In healthy individuals, glucose levels are maintained within tight upper and lower bounds because of a complex physiological closed-loop regulatory process based on the accurate and timely secretion of insulin and glucagon into the portal vein by pancreatic islet cells [1]. To achieve this, the human body contains numerous sensory mechanisms that track and even anticipate fluctuations in glucose owing to meal intake, exercise, and other factors. This information is used to estimate the optimal secretion of insulin as well as other hormones, such as incretins, glucagon, and adrenaline, which play important modulating roles [2].

In persons with type 1 diabetes (T1D), insulin secretion by the pancreas is absent because of the autoimmune destruction of the pancreatic beta cells, breaking the normal closed-loop regulation process [3]. The absence of insulin results in the inability to metabolize glucose and an unregulated catabolic state leading to hyperglycemia and ketoacidosis, a recognized complication of diabetes [4]. Persons living with T1D are dependent on exogenous insulin (usually by subcutaneous administration) to regulate their glucose levels [5].

Poor glucose control leads to both acute and chronic complications, which may be life-threatening [6]. One common acute complication is hypoglycemia, a state of low blood glucose (BG) concentration that results from excessive insulin administration. Hypoglycemia can lead to loss of consciousness, coma, and even death [7]. Those who develop impaired awareness to hypoglycemia (IAH), a condition in which the individual does not experience the usual early warning symptoms of low BG, are at a much higher risk of severe hypoglycemic events [8].

On the other hand, chronic hyperglycemia leads to long-term complications, one of the most common being cardiac autonomic neuropathy (CAN). The reported prevalence of CAN in persons with T1D spans a very wide range, indicatively 17%-90%, depending on the criteria used for its diagnosis and the population studied [9]. CAN results in impaired cardiovascular autonomic control as a consequence of autonomic nerve neuronal metabolic and ischemic damage. Apart from CAN, T1D is linked to other long-term complications such as peripheral neuropathy, retinopathy, nephropathy, and atherosclerotic vascular disease [6].

Maintaining glucose levels within a healthy range in T1D represents a therapeutic challenge, and the accurate estimation of an individual’s insulin requirements is key to achieving this goal [10,11]. Daily glucose management places a significant physical and cognitive burden on the persons living with T1D and their families. As part of ambulatory care, long and rapid acting insulins are administered using insulin injections (insulin pens) or rapid acting insulin is administered continuously and subcutaneously by insulin infusion pumps. Both insulin pumps and injections may be used in combination with continuous glucose monitors (CGMs). Insulin pumps integrated with CGMs are referred to as sensor-augmented pumps and their use has been shown to improve glycemic control [12,13]. More recently, control algorithms have been incorporated to automate basal insulin delivery; however, the user is still required to administer a bolus before meals. This treatment scheme is referred to as a hybrid closed-loop system. Hybrid closed-loop schemes have been shown to improve glucose control in comparison with insulin dosing entirely determined by the user [14] and have paved the way for the holy grail of T1D management, which is the development of fully closed-loop functionality or an artificial pancreas (AP) [15,16].

For optimal T1D management, accurate real-time sensing of key physiological parameters (ie, biomarkers) is essential. Sensing is crucial for daily acute glycemic management to ensure the correct estimation of insulin dose, as well as for the early recognition of long-term complications. To date, T1D sensing for acute glycemic management is focused on glucose monitoring, measured either by finger pricking or CGMs, which are associated with painful and sparse measurements or significant sensing lags, respectively [17,18]. However, monitoring of glucose alone has significant limitations in informing optimal insulin delivery, whether in open or closed-loop systems. In addition, monitoring of diabetes complications such as CAN is predominantly performed through infrequent assessments during clinic visits. Please refer to Multimedia Appendix 1 [9,17-23] for a more detailed discussion on the current status and challenges of sensing biomarkers of T1D. This survey, which reviewed 77 publications, investigated novel physiological parameters with the potential to be incorporated into T1D management systems and reduce the impact of T1D on quality of life, specifically on parameters that can be monitored with the wearable sensors available today.

Wearable sensors or wearables are becoming increasingly popular because they can provide seamless and continuous monitoring at low cost. Wearable sensors are available in various forms and shapes and can be worn at different body sites (Figure 1). Currently, wearables monitor physiological parameters, such as heart rate (HR), respiratory rate, oxygen saturation (SpO₂), skin temperature (ST), electrochemical skin conductance (ESC), and galvanic skin response (GSR). The most recent devices provide electrocardiogram (ECG) monitoring, which carries rich information about features of the cardiac state, such as HR variability (HRV) and QT interval (time interval from the start of the Q-wave to the end of the T-wave in an ECG).

This offers new potential for wearable devices to move beyond their original purpose of fitness and wellness monitoring to that of continuous health care monitoring. Although not yet integrated into clinical practice, research community is currently investing in the development of medical applications using wearable sensors that can assist or complement routine medical procedures and disease monitoring practices. These efforts usually combine wearable sensors with advanced data processing methods such as machine learning (ML) algorithms to address the volume and complexity of the produced data (noise, motion artifacts, and gaps) and to build smart decision support or diagnostic systems (Multimedia Appendix 2 [24-37]).

Despite this recent move toward health care, the potential of wearables in T1D management has not been investigated much to date. The purpose of this survey was to explore the potential of monitoring physiological parameters with wearable sensors to assist in acute glycemic management and diagnosis and monitoring of complications in T1D. We devised a search and analysis framework to investigate the potential of wearable, noninvasive sensors (hereafter referred to as wearable sensors without explicitly repeating their noninvasive property). To the authors’ knowledge, a similar survey has not been conducted thus far. We hypothesized that the demonstration of strong links between physiological parameters measurable with wearable sensors and T1D, combined with their intrinsic advantages for use, will support and inspire developments and research focused on the enhancement of real-life T1D applications.

The survey aimed to answer the following research question: Do the existing wearable, noninvasive sensors have the potential to improve how T1D is monitored and managed? We focused only on noninvasive wearable sensors. Minimally invasive wearables (eg, CGMs) or noninvasive portable sensors (eg, breath sensors) were outside the scope of the survey. We did not limit our search to studies that used wearable sensors but rather formulated our research question toward exploring the existence of clinical relationships between T1D and physiological parameters that could be (but not necessarily) monitored with the wearable, noninvasive technology available in the market today.

As a first step, a list of physiological parameters that could be monitored with wearable, noninvasive sensors available in 2020 was compiled, based on a review of the wearable devices available in the market today. These parameters were used in the search query combined with keywords related to T1D (Textboxes 1 and 2).

Figure 2 illustrates the methodological steps we took to select the final set of articles, following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [38]. The details of the search and selection methodology are described in Multimedia Appendix 3 [38]. Following these steps, 77 articles were retained for analysis.

The 77 retained articles shared similar structures. Data were collected through a clinical protocol whereby each study collected two types of data: (1) phenotypic characteristics of the participants with T1D (eg, age, sex, duration of diabetes, and glucose and hemoglobin A_1c [HbA_1c] levels), with or without the presence of nondiabetic control participants and (2) physiological function of interest (eg, cardiac autonomic function), with one or more physiological parameters measured (eg, HRV for autonomic function and ESC for sudomotor function). Through the extraction of features and data analysis, every study explored the relationship between T1D phenotypic characteristics and the physiological function of interest. All articles referred to clinical studies involving human participants. Although the articles did not necessarily use wearable devices in their methods, they all measured physiological parameters monitorable with commercially available wearable sensors, as guaranteed by our exclusion criteria (please refer to Multimedia Appendix 3).

Our analysis of the retained articles evolved along the axis of this survey’s research question: Do the existing wearable, noninvasive sensors have the potential to improve how T1D is monitored and managed? The surveyed studies were clustered into the following two broad categories: (1) those that performed a comparison between a cohort of persons with T1D and a matched group of healthy control participants and (2) those that explored relations within a cohort of persons with T1D. The first category informed about the physiological functions and parameters that were altered in persons living with T1D. The second category explored whether the monitoring of these parameters could benefit the management of T1D and its complications. All articles included in this survey are tabulated in Multimedia Appendix 4 [39-114].

The data from many studies in this survey showed that a variety of physiological parameters (1) can differentiate persons with T1D from healthy control participants and (2) are associated with aspects of glycemic control, in particular hypoglycemia, and the presence of diabetes complications within cohorts of persons with T1D.

The most explored physiological functions were cardiac autonomic, cardiorespiratory control balance, and thermal homeostasis. The survey identified that monitoring of physiological parameters, such as HR, HRV, QTc, ESC, baroreflex sensitivity, and VO₂, as well as the autonomic, cardiorespiratory, and thermal homeostasis functions, can be used to identify differences between persons with T1D and healthy control participants. The most pronounced differences between the 2 populations were shown in aspects of cardiac autonomic function, including parasympathetic activity, vagal tone, and cardiac repolarization and depolarization.

With respect to noninvasive hypoglycemia prediction and detection, physiological parameters of the ECG, including HR, HRV, and QTc, have considerable potential to be leveraged. Apart from hypoglycemia detection, these physiological parameters, as well as ESC and ST, can be monitored in conjunction with glucose to compensate for the deficiencies in CGM, such as signal lag, which is vital to the enhancement of AP development.

The screening of long-term T1D complications can also be enhanced. The demonstrated relationship between CAN and HRV paves the way for its continuous, and at-home, assessment. Other complications such as retinopathy and diabetic peripheral neuropathy were shown to relate to cardiac autonomic function and CAN and their onset could be predicted through monitoring of these functions.

Most of the studies discussed in this survey followed conventional statistical analysis methods to assess the existence of correlations in their measured data. ML methods have been used in some studies for the recognition of hypoglycemia and near-future glucose prediction [72-74,77-80].

Most of the surveyed studies used conventional devices; however, the measurement of the above physiological parameters can be performed with the 2020 commercially available wearable, noninvasive sensors. Table 1 lists the identified physiological functions and parameters, together with examples of commercially available wearable sensors that can monitor them today. Only a few studies found in this survey used wearable sensors [64,66,77-82]. These studies mostly focused on the detection of hypoglycemia, glucose prediction, or improvement of CGM glucose measurement accuracy.

The existence of wearable technology that can perform this type of physiological parameter monitoring is a crucial first step in the confirmation of our survey hypothesis that wearables have the potential to enhance T1D sensing with richer information seamlessly and continuously toward improved daily management decisions, and mitigation of complications. In view of this potential, the challenges and perspectives of this endeavor are discussed further below.

An important parameter when considering the potential of wearables is the quality of the generated data and the ability to extract the required information from them. A clinical ECG setup offers much higher accuracy than a wearable bracelet. Wearable sensors must compromise accuracy for small size, low cost, and high autonomy. Moreover, their default use involves people undertaking daily activities, which introduce motion artifacts and data corruption. Although in most of the studies reported in this survey, the involved participants followed a specified protocol under the supervision of a clinical staff member, this condition cannot be guaranteed or controlled in a daily life setting. To this end, a one-to-one comparison between a medical grade device and its wearable counterpart would always be an uneven battle. However, the claim of wearable sensors is not to substitute medical grade devices but rather to take up the space where the latter cannot be used; that is, the space of at-home, daily life routine. This different use case offers the following critical advantage over medical grade devices: the massive generation of data [24].

Compared with a clinical study, a wearable monitoring scenario produces vastly larger volumes of data over much longer periods and during various conditions, such as sleep, physical activity, eating, resting, stress, and working. Although wearables compromise accuracy, they can offer a significantly better and more representative coverage of a human life’s spectrum. The big data return can compensate for data noise by leveraging redundancies and information fusion coming from different wearable sources. Long-term information, such as seasonality, might be revealed, which would not be feasible for short-duration clinical protocols. Finally, the possibility of monitoring a large number of participants opens the potential for population studies that are currently very difficult or very costly to perform. At the same time, efforts for disease management personalization can be significantly boosted by the increase in data availability per user. To this end, the usability of wearable sensors in health care should be considered in light of their space of function, the added value that this space can encompass, and the additional knowledge extraction possibilities that follow from their much larger volume of generated data.

In view of their high data quantity and complexity, wearable devices usually require advanced processing techniques for the harvesting of their data. Although classical signal processing techniques may be sufficient for the extraction of the HRV or QT interval from a medical grade ECG signal, ML strategies may need to be used to perform the same task on ECG data collected from a wearable device. At the same time, there is a requirement for increasingly complex solutions to support clinical judgment and decision-making to meet the current medical and user demands. To this end, wearable-based applications require a postsensing stage of complex data processing to drive usable and viable solutions. The development of processing techniques for information extraction and decision-making based on data generated by wearable sensors is a field that currently receives intensive attention and research. This manifests itself in most of the studies that propose the use of wearables in health care, as discussed in the Introduction and Results sections.

Despite the depth of research dedicated to the exploration of relations between T1D and other physiological functions, only a few studies found in this survey used wearable sensors. These studies, although still small in number, support the contention that off-the-shelf wearable devices could be readily used in T1D interventions. However, the adoption of this technology in the field is very slow. In addition to the data quality–related challenges, the complex relationships between these biomarkers and glucose regulation provide an additional hurdle. Further research is required for the development of models and simulations, design of treatment strategies for the inclusion of new inputs, and finally, conduction of further clinical trials to demonstrate the added clinical impact of the methods. Finally, the strict safety constraints that need to be guaranteed during the daily management of T1D render the validation of new methods very demanding. In view of these challenges, publicly available data sets have been released to support the research output toward the development of data analytics tools to incorporate information received by wearable devices into T1D interventions [116,117].

In summary, the major challenges in the adoption of wearable technology in the management of T1D and its complications are as follows:

Despite the hurdles, this survey makes the case that wearable sensors have a significant potential to enhance the life of those living with T1D. The identified links between physiological parameters that wearables can monitor and T1D can be used to augment the T1D sensing space and develop better management tools. The continuous monitoring potential and the abundance of generated data per person can assist in the personalization of interventions. At the same time, wearable sensors that provide seamless and noninvasive monitoring are expected to add a minimum sensing burden compared with other types of sensors, whereas the automation of management processes that today require the cognitive effort of the persons living with T1D (eg, insulin bolus calculation) or induce stress (eg, fear of hypoglycemia) is expected to lead to better quality of life and lower daily burden.

To harvest this potential, 2 main directions for future research can be identified. First, advanced data processing strategies need to be developed to extract the information obtained from the data collected through wearable sensors. This research direction is not specific to T1D, and T1D can profit from the research outcomes of every field (health care or other), which opts to use wearable sensors. Second, further simulations, models, and clinical studies need to be conducted to support the development of decision tools for T1D based on the data collected through wearables. This direction is T1D-specific and involves multidisciplinary collaboration among data scientists, engineers, clinicians in the field, and persons with lived experience of T1D.

This survey aims to bridge on one side the large volume of research dedicated to identifying correlations between T1D glycemic control and complications with measurable physiological functions and on the other side, the novel potential of wearable technology in medical applications. The bulk of existing work dedicated to wearable technology and T1D is related to glucose sensing and omits other biomarkers that can be readily and noninvasively monitored with the available wearable sensors. To the authors’ knowledge, a survey that explores the potential of wearable sensors in T1D has not been conducted till date. This is the first attempt to bring the fields of T1D and wearables together to highlight the potential of these sensors in the daily management of this disease and the mitigation of its long-term complications.

The main limitation of the survey was that the search was conducted based on a list of wearable-enabled biomarkers (Textbox 1), which may not be exhaustive. Additional knowledge models related to wearables and T1D may exist that were not identified. Moreover, the focus was only on noninvasive sensors and did not consider minimally invasive wearable technology, which can be another pathway to further enhance T1D diagnosis and management. Similarly, the survey focused only on wearable sensors and did not discuss noninvasive, nonwearable devices, such as breath or saliva sensors, which can also be highly advantageous in T1D (eg, breath acetone sensing).

Considering the wearable sensor boom and its gradual adoption in the health care domain, this survey aimed to investigate their potential impact on T1D, a chronic disease that affects millions of people worldwide and requires daily and costly management and care. The survey search strategy targeted the discovery of studies that examined the relationship between physiological functions or conditions measurable by wearable sensors and T1D. Our analysis showed that T1D greatly affects cardiac, cardiorespiratory, and thermoregulatory functions, and its impact can be readily observed through features of the ECG, such as HRV, QT interval, and T-wave, as well as skin properties such as ESC, temperature, and sweat profile. The effects of T1D on these functions manifest themselves at rest, overnight, during and after exercise, and during daily life activities. Importantly, they can be leveraged to improve the prompt detection of hypoglycemia, the efficiency of the AP, and the diagnosis of CAN and other complications.

Commercially available wearable technology exists for continuous, noninvasive monitoring of the above parameters. For the successful adoption of this technology in health care in general, and T1D in particular, several challenges still need to be resolved, such as issues related to motion artifacts and noise removal, accurate extraction of the features of interest, and development of decision algorithms for improved and safe disease management. Current research efforts are working toward advanced algorithmic solutions for the efficient processing of massive amounts of data produced by wearable sensors. Their promising results can pave the way for similar endeavors for T1D.