1 Introduction

Diabetes Mellitus (DM) is among one of the leading worldwide epidemic metabolic disorder that imposes inadmissibly significant human, social, and economic costs. Despite all the recent advancements in drug therapies, technologies, healthcare education, and preventive intervention strategies, the prevalence of diabetes is still on the rise and its associated health complications have become even more prominent and life-threatening [1]. Globally, 425 million people have been affected by diabetes and around 693 million people are expected to be affected by 2045 if the current trends continue. Moreover, USD 727 billion is being spent yearly by diabetic people in their healthcare treatment and approximately four million people have lost their lives due to diabetes in 2017. The total healthcare expenditure has increased from USD 232 billion in 2007 to USD 727 billion in 2017 and the economic burden is expected to increase to USD 776 billion by 2045 [2].

The high prevalence of diabetes is due to unmanaged and uncontrolled diabetes levels among patients. Despite no cure for diabetes has been found so far, the symptoms can be alleviated, and complications can be reduced to a great extent. In addition, treatment efficiency can be improved significantly through continuous glucose monitoring coupled with proper medication, dietary habits, and physical exercise.

Ongoing monitoring of blood glucose (BG) level has become imperative in the management and treatment of diabetes. Patients can perform self-monitoring of blood glucose (SMBG) to control their insulin levels, to improve adherence, and to keep BG levels within a normal range. Thus, the mortality and associated healthcare complications of diabetes will be better controlled. Very often, failure to manage diabetes results into lower quality of life, increased economic burden, and social problems [3, 4]. Consequently, societal costs related to hospitals, readmission rates and hospital visits with cases of hypoglycemia/hyperglycemia can be reduced with continuous monitoring. Therefore, there is an increasingly high need for cost-effective healthcare services that can be provided to everyone, everywhere, and anytime ubiquitously to support and monitor patients to avoid expensive hospital-based care [5].

The evolution of Pervasive Healthcare Systems (PHSs) is a promising potential for unobtrusive remote health monitoring anywhere and anytime. It empowers the patients with the ability to detect their symptoms at an earlier stage and to easily share their medical information with healthcare professionals for further real-time analysis, diagnosis and timely intervention without spatial–temporal restrictions. In case of emergencies, the concerned parties can be alerted through message or email notifications. Consequently, the wide usage and adoption of PHSs can lead to better self-disease management, proactive monitoring of conditions and significant reduction in healthcare economic burden and hospital visits.

PHSs include WBAN which is an emerging technology involving the use of wireless communication whereby low-power, intelligent, small-sized, lightweight, and invasive or non-invasive sensors function in the vicinity of the human body to detect the patient’s vital signs. WBAN is gaining major interest with the widespread plethora of available technologies supporting medical and healthcare applications [6, 7]. An increase in prevalence of diabetes is leading to a worldwide paradigm shift from doctor-centric to patient-centric whereby WBANs have a fundamental role in addressing the multifarious challenges in healthcare [8]. The recent advancements and evolution in WBAN have demonstrated the huge potential to improve the quality of life of diabetic patients [9]. Nevertheless, these WBANs require high level of data reliability and quality to be effective and widely adopted.

Alarming reasons such as high societal and economic burden are triggering various research works in the field of healthcare especially in continuous monitoring of diabetic patients using WBAN. Furthermore, the benefits of the WBAN are beyond argument as they have the capability to improve the health conditions and lifetime expectation of diabetic patients.

This chapter presents an overview of the current BAN for monitoring patients by focusing on the traditional systems and continuous glucose monitoring systems (CGMs). The challenges of CGMs are further described and some leading and latest CGMs on the market are presented. It also discusses recent advances in sensor technologies and some emerging devices for non-invasive glucose monitoring via other human fluids than blood. This chapter also outlines the need for data quality and reliability in BAN for diabetes monitoring. The different metrics to measure the dimensions of Quality of Information (QoI) are presented and the research directions in BAN with regard to data quality and reliability are discussed at the sensor level, network level and human-centric level.

2 Body Area Networks for Diabetes Monitoring

2.1 Traditional Systems

The gold standard way for diabetic patients to perform SMBG is through a blood glucose meter, an uncomfortable and slow process. A small blood sample (<1 μL) is retrieved by a pricking the finger using a lancet and the sample is transferred immediately to a test strip. The latter is then put into a blood glucose meter to provide the glucose readings. This traditional method is invasive and painful, and thus results in poor patient compliance. It is also inconvenient and causes skin injury since the patients need to draw out their blood several times per day. A recent report revealed that over 65 blood glucose meters is currently available off-the shelf in UK which vary in size, weight, accuracy, response time, storage capacities and special features [10].

Besides the conventional methods, there are other alternative glucose monitoring technologies such as continuous glucose monitoring systems (CGMs) , invasive, minimally invasive and non-invasive glucose monitoring systems [11].

2.2 Continuous Glucose Monitoring Systems

CGMs uses Food and Drug Association (FDA)-approved glucose sensors placed into the human body, which detect the BG levels in the interstitial fluid. CGMs can improve glycemic control of patients through continuous and automatic monitoring without having to do finger pricks. CGMs usually provide recordings of at least 72 h with the possibility of obtaining up to 288 BG level readings/day. The CGMs’ glucose sensor is implanted right beneath the skin and is connected to a transmitter that continuously sends the data to a receiver. The latter is a small external monitor usually built in to an insulin pump to monitor BG levels and take necessary actions in case of anomalous readings. These devices provide in depth information about glucose such as frequency and duration of hypoglycemia/hyperglycemia on a daily basis.

