucometer, CGM offers the advantage of collecting real-time glucose data continuously, providing a comprehensive overview of glucose levels and trends. These data can be used to inform personalised behavioural and pharmacological interventions aimed at improving glycaemic control outcomes [10]. The significance of CGM is underscored by its dominance in the biosensor market [1]. CGM was initially introduced in 1999 as a diabetes management tool for people living with type 1 diabetes mellitus, reducing reliance on fingerpricks from glucometers [11]. Nearly a quarter-century later, CGM-based biological feedback is in use within a broader market, fuelling the rise of global digital health startups. These companies mainly target people without diabetes, people desiring weight loss, athletes, and health enthusiasts. Using advanced data analytics, individuals’ CGM data are integrated with their related behavioural, biological, and psychosocial data to offer real-time insights into how food, sleep, exercise, and stress impact their glucose trends with a goal of optimising health and performance.

Despite the increasing popularity of CGM as a health behaviour change tool, there is a paucity of literature characterising the use of CGM in behavioural intervention research [12, 13]. The use of CGM in research is diverse, with CGM wear periods ranging from a couple of days to several months, and includes variations in whether participants can view CGM data in real time, as well as differences in how this data is interpreted. This leaves a significant gap in the collective understanding of how wearable biosensors can be best employed to affect meaningful health behaviour change. As technology and healthcare continue to intersect, it is becoming increasingly essential to develop best practices that optimise the effectiveness of behavioural interventions leveraging these tools. Therefore, the objectives of this scoping review were to: (1) describe the patient populations, health behaviours, and health-related outcomes targeted by CGM-based biological feedback interventions, and (2) characterise the methods by which CGM is used as a behaviour change tool within RCTs aimed to support health behaviour change.

Methods

Overview

Our aims align with the indications for a scoping review, which include identifying what evidence is available and which knowledge gaps remain, investigating the methods of research conduct, and utilising the findings as precursor to the feasibility of a systematic review and meta-analysis; thus, justifying the scoping review approach [14]. The Joanna Briggs Institute Reviewer Manual [15] was used to guide the review methods. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews (PRISMA-ScR) checklist was followed [16]. The review was registered in Open Science Framework Registries (https://doi.org/10.17605/OSF.IO/SJREA) [17].

Search strategy, selection criteria and review management

We collaborated with a research librarian to devise a search strategy based on our prior scoping review of 767 RCTs utilising biological feedback to support health behaviour change [6]. The prior search was conducted in June 2021 with no limitation of publication date. Here, relevant subject terms and text-words were included to capture behavioural interventions that incorporated feedback and biological measures, including glucose monitoring. For the current review, we updated the prior search and added terminology specific to CGM. The full search strategy has been included as Appendix 1. The updated search strategy was applied to articles published through January 2024, with no limit on year of publication. The search strategy was modified for the following electronic databases: Ovid MEDLINE, Elsevier Embase, Cochrane Central Register of Controlled Trials, EBSCOhost PsycINFO, and ProQuest Dissertations & Theses Global. Bibliographies of 17 additional reviews were also searched, and relevant articles were retained. There were no restrictions based on language.

Records returned by the search were deduplicated using EndNote 20 (Clarivate Analytics, Boston, MA) and added to the literature review software, DistillerSR® (Evidence Partners; Ottawa, Canada) for screening and data extraction. An additional deduplication process (using artificial intelligence) was applied in DistillerSR® to confirm all duplicate records were removed. Retracted articles were additionally identified using EndNote 20 and removed.

A multistep process was followed to determine study eligibility based on the following inclusion criteria: human adults ≥ 18 years, primary analyses of RCTs published in a peer-review journal or as a thesis or dissertation, and have at least one study arm receiving CGM-based biological feedback to support a health behaviour change. First, two trained reviewers completed an independent, single-entry title and abstract screening phase for initial eligibility. An artificial intelligence feature within DistillerSR® was used to confirm no abstracts were erroneously excluded. Then, full text versions of initially eligible articles were retrieved. Two trained reviewers completed a full text screening phase in which the preliminary inclusion criteria were confirmed and the use of CGM data to promote behaviour change was determined. If the use of CGM was unclear from the full text, an in-depth review of the study protocols available from trial registrations or published protocol was conducted. Articles not available in English were translated using Google Translate. Double-data entry by two independent reviewers for the full text screening phase was used for quality assurance. Conflicts were discussed between the two reviewers and resolved. If a conflict could not be resolved by the two reviewers, a third qualified reviewer made the final determination.

Data extraction

Extracted data were selected based on the Taxonomy of Technology-Enabled Self-Management Interventions [18] and CGM-specific reporting guidelines by Wagner and colleagues [19]. Data were also consistent with the three active components of personalised interventions: (1) sensing, (2) reasoning, and (3) acting [20]. Sensing describes the input parameters (ie, glucose) needed for the personalised intervention and how the measurement is performed (ie, CGM) [20]. Reasoning refers to providing feedback that is based on the input data (ie, biological feedback), including personalised behaviour recommendations or disease management guidance. Lastly, acting refers to how the biological feedback is communicated to the consumer to promote behaviour change (e.g., the mode, channel, frequency, and timing) [20]. Based on these criteria, a data extraction form was developed within DistillerSR®. The data extraction form was piloted by the three reviewers and refined prior to use. Extraction items included bibliographic data, participant characteristics, study design, CGM characteristics and wear durations, and CGM use (Appendix 2). Information related to the study design and treatment of all study arms were extracted, for reference. Results of included RCTs were not extracted as a synthesis of findings was not the objective of our scoping review [14], hence a risk of bias assessment was not completed. Double-data extraction of the included full text articles was then performed by the two primary reviewers. When necessary and if available, previously published study protocols or protocol details from clinical trial registries were reviewed. Data that were unobtainable have been described as “unclear.” Conflicts were discussed between the primary reviewers and resolved. If a conflict could not be resolved, the third reviewer made the final determination. The extracted data in DistillerSR® was downloaded and cleaned in OpenRefine [21].

Results

The updated database search resulted in 5355 articles. After removing 1394 duplicates, 3961 articles were screened for eligibility. An additional 24 studies from our original scoping review, and 10 studies from citation searching, were screened. N = 31 eligible studies were identified (Fig. 1) [22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52]. Characteristics of the included studies appear in Table 1.

Fig. 1
figure 1

Preferred reporting items for systematic reviews and meta-analyses extension for scoping reviews (PRISMA-ScR)

Full size image
Table 1 Characteristics of included randomised controlled trials that use CGM as a behaviour change tool (N = 31)
Full size table

Characteristics of CGM-based health behaviour RCTs

Included RCTs were conducted in 14 countries across 4 continents with the United States being the most frequently cited location (n = 6/31, 19%), followed closely by South Korea (n = 5/31, 16%). As displayed in Fig. 2, the first included RCT was published in 2006, with almost half of the RCTs (n = 15/31, 48%) being published in the most recent three years (2021–2023). Included studies ranged in duration from 2–52 weeks (median 13 weeks, IQR 12–26). Most of the studies were two arm RCTs (n = 20/31, 90%), with two 3-arm studies (n = 2/31, 7%) and one 4-arm study (n = 1/31, 3%). The total number of study participants ranged from N = 14–300 (median 70, IQR 40–149).

Fig. 2
figure 2

Overview of CGM-based health behaviour RCTs: study duration, targeted population, and number of sensor days (2006–2024). This figure illustrates that CGM-based health behaviour RCTs are increasing in frequency, duration, and number of days participants were asked to wear CGM sensors from 2006 to 2024. Since 2020, the target population has started to include participants without diabetes

Full size image

Characteristics of the targeted populations

Out of the 31 studies, a majority (n = 20, 65%) included people with type 2 diabetes (T2DM). The remaining studies included people with pre-gestational or gestational diabetes (n = 6/31, 19%), type 1 diabetes (T1DM) (n = 4/31, 13%), overweight or obesity (without diabetes) (n = 4/31, 13%), and/or prediabetes (n = 1/31, 3%). Insulin use among study participants was mixed with n = 10/31 (32%) studies including both insulin users and non-users, n = 8/31 (26%) studies exclusively included non-insulin users, n = 6/31 (19%) exclusively included insulin users, and n = 3/31 (10%) studies did not specify participants' insulin use.

Design of health behaviour change interventions incorporating CGM

Targeted health behaviours were dietary intake (n = 27/31, 87%), physical activity (n = 16/31, 52%), and/or unspecified healthy lifestyle changes (n = 2/31, 6%). All the included studies were complex interventions (i.e., included multiple components) incorporating other behaviour change strategies in addition to CGM (n = 31/31, 100%). For example, one additional component present in most CGM-interventions was guidance (n = 28/31, 90%), delivered prospectively, in-real time, or retrospectively by a professional (diabetes educator (n = 7/28, 25%), researcher (n = 6/28, 21%), general healthcare provider (n = 5/28, 18%), healthcare specialist (n = 5/28, 18%), or unspecified provider (n = 5/28, 18%)) based on reviewing the participants’ CGM data. Prospective CGM-based guidance took place prior to the participants’ CGM wear period and involved a professional instructing participants on how to use their CGM glucose values to inform personalised dietary and physical activity changes. Real-time CGM-based guidance occurred during the CGM wear period. It used data generated from the CGM combined with physiological and/or behavioural data to generate intervention messages. Retrospective CGM-based guidance occurred after the CGM wear period, and involved a professional providing personalised recommendations for diet, physical activity, or unspecified therapy changes based on the participant’s CGM glucose values. In n = 6/31 (19%) of studies, participants received both prospective and retrospective CGM-based guidance. Most often (n = 19/31, 61%), participants received retrospective CGM-based guidance, while in n = 13/31 (42%), participants received prospective CGM-based guidance. In n = 3/13 (23%) of these studies, participants were instructed by a professional to follow a simple algorithm to make dietary or meal timing decisions based on the CGM-provided information. In two studies (n = 2/31, 6%) participants received real-time advice based on their CGM data. In the intervention arms, CGM was often combined with other intervention components that included health-related education (individual or group) (n = 20/31, 65%), diet tracking (n = 15/31, 48%), physical activity tracking (n = 11/31, 35%), and/or medication tracking (n = 5/31, 16%).

The comparison arms (N = 35) commonly included health-related education (n = 20/35, 57%), the use of a glucometer (n = 19/35, 54%), and/or diet tracking (n = 9/35, 26%). In seven comparison arms (n = 7/35, 20%), participants wore a CGM and received biological feedback; the distinguishing factors between the intervention and comparison arms were either the additional intervention components that were offered alongside CGM, and when the biological feedback was delivered (i.e., in real-time versus retrospectively). One study was a three-arm crossover trial, where all participants received 14 days of unblinded CGM and were randomised based on the order in which they consumed three standardised mixed dishes, varying in glycemic indices [28].

Characteristics of CGM device and wear

CGM manufacturer was specified in most studies (n = 27/31, 87%). Abbott (n = 14/31, 45%) was most frequently used, followed by Medtronic (n = 9/31, 29%), Dexcom (n = 3/31, 10%), and A. Menarini Diagnostics (n = 1/31, 3%). The Abbott Freestyle Libre (n = 12/31, 39%) was the most commonly used model of CGM. Single CGM wears ranged from 2–14 days in duration, depending on the manufacturer (Medtronic = 2–10 day wears, Abbott = 10–14 day wears, Dexcom = 7–10 day wears). Across the reporting studies (n = 30/31, 97%), the number of sensors worn ranged from 1–18 (median 3 wears, IQR 2–6), which resulted in a total number of CGM wear days of 2–252 days per intervention (median 28 days, IQR 14–63). For studies with multiple CGM wears (n = 24/30, 80%), CGM was worn continuously during the intervention in n = 11/24 (46%) studies; whereas, in the other n = 13/24 (54%) studies, participants wore CGM intermittently (median 3 wears, IQR 2–4) with breaks between wears (median 5 weeks, IQR 4–11).

Communication of CGM-based biological feedback

The communication of CGM-based biological feedback varied by whether CGM data were made visible (“unblinded”) or not visible (“blinded”) to participants during the CGM wear(s), and whether one-way (e.g., via one-way email) or two-way (e.g., via in-person discussion) delivery of CGM-based biological feedback was provided (Fig. 3). There were 3 predominant forms of communication: (1) via unblinded CGM device with one- or two-way communication (n = 17/31, 55%); (2) via blinded CGM device with one- or two-way communication (n = 6/31, 19%); and (3) via unblinded CGM device without one- or two-way communication (n = 7/31, 23%). One study was unclear about blinding but did provide two-way communication.

Fig. 3
figure 3

Delivery of CGM-based biological feedback in behaviour change interventions l (N = 31). This figure illustrates how studies delivered CGM-based biological feedback. The size of the band indicates the number of studies. “CGM blinding” describes whether CGM data were visible (unblinded) or were not visible (blinded) to a study participant in real-time during the CGM wear period(s). “Mode”, “Channel”, “Frequency”, and “Timing” are specific to how CGM-based biological feedback was communicated. “Frequency” was calculated by the number of one- or two-way feedback sessions divided by the number of sensors worn. “Unclear” was used when the study protocols did not provide related information. From this figure we can see that the plurality of studies used unblinded CGM, with device and two-way communication, which was usually in-person, at a frequency of 1 communication session per CGM sensor, which was provided after CGM wear

Full size image

There was variability—and occasionally a lack of clarity—in how the feedback was conveyed to participants in terms of the mode, channel, frequency, and timing. Most commonly, when reported, CGM-based biological feedback was provided by the mode of CGM device and two-way communication (n = 12/31, 39%), through two-way communication alone (n = 7/31, 23%) or device alone (n = 7/31, 23%). Two-way communication was most often delivered in-person (n = 13/31, 42%) and/or over the phone (n = 6/31, 19%), and typically occurred after CGM wear (n = 19/31, 61%), once per CGM wear (n = 13/31; 42%). All feedback for one- and two-way communication was delivered by a human, as opposed to automated feedback (digital or artificial intelligence).

Targeted biological, behavioural and psychosocial outcomes

Multiple biological, behavioural, and psychosocial outcomes were reported in the included RCTs (Table 1). Biological outcomes were reported by all included studies and were often the primary outcome(s) (n = 25/31, 81%). Change in HbA1c was reported as an outcome in a majority of studies (n = 29/31, 94%). Other commonly reported biological outcomes were anthropometry (n = 18/31, 58%), time in range (n =16/31, 52%), hypoglycemia (n = 15/31, 48%), mean glucose (n = 11/31, 35%), lipids (n = 10/31, 32%), standard deviation of mean glucose (n = 9, 29%), and fasting glucose (n = 9/31, 29%). Seventeen studies (55%) included behavioural outcomes, which were most frequently diet (n = 11/35, 32%), physical activity (n = 10/31, 32%), and diabetes self-care (n = 5/31, 16%). Eight studies (n = 8/31, 26%) included psychosocial outcomes, including depression/anxiety (n = 6/31, 19%), and diabetes distress (n = 4/31, 13%). Six studies (19%) included intervention feasibility and acceptability as an outcome.

Discussion

As we enter the precision health era, biosensors like CGM exemplify how biological feedback can potentially revolutionise health behaviour change interventions. To our knowledge, this is the first review to comprehensively explore the characteristics of CGM-based interventions that use biological feedback to support health behaviour change. We found that a significant portion of the included studies were published recently, with nearly half (N = 15/31, 48%) published within the last 3 years, indicating considerable growth of the CGM evidence base. Most studies involved people with T2DM and assessed HbA1c as an outcome. All were complex, multi-component interventions, often combining CGM with prospective or retrospective guidance; health-related education; and diet, physical activity, or medication tracking. CGM-based biological feedback was often delivered through in-person discussions after wearing CGM. These detailed understandings of CGM interventions—how they were operationalized, what they involved and what they targeted, alone and in combination with other behaviour change components—is an important first step to systematically understanding the relationship of these various elements with intervention effects.

The first objective of this review was to provide an overview of patient populations, health behaviours, and health-related outcomes associated with CGM-based biological feedback interventions. We found a lack of RCTs investigating the benefits of using CGM for behaviour change among individuals without diabetes, despite interests in this application of the technology in the digital health market. Nevertheless, research in this area appears to be on the rise, with four RCTs investigating the use of CGM-based biological feedback in individuals without diabetes since 2020, and one RCT including individuals with prediabetes published in 2023. CGM interventions primarily targeted diet and physical activity, aligning with general biological feedback [6], and precision health interventions [53]. Most interventions assessed HbA1c as an intervention outcome, likely due to the prevalence of diabetes in the studies. Future research should explore CGM's impact on other health biomarkers (e.g., weight, CVD risk factors), potentially benefiting individuals without diabetes. This research could provide a scientific basis for the goals of digital health startups focusing on outcomes like weight loss and chronic disease prevention.

The second objective of this review was to describe how CGM is used in biological feedback interventions. In most of the reviewed RCTs, CGM-measured glucose levels were used as input to generate guidance to improve healthy lifestyle behaviours, often through retrospective feedback by professionals on diet, activity, or disease management plans. However, there was considerable variation in how CGM-based feedback was delivered to participants, including differences in mode, channel, frequency, and timing. The noted variability in communication has been observed previously in another context [54] and may vary depending on the population, biomarker, and targeted outcome [6, 55]. More recent studies have provided CGM-based biological feedback from an unblinded CGM over longer durations, and have incorporated the use of one-way communication (e.g., via a mobile app). Nevertheless, the delivery of CGM-based guidance was mainly reliant on human interaction versus artificial intelligence. Consistent with precision health literature [53], a majority of personalised feedback in the present review relied on human interaction for developing and communicating CGM-based guidance. Despite human interaction being potentially more effective in achieving health outcomes [56], limitations like cost, availability, and reach limit widespread use. This highlights a potential research gap and opportunity for more novel approaches, such as artificial intelligence, to be integrated into mobile platforms to automate the delivery of meaningful, personalised biological feedback. An example of this was showcased in a recent RCT, where Guo and colleagues instructed intervention participants with T2DM to use a mobile app, which used artificial intelligence to analyse and integrate unblinded CGM data and participant self-reported diet and activity data to provide personalised feedback on foods and exercises that were least and most beneficial for the participant’s personal glucose management [33].

The main strength of this review was our application of a systematic method to capture and characterise CGM-based biological feedback interventions in unprecedented detail. This thorough mapping provides a starting point for further examination of individual intervention components and their impact, paving the way for inventive intervention designs. However, there are limitations. Our inclusion criteria focused only on RCTs and adults, with the purpose of laying the groundwork for a future meta-analysis of study effects based on commonly targeted outcomes (e.g., HbA1c) identified through this review. Some studies lacked clarity in how CGM was used and how intervention components were implemented, which we addressed by searching for protocols, corresponding with authors, and conducting a thorough search of clinical trial registries.

To our knowledge, this is the first scoping review to describe how CGM is used within interventions that promote behaviour change. Despite the burgeoning interest in CGM and its application in the digital health market, academic evidence supporting the use of CGM-based interventions for behaviour change is mostly limited to people living with diabetes. To advance CGM-based precision health interventions, collaboration between academia and industry will be crucial. This collaboration can expedite the translation of research to real-world applications, enabling more effective data-driven interventions.

Based on the findings of this scoping review, we have identified a substantial body of literature on the effects of using CGM as a tool for biological feedback to reduce HbA1c levels. We plan to evaluate these effects in a subsequent meta-analysis (CRD42024514135). In addition to this, given the multi-component nature of these interventions, we plan to further investigate the behaviour change techniques that accompany CGM-based biological feedback interventions, with the long-term goal of identifying optimal combinations of behaviour change techniques to offer in combination with CGM to improve health outcomes (CRD42023398390). These future directions underscore the importance of our review, which serves not only as a current snapshot but also as a foundational resource for upcoming research efforts. This review has the potential to guide the design of future research to determine best practices for implementing CGM-based precision health interventions and contribute to guidelines for precision health interventions using biological feedback. Best practices can address key aspects such as the duration and frequency of sensor wear, communication of CGM data, and behaviour change techniques to deliver alongside CGM-based biological feedback. As biosensors like CGM play an expanding role in healthcare, rigorous evaluation is essential to inform public health and clinical guidelines.