CGMs’ benefits are beyond argument and have proven to be very effective in providing instantaneous and real-time BG readings to patients. These information and trends are very helpful to both patients and healthcare professionals to perform retrospective analysis, detection and prediction of hypoglycemia/hyperglycemia [1]. Moreover, CGMs empower patients and give them insights to proactively manage their diabetes 24/7. Consequently, the glycated hemoglobin (HbA1c) and risk for hypoglycemia/hyperglycemia are revealed to be significantly reduced even though the patients are on insulin injections or pump therapy [12]. CGMs can notify, educate, encourage, and alert people with diabetes [13].

2.3 Challenges of CGMs

While the numerous benefits make the use of CGMs a highly remarkable method for diabetes management, significant limitations remain and adoption of CGMs is fairly limited. CGMs are usually quite expensive and it is challenging to estimate the total costs associated with them since the pricing systems and indirect costs are too diverse. Patients are thus reluctant to invest in such costly systems due to poor reimbursement facilities as insurance companies throughout the world do not cover and exclude CGMs in the health plans [13]. Consequently, user acceptance and adoption is bound to improve as health insurance companies approve reimbursement of CGMs. Furthermore, the CGMs’ technical advances and functionalities are evolving so fast that the current ones are becoming obsolete rapidly. Thus, patients have to perform new hardware/software upgrades frequently or they have to rely on the previous versions’ support.

The glucose sensors’ lifetime is another barrier as the sensors have to be disposed and replaced approximately within 7–14 days. The frequent need for recalibration and periodic replacement of sensors usually contribute to other factors such as inconvenience, slow user acceptance, cumbersome process, discomfort, cost and toxicity [14]. New research works are highly needed to increase the sensors’ battery lifetime to promote user acceptance and better adoption of such systems. Additionally, CGMs manufacturers can incorporate new technical advances such as pre-calibration so that patients are free from this hassle.

Accuracy and precision of readings are other major concerns to be considered for the wide adoption of CGMs. Interfering substances in the interstitial fluid can lead to erroneous glucose readings and false alarms which can have significant impact on the self-adjustment of insulin dosage and particularly the detection of hypoglycemia/hyperglycemia [15]. Consequently, patients are most likely to adjust hyperglycemia and hypoglycemia through insulin boluses and carbohydrate doses respectively. This will ultimately increase the risk for hypoglycemia/hyperglycemia. Thus, user acceptance will definitely rise as the accuracy margins of CGMs improve. Furthermore, sensors should be calibrated when BG levels are steady to output the most accurate, reliable, and precise readings.

CGMs sensors generate enormous amount of data, 288 readings per day and 2016 records per week per patient. Besides storage issues, analysis and interpretation of vast amount of data are quite challenging and complex since not all healthcare professionals are well trained or have systematic approaches. In addition, the healthcare professionals already have a huge amount of duties and might have insufficient time to get acquainted and familiarize themselves to the new CGMs [16]. Thus, patients are often left on their own to make their analysis, interpretation, and decisions. It is of paramount importance that both patients and healthcare professionals attain a decent level of numeracy and literacy with regard to the use and interpretation of the results for better adoption and usage.

Some studies have also shown that high rates of CGMs discontinuation and patients’ reluctance to adopt such systems are due to some prominent factors such as discomfort, problems to insert the sensor, sensitivity issues with the adhesive, improper functioning of the sensor, too many alarms, inaccurate CGM data, interference with sports activities, and skin reactions [16, 17]. These factors are of great concern to patients and they need to be considered and addressed when developing the CGMs.

2.4 Existing Continuous Glucose Monitoring Systems

Some leading CGM systems on the market that are widely used in clinical practice are (1) DexCom G5 from DexCom Inc., (2) MiniMed® 670G from Medtronic Inc., and (3) FreeStyle® Libre Flash from Abbott Diabetes Care.

DexCom G5, DexCom Inc. (San Diego, CA) . Recently, DexCom G5, the first mobile CGMs has been recently approved by FDA in 2016 to be used by persons 2 years and older [18, 19]. It includes a discrete sensor that is implanted right beneath the abdomen’s skin with a 7-day sensor lifetime. The sensor is attached to a transmitter that continuously sends data directly to either a handheld receiver or a mobile phone every 5 min to display the glucose readings through Bluetooth® wireless technology. However, for proper functioning, the sensor/transmitter and receiver/mobile phone should be within a range of 20 ft. It takes a 2 h warm up time to send the glucose readings after the sensor has been inserted and the latter needs to be calibrated every 12 h. The BG levels should be within a range of 40–400 mg/dL for proper calibration. The system offers customizable high and low alarms as well as glucose rate of change alarms to alert the patients in case of hypoglycemia/hyperglycemia.

The sensors/transmitters are water resistant for up to 8 ft deep for 24 h and are suitable for water sports. However, the receiver is not water resistant. Several events such as insulin, carbohydrates, blood glucose, exercise, and health events can also be captured to provide a comprehensive representation of the possible reasons of glucose excursions. Nevertheless, the system is susceptible to incorrect readings due to medication such as acetaminophen [20]. Furthermore, it is advisable not to use the system during Magnetic Resonance Imaging (MRI), Computed Tomography (CT) scan, or high-frequency electrical heat (diathermy) treatment since magnetic fields and heat could damage the sample leading to erroneous values [18].

DexCom G5 works with only one insulin pump, the T: slim X2™ Insulin Pump, and can be used by diabetic patients requiring insulin infusion [21]. The system comes with the DexCom CLARITY® software to generate necessary reports, compare glucose data based on customized dates and analyze trends. However, the CGM system is only compatible with iOS or Android-enabled devices. Additionally, it consists of DexCom’s Follow app on iOS or Android-enabled device whereby up to five relatives or healthcare professionals can remotely monitor the glucose data and analyze the trends in real-time .

MiniMed ® 670G, Medtronic Inc. (Northridge, CA) . The Medtronic MiniMed® 670G is the first hybrid closed loop system that has been approved by the FDA in 2016 to track BG levels and automatically adjust the delivery of basal insulin. The system uses SmartGuard® HCL technology and should be used as a supplementary device rather than replacing the readings gained from blood glucose monitoring devices. The system should be used by diabetes type 1 patients 14 years and older. It also includes the MiniMed 670G Pump and a subcutaneous sensor (New Guardian Sensor 3) that must be calibrated by patients two times per day before meals and at bedtime. It is FDA-approved for 7-day wear. The sensor warm-up time may vary from 40 min up to 120 min before BG values are obtained, after which the readings are shown every 5 min.

The system has two main modes, manual and the auto mode. In the “auto mode,” the pump automatically adjusts the basal insulin delivery based on the glucose values obtained every 5 min from the device. The insulin delivery is continuously increased, decreased, or suspended, targeting a BG level of 120 mg/dL. However, the patient should still manually input the carbohydrate information entries and should manually deliver insulin therapy during meals. In the “manual mode”, the basal insulin is delivered at a constant rate as programmed by the patients. It consists of a “suspend before low” option where the delivery of insulin is temporarily suspended when the BG value falls to or approaches a preselected low-glucose limit. Then, it automatically restarts the insulin delivery once the BG value increases or is predicted to increase above a pre-selected limit.

The system can resist water up to 12 ft deep for approximately 1 day and is thus suitable for water sports. The information can be transferred via radio frequency communication to the Medtronic CareLink software which is web-based and works practically on all operating systems. The sensor data with pump, meter data and generated reports can easily be shared with the healthcare professionals .

FreeStyle ® Libre Flash, Abbott Diabetes Care (Alameda, CA) . The Abbott Diabetes Care FreeStyle Flash is the first FDA-approved CGMs in 2017 to be used by diabetic patients of 18 years or older. As compared to all existing sensors, this system is pre–calibrated in the factory and does not require any finger-prick calibration thereby reducing all the discomfort often experienced by patients [22]. It consists of a small subcutaneous sensor wire (FreeStyle Libre sensor) that is FDA approved for 14-day continuous wear. The sensor warm-up period is around 12 h before BG values are displayed.

The sensor records the glucose readings every 15 min and the readings can be scanned within 1 s anytime using the FreeStyle Libre Reader. The reader has an in-built FreeStyle Precision BG meter that can detect the sensor within a range of 1–4 cm to retrieve and display the glucose trend arrow and glucose trend graph with a 15 min frequency till 8 h. Although the glucose trends and alerts are displayed on the reader, the system does not include real-time alarms that can be a disadvantage for patients who are hypoglycemia unaware. Moreover, the system is water resistant for up to one meter in water for a maximum of 30 min and is thus suitable for water sports. One major limitation is that the system does not work with an insulin pump so the patient will have to administer the insulin dosages .

2.5 Comparisons of CGM Systems

Sensor technologies used in CGMs and their accuracy has been improving with new innovations, technological advancements, and time [23, 24]. Table 1 summarizes the accuracy of CGMs, determined by the mean absolute relative difference (MARD) , whereby the lower values correspond to better accuracy. Online data from DexCom suggest that the DexCom G5 Sensor has a MARD of 9% with adults and 10% with pediatrics [25]. Medtronic Guardian Sensor 3 has a reported MARD of 9.64% when calibrating 3–4 times/day and 10.55% when calibrating 1–2 times/day for children and adults [26, 27] . Abbott’s FreeStyle® Libre system has a reported MARD of 11.4% (all ages) and does not require finger-stick calibration [22, 26]. Thus, DexCom G5 seems to have better accuracy among the three CGM systems discussed.

Table 1 Accuracy comparison of CGMs
Full size table

All the three CGMs have their benefits and limitations in terms of their features, sensor lifetime, calibration requirements, ease of use, and data analysis. In 2016, DexCom G5 mobile CGM did a major breakthrough where it was reported to have adequate accuracy and reliability. It is to be used without adjunctive SMBG in clinical decision-making [28]. Additionally, in 2016, the FDA has approved MiniMed 670G for use of continuous basal/bolus insulin delivery. FreeStyle Libre system has undergone several clinical trials whereby its overall accuracy have been proved to be similar to those CGM-systems with high accuracy. Hence, there is a very close cut throat selection of the best CGM system and the cost factor also plays a major role in the choice .

2.6 Recent Advances in Sensor Technologies

Traditional glucose monitoring , SMBG , is the most commonly used method to monitor glucose. Despite it is widely accepted, it is painful and inconvenient especially if patients have to check their BG levels multiple times per day. Moreover, it is not suitable for continuous monitoring. Minimally invasive CGMs have the potential to continuously monitor the BG levels of patients, manage diabetes and reduce the overall societal and economic burden. However, they are still invasive, and have contributing factors such as cost, accuracy and discomfort that remain pertinent factors hindering their wide acceptance and adoption. These major setbacks have triggered the motivation amongst many researchers to explore new frontiers towards non-invasive methods for glucose measurement [29].

The unprecedented advancement in healthcare has led to the development of numerous wearable and miniaturized sensor technologies to enhance the patients’ quality of life and to prevent the risk of complications of diabetes. Different sensors are being developed to measure key biomarkers such as glucose in human fluid other than blood.

Emerging non-invasive blood glucose monitoring techniques are currently being developed through the analysis of tear, sweat, saliva, breath and urine [30]. These techniques are highly attractive alternatives to blood measurements. They can play a major role in blood glucose monitoring by alleviating the pain and discomfort of patients and thus contributing to a higher quality of life and reduced healthcare costs.

Tear sensors. Researchers have been working on tear glucose sensing for over 80 years to explore the non-invasive measurement of tear glucose concentrations (TGC) and converting it to an estimated level of BG [30]. The advancement and wide usage of contact lenses have triggered tear glucose analysis through tears in many studies so that diabetic patients can detect and monitor their BG levels [31].

Miniaturized wireless tear glucose sensors to monitor BG levels in the basal tear fluid have been developed that are integrated under the eye non-invasively or on the inner side of a flexible contact lens [32]. The TGC vary with both the amount of glucose concentration in tear fluid and the volume of the tear fluid fraction retrieved. The glucose concentrations are found to correspond well with the BG levels [33]. The TGC in normal individuals range roughly from 0.1 to 0.6 mM, whereas TGC are as high as 5 mM for diabetic patients [34]. Despite several tear collection methods have been suggested, the precise concentration of tear glucose needed and its correlation with BG level is still debatable since the collection method highly influences the measured TGC. Thus, a sampling method is fundamental to determine the TGC [31].

Tears are being continuously replenished and are less susceptible to dilution than urine, resulting into better glucose sensing [35]. Table 2 presents some recent achievements to monitor glucose via tear-based sensors. Tear based glucose monitoring has enormous potential since tears are highly accessible and available as compared to blood. Consequently, it can be more cost-effective that SMBG if proper correlation between TGC and BG are found.

Table 2 Comparison of glucose monitoring tear-based sensors
Full size table

Despite many developments , the correlation coefficient is not always computed in the studies. The accuracy and performance of the system become more tangible as correlation are computed in more studies. Moreover, the sensor design should cater for lower detection limit and higher sensitivity since the glucose in tear fluid are much lower than in blood. Additionally, the sensor’s selectivity is highly influenced with the lactate and pH levels. The selectivity rises with rising level of lactate and varying pH levels. The size and layout of the sensors are very important to ensure unobtrusive vision, ease of use and higher comfort level during the use of sensors. It is important to implement safe biocompatible battery powered device since lack of power source is another major limitation during implementation and testing [29]. Furthermore, the devices should be able to transfer the sensed response wirelessly to be more practical [43].

Saliva sensors. The development of non-invasive techniques is highly desirable by diabetic patients to diagnose and monitor their health with more comfort and lesser strain due to frequent SMBG. Recent studies have shown that, many diseases [44, 45] including diabetes can be assayed using human saliva produced by the salivary glands [46]. In contrast to glucose monitoring via tear fluid, salivary glucose monitoring is far more advantageous since the oral cavity is less delicate and inconvenient than the eye. The collection of saliva is quite easy and no specific equipment or trained personnel are required as compared to the retrieval of the other human serum [47]. Additionally, a higher amount of saliva is easily available than the secretion of tears and the mouth is more convenient for the placement of sensors that the eyes.

Recently, researchers have attempted to identify and diagnose diabetes cases non-invasively [48, 49]. Another study revealed that a rise in the BG level in diabetic patients leads to a rise in the level of salivary glucose as compared to non-diabetic patients [49, 50]. Many other studies have revealed a change in salivary glucose for diabetic patients [51, 52]. In addition, other studies have shown that diabetes type 2 patients have a higher level of salivary glucose as compared to non-diabetics [53, 54]. Consequently, a positive correlation was found between BG level and salivary glucose in many research works [55, 56]. Therefore, salivary analysis seems to be a very promising diagnostic tool to screen pre-diabetic and undiagnosed diabetic people without complicated and expensive procedures [57]. Moreover, this technique can also be very helpful and cost effective for a larger population yielding better compliance especially with children and elderly people as it is non-invasive and far less painful than retrieval of blood samples [58]. The saliva glucose levels in normal individuals are between 8 μM and 0.21 mM [59].

Zhang et al. [60] have developed a unique on-chip disposable non-invasive saliva nano-biosensor . Besides being cost-effective, it also provides accurate, reliable and continuous glucose levels in saliva for screening of diabetes and glucose monitoring. The sensitivity and accuracy levels of salivary glucose are quite excellent as compared to that obtained by UV Spectrophotometer. Liu et al. [42] performed salivary analysis using a dual-enzyme biosensor for non-invasive continuous diabetes monitoring. Petropoulos et al. [61] constructed a disposable electrochemical biosensor to measure lactate in salivary glucose levels. It is a user friendly and cost effective device and has proved to be very helpful to monitor the levels of glucose in athletes. Arakawa et al. [62] proposed a non-invasive detachable mouth-guard biosensor with telemetry system that could to be applied to the human oral cavity for real-time wireless continuous monitoring of saliva glucose to manage diabetes. Soni and Jha [63] developed a non-invasive low-cost paper strip using an optical glucose biosensor to analyze the saliva for diagnosis of diabetes. The strip color is influenced with the saliva glucose levels and this change was detected using a smartphone camera via RGB color profiling. Dominguez et al. [64] developed a low-cost, portable and user-friendly RGB colorimeter. The device has the ability to provide real-time and accurate measurement of glucose in low concentration of salivary samples with high accuracy and sensitivity.

Despite the numerous advantages presented in several studies, some concerns still remain to be dealt with for a more successful deployment of this technique. The samples need to be properly collected, processed, stored and preserved to prevent any types of contamination and for successful determination of the salivary glucose levels [45]. Some studies have revealed that this technique might not be suitable for diabetes type 1 patients because of the lag in salivary glucose levels. Similar to the tear-based glucose monitoring, this technique also needs high sensitivity and selectiveness to deliver substantial results. Furthermore, the sensor devices should be equipped with wireless capabilities to be able to transmit the data easily [65].

Table 3 presents some devices that are currently being undergoing clinical trials for non-invasive glucose monitoring .

Table 3 Comparison of glucose monitoring saliva-based sensors
Full size table

Sweat sensors. Sweat-based sensors have enormous advantages since they usually make use of the skin as compared to other sensors using tears and saliva [66]. The skin has the largest surface area than the eye and mouth, making it less delicate and inconvenient to interact with the sensors. Compared to other non-invasive medium such as tear and saliva, sweat is readily available, accessible and reproduced in sufficient quantities in human beings [67]. Recent studies have shown than sweat-based glucose monitoring is a highly potential method to perform real-time and continuous monitoring since the sweat glucose concentration (SGC) is correlated with the BG level [14, 68, 69]. The SGC in normal individuals range from 5 to 20 mg/dL [70].

Despite all the advantages of sweat-based monitoring, some major challenges still remain to be tackled. The amount of sweat secreted can differ significantly with regard to the amount of physical activity performed by the patients. Sedentary patients are more likely to have lower amount of sweat secretion than tear fluid or saliva. Additionally, sweat collection can prove to be a difficult and tedious process since sweat has a much lower glucose concentration than that in blood [71]. Moreover, the readings may vary by ambient temperature changes and thus become more complex to be calibrated. Thus, extraction of adequate amount of sweat for reliable glucose sensing using iontophoresis methods is still an open research challenge. The sweat glucose sensing is also influenced by medications and can lead to incorrect readings due to contaminants present in the system of diabetic patients. In addition, the enzyme could be delaminated from the glucose sensor if subjected to friction and too much pressure on the skin can result into skin irritation and rashes [32].

Olarte et al. [72] developed a multi-sensor olfactory system to detect glucose in sweat using an electronic nose. Although the overall accuracy shows that the electric nose is a potential method to measure glucose non-invasively, the sensitivity still needs to be optimized. Liu et al. [73] developed a wireless and real-time sweat-based wearable sensor to monitor the glucose level of patients. The study also demonstrated that sweat conductivity is much better when the patients are hydrated. However, further testing and validation are required to correlate the sweat glucose with episodes of hypoglycemia/hyperglycemia. Cho et al. [74] designed and developed a cost-effective, non-invasive disposable paper-based sensor to monitor the sweat glucose level. The latter is self-powered and obviates the need of any external power source to operate.

Lee et al. [75] developed an innovative closed-loop device to monitor the sweat glucose level of diabetic patients and to adjust the level of insulin with an integrated drug delivery module (Table 4). The device can either be used as a wearable-patch or a disposable one. The study presents very promising results. However, some further works such as correlation testing, calibrations, and extensive clinical trials are yet to be completed for a more optimized system. Anastasova et al. [76] proposed a highly sensitive wearable system with several sensors, and additional capabilities for wireless data collection for real-time monitoring. It also performs multiparameter analysis of sweat (metabolite, electrolytes, temperature) during exercise to measure the glucose non-invasively. However, the performance could be tested with longer duration of exercise and greater number of subjects .

Table 4 Comparison of glucose monitoring sweat-based sensors
Full size table

Urine sensors. Urine has been used as another biomarker to detect glucose concentration levels. It has many advantages over other biological fluids since it is affordable, portable and can be easily reproduced. Moreover, the collection process is non-invasive and painless as compared to retrieval of blood samples. It has been reported that the glucose overflowing into the urine for the diabetic patients leads to a sweet and fruity odor of the urine [77]. High glucose concentration in the urine is quite alarming and the glucose concentration is above 50–100 mg/dL for diabetic patients [78].

Park et al. [79] proposed a portable non-invasive urine glucose biosensor to detect the urine glucose level. It is a light-weight card sized device that uses amperometric technology. Miyashita et al. [80] used amperometric technology to develop a urine glucose meter. The latter was also commercialized. The device could measure urine glucose in a wide range of 0–2000 mg/dL with a very quick response time of 6 s. Moreover, the device showed very promising results despite the presence of other interferents (Table 5). Su et al. [78] proposed a simple and low cost sensor using colorimetric method for the measurement of urine glucose through ZnFe2O4 magnetic nanoparticles. Siyang et al. [24] used an electronic nose to measure urine odor with varying temperatures and corresponded the readings with urine glucose levels for diabetes monitoring. A gas sensor from Taguchi Gas Sensors yielded the highest sensitivity of ammonia detection in the urine sample. Moreover, the glucose concentration was estimated by testing the urine samples with urine reagent strips .

Table 5 Comparison of glucose monitoring urine-based sensors
Full size table

Breath sensors. Several studies revealed that as the BG level rises, there is a significant rise in breath condensate glucose [81, 82]. Guo et al. [83] developed a novel breath analysis system whereby breath signals are analyzed to perform blood glucose monitoring. An ordinal regression and a mapping technique were used to predict the condition of the diabetic patients. Although the accuracy level was not very high, the technique is still a major non-invasive advancement to monitor BG through breath. Saraoglu and Kocan [84] used an electronic nose sensor to detect the glucose level in the breath of patients by analyzing the level of acetone odor (Table 6). Yan et al. [85] proposed a non-invasive and portable breath analysis system to screen diabetic patients and predict the BG level with the use of algorithms. The performance of the system was tested comprehensively and was claimed to be not good enough for practical use. Although the accuracy and performance need to be improved, the studies have offered a potential prospect of non-invasive approach for diabetic patients by using their breath .

Table 6 Comparison of glucose monitoring breath-based sensors
Full size table

2.7 Non-invasive Devices for Glucose Monitoring

Table 7 briefly describes some recent non-invasive devices for glucose monitoring. The target site, technology used, and status of each of the devices are also provided.

Table 7 Non-invasive devices for glucose monitoring
Full size table

3 Data Reliability and Quality in BAN for Diabetes Monitoring

In a BAN , the vital signs of the diabetic patients are captured at the sensor level, and are transmitted across multiple networks and mobile devices until they reach the healthcare professionals [91]. Guaranteeing high data reliability and quality in WBAN is practically impossible since by their nature, BANs are susceptible to faults and failures during data acquisition, data processing and data delivery. Thus, it is critical to ensure a high degree of confidence for data reliability and quality in such an erratic environment [92]. High QoI in BAN is a crucial aspect for its development, deployment and acceptance for diabetes monitoring in the near future. Moreover, with the help of a BAN exhibiting high data reliability and quality, patients diagnosed with diabetes can make better informed decisions, self-manage their vital signs, be more patient-centric than doctor-centric, and lead a higher quality of life.

QoI is a multi-dimensional concept referring to the “fitness for use” [93, 94]. It aims to measure and manage data quality along with information. Although a vast number of dimensions related to accuracy, consistency, completeness, and timeliness have been mentioned in several studies, a defined set of data quality dimensions has not been formulated specifically for diabetes monitoring [93, 95,96,97].

Additionally, many frameworks of dimensions have been introduced and developed to ensure a higher degree of confidence of QoI for several fields such as information systems and e-health monitoring applications. However, there is no specific QoI framework to evaluate the data reliability and quality in the BAN for diabetic monitoring. Data reliability and quality can have a significant effect on the overall performance of the BAN for diabetic monitoring with regard to its adoption. A lack of adherence to high data reliability and quality such as medical errors can lead to patients’ loss of lives and incorrect diagnosis [98].

3.1 Stages of Diabetes Data Management

Due to the recent advancements in wireless sensors and mobile phones, massive volumes of real-time data are generated, and these data can be helpful in the diagnosis and decision-making by continuous monitoring of patients. However, an explosive increase in data has led to prevalent challenges to design reliable data collection and transmission strategies for sensor network [99]. Efficient data interpretation and decision-making is often very challenging and critical since real-time processing should be made from possibly uncertain data in a time-critical fashion. Since the transmitted data have different size, priority, and security levels, some of them have to be compressed or encrypted before being transmitted over the network.

Hence, QoI is highly impacted during the five main stages namely data acquisition, pre-processing, transmission, storage, and analytics .

Data acquisition challenges. Data acquisition is a huge challenge due to erroneous and incomplete data in the acquisition process [100]. Since the sensors are battery powered, it is highly important to minimize the data to be transmitted, to reduce the communication cost, and at the same time to maintain the accuracy of the sensed values. Due to interferences, path fading, and multi-hop routing paths, the sensed data can be delayed during data transmission from sensor to mobile thus causing latency, unreal-time data, incorrect diagnosis, economic burden, and even deaths [101]. Furthermore, as sensors are resource-constrained and error-prone wireless devices, data loss from sensor to mobile eventually leads to incomplete data and creates uncertainty in the data representation [9]. Two major approaches used to acquire data are pull-based or pushed-based. Data is captured based on only the frequency set by the user in the pull-based approach, whereas in the push-based approach, the sensors transmit the data to the mobile computer tier only if any anomalous behavior is detected .

Data preprocessing challenges. Preprocessing raw data is a fundamental step in WBAN for diabetic monitoring due to several factors such as noise in the data and sensor errors [101]. Some major tasks during data preprocessing phase include (1) data cleaning and transformation (2) feature extraction and selection, and (3) data compression.

Data cleaning and transformation challenges. Data cleaning is the initial step that needs to be performed effectively to clean the sensed data since they are often unreliable and contain other sources of errors. The aim is to identify the incorrect, faulty, irrelevant, inaccurate, incomplete, omitted, or duplicate data in order to transform them for higher data quality and reliability. Effective, correct, and complete information is essential for appropriate data analytics and prediction. Another persistent challenge that needs to be addressed is the detection of anomalies by comparing raw sensed data with their respective inferred sensed values and thereby developing new approaches for data cleaning in a computationally efficient way. Data cleaning is often achieved through identification of missing values, removal of redundant values, detection and correction of erroneous data, and data transformation [102]. Common techniques such as threshold-based methods, low pass and high pass filters, Wavelet Transform, mathematical morphology filters, piecewise linear representation, statistical tools, Fourier Transform, Laplacian Transform, and declarative data cleaning approaches are used for this purpose [101, 103, 104]. The Kalman filter algorithm has been widely used to detect the anomalies, substitute missing, or incorrect values among the sensed data so as to smooth the noisy dataset [102, 105].

Feature extraction and selection challenges. The extraction and selection of data fields is a major step to provide a meaningful representation of the sensor data that are relevant to the data analysis for better decision-making. Selection of pertinent features reduce redundancy, noise in the data and eventually result in lesser amount of data to be transmitted [106].

Data compression challenges. Multitude amount of sensors lead to a tremendous amount of sensor data generated and pose huge challenge with regard to data storage and management. In such situation, storing all the data locally on the sensors or the mobile phone is quite impractical and consequently these data should either be compressed or dropped. Motivated by this challenge, many data compression mechanisms and algorithms have been developed to significantly decrease the quantity of data to be sent over the sensor network for better query processing and data storage [107]. The choice of compressing or dropping the data is quite a complex one.

Data compression factor is deduced through the QoI requirements of the patients, which is frequently achieved through trade-off among several dimensions such as cost-effectiveness, reliability, accuracy, and power efficiency [108]. Various techniques have been used for data compression by transmitting only necessary data such as any change in values since the last transmission or any anomalous signals deviating from a preset threshold value [109]. Furthermore, a threshold is often proposed and other sensed values are eventually approximated based on preset threshold. In other studies, only relevant features are selected such as the anomalous data .

3.2 Data Transmission and Data Collection Challenges

Data transmission is another major step whereby the remote monitoring server receives data sent from the mobile computing tier. Data reliability, security, and privacy are important QoI dimensions to be considered to ensure reliable performance of the WBAN and effective data analytics. Often, delay or failure of data transmission occurs due to internet bandwidth or network failure. Retransmission schemes need to be put into place in case message sent acknowledgements are not received before a timeout for better message delivery and optimized battery performance of mobile phones [110]. Rule-based methods are during data summarizing and transmitting to ensure effective transmission [111]. Moreover, the privacy and security of sensitive and confidential patient medical data might be compromised since various users such as the doctors and healthcare professionals will have access to them and a break into the system is possible [112].

3.3 Data Storage Challenges

One of the most critical aspects for a successful WBAN adoption is data security .

The remote server acquires the patients’ data that are sent via the GPRS network, categorizes and stores them into a database or on the cloud for better management of huge volume and complex data. Moreover, the WBAN should ensure proper backup facility in case of loss of data from the server or the cloud so that timely data can be accessed and retrieved for real-time monitoring. The cloud is vulnerable to human mistake, faulty equipment, network connectivity error or any third party’s spoofing intentions such as Byzantine attacks [113]. Furthermore, it is vital to enforce and guarantee data security by limiting the access only to authorized users [114].

3.4 Data Analytics Challenges

It is a big technical concern to work towards promoting meaningful visualization of the huge volume of sensed data that is acquired and stored in WBAN. Consequently, data analytics techniques have the potential to contribute to proactive healthcare services and remote monitoring which will ultimately benefit the global society and economy. Among data analytics methods, machine learning techniques, are being highly used in WBANs to analyze the sensor data and obtain significant insights for diagnoses with high precision and effective decision-making [115]. Supervised learning algorithms are used to predict the known properties of data whereas unsupervised learning algorithms focus on discovering unknown knowledge from data.

Acampora et al. [116] categorize data analytics into six main components such as (1) activity recognition, (2) behavior pattern discovery, (3) anomaly detection for emergency situations, (4) planning and scheduling, (5) decision support, and (6) anonymization and privacy preserving. Data analytics can also be helpful to process huge amount and complex data and thereby providing decision support with regard to the correct amount of carbohydrate intake, insulin doses, and levels of physical activity. The insulin dosage could be adjusted based on the planned exercise activity or carbohydrate intake.

Since there is a vast range of standalone software on the market that performs decision support for diabetes, patients are highly prone to potential consequences of inaccurate or erroneous diagnosis, feedback and decision support guidelines. The FDA has planned to enforce proper policies and procedures for standalone software and medical mobile applications that deal with patients’ information, provide diagnosis and decision-making support [117].

4 Data Reliability and Quality Metrics for BAN

Data quality (DQ) is often defined as ‘data that are fit for use by data consumers’ and data quality dimension is defined ‘a set of data quality attributes that represent a single aspect of data quality’ [118]. DQ dimensions have been widely researched and elaborated in many research works. Wang and Strong [118] published a framework which described a DQ model consisting of four broad categories with their respective DQ dimensions namely intrinsic, contextual, representational and accessibility. Fox et al. [119] also proposed four different dimensions, namely accuracy, completeness, consistency, and currentness.

The Canadian Institute for Health Information described five main dimensions such as comparability, relevance, timeliness, accuracy, and usability [120]. Kahn et al. [121] proposed a healthcare specific framework using a “fit-for-use” data quality model in which he proposed five high-level dimensions such as appropriate amount of data, timeliness, believability, objectivity, and accuracy. Liaw performed an extensive literature review looking for commonalities on data quality dimensions and he also found consensus on the five most common occurring dimensions (accuracy, completeness, consistency, correctness, and timeliness) [122]. Based on literature review, Almutiry et al. [123] proposed an initial framework considering the type of c1oud-based healthcare systems that fits the data quality dimensions (accuracy, completeness, relevance, timeliness, usability, consistency, privacy, confidentiality, secure access, provenance, and interpretability).

Despite numerous attempts have been made to define healthcare data quality, they do share overlapping dimensions and it is challenging to apply them in the BAN context for diabetic monitoring. Thus, a framework towards QoI for BAN for diabetes monitoring is needed to ensure high data quality and reliability.

4.1 QoI Dimensions and Metrics for Diabetes Monitoring

Table 8 summarizes a set of data quality dimensions filtered from existing data quality dimensions and their respective metrics which will be relevant for the BAN for diabetes monitoring.

Table 8 QoI dimensions and metrics for diabetes monitoring
Full size table

4.2 QoI Framework for Diabetes Monitoring

The QoI framework as shown in Fig. 1 is an illustration of how the BAN systems will be evaluated. The framework consists of four main blocks including the (1) sensor data sources, (2) different data types required for processing, (3) process of diabetes data management, and (4) QoI dimensions and respective metrics.

Fig. 1
figure 1

QoI framework for diabetes monitoring

Full size image

Data sources. The data can be retrieved from several sources such as traditional glucometers, CGMs or other non-invasive/minimally invasive wearable sensors. A wide range of off-the-shelf glucometers are currently available on the market and they have different specifications, weight, size, accuracy, cost, readability and portability. Majority of the glucometers generate the results within 15 s and have adequate storage capabilities to save the results. CGMs devices such as DexCom G5, Medtronic 670G® and Abbott Freestyle™ Libre can be used to detect the BG of the patients. Non-invasive sensor devices using other human fluids than blood namely tears, saliva, breath, urine, or sweat can be used to detect the glucose concentrations .

Data types. Different data types such as the patients’ demographical data (age, diabetes duration, weight, height and body mass index), types of diabetes, glucose level (HbA1c), insulin dosage (IU/kg per day), smoking status, physical activity (aerobic energy-producing systems, resistance training, flexibility or balance exercises), frequency of exercises (0, 1–2, 3 or more times per week), duration of exercises (number of minutes per week), and food intake (type of food, serving sizes, and food intake timings) are necessary information that are required to be regularly recorded. These information are very important to make effective diagnosis of the patients .

Stages of diabetes data management. In the mobile computing tier , the main stages are (1) data acquisition, (2) data pre-processing, and (3) data transmission. In the remote server tier, the main stages are (1) data collection, (2) data storage, and (3) data analytics.

QoI dimensions and metrics. The QoI dimensions and metrics will be very useful to evaluate the degree of confidence of QoI and reliability in BAN for diabetic monitoring with regard to the data being delivered to the patients or healthcare professionals and the processes to transform the data. Some dimensions such as usability, provenance and interpretability cannot be measured directly and will thus have to be measured through surveys or semi interviews .

As future work, the QoI framework will have to be applied on the different BANs to further evaluation.

5 Research Directions

WBANs have immense opportunities in monitoring chronic diseases, including diabetes, but at the same time, such systems give rise to many research directions. These systems are very critical and complex since they demand high data quality and reliability for successful adoption and acceptability. Due to these complexities, many research directions are still open. The future research directions are discussed in the section below and are categorized under the sensor level, network level, and human-centric level.

5.1 Sensor Level

Despite the fact there is a plethora of sensors , there are many limitations associated with them such as inaccuracy, inconvenience, discomfort, and lack of power source and wireless capabilities. While developing future sensors for CGMs and non-invasive devices using other human serum than blood, it is of paramount importance that sensors require minimum calibration to prevent lengthy and complex procedures. Moreover, besides longer battery lifetime, sensors should also be miniaturized, flexible, stretchable, and ultralight so that they are less obtrusive, thus ensuring that patients can lead their life normally.

Additionally, it would be impractical to have sensors with wires for data transmission, therefore, wireless capabilities should be integrated within those sensors for seamless data transmission and sharing with healthcare professionals. In addition, sensor accuracy, sensitivity and selectivity should be further tested to obtain accurate and reliable measure of glucose concentration for effective diagnosis and diabetes management. Lastly, more CGMs sensors should be integrated with insulin pumps to facilitate the process of insulin delivery and control.

Furthermore, intense biocompatibility testing should be performed to ensure that the sensors are safe for patients and will not cause any allergic reactions, skin inflammations or irritations. Glucose concentration is highly influenced by environmental, human factors, medications, physical activity and dietary habits, therefore, a direct comparison will not necessary lead to conclusive results [73]. Thus, further correlation testing should be performed to ensure that the results are accurate .

5.2 Network Level

Ensuring QoI during data collection, data processing and data delivery in WBANs is a major challenge as guaranteeing QoI is hardly possible. WBANs need to satisfy the degree of confidence of QoI by conforming to a set of appropriate quality dimensions so that relevant and timely decisions can be taken by patients and medical experts. Higher QoI in WBANs will allow capture of regular measurements and thus promoting better day-to-day healthcare follow-ups through reliable diagnostics, monitoring, and clinical guidance.

Incorrect, erroneous and missing sensed data or delayed information transmission can be ambiguous, leading to incorrect diagnosis, untimely decision-making, unnecessary expenditures, and even deaths. Sources of such information can be due to (1) incorrect sensed information, (2) lack of calibration of the sensors or (3) errors during data transmission because of path fading, unstable multi-hop routing paths and other interferences. Therefore, the number of errors during data transmission and the delay must be as minimal as possible. CGMs and other sensors should integrate data transmission modules so that information can be uploaded automatically to remote servers, cloud or smartphones for further storage and analytical purposes. Thus, new systems should ensure that high priority level data are continuously transmitted in a real-time manner despite frequent deep fading and packets loss.

High reliability of data is another key challenge since inappropriate handling of exceptional situations, false-alarm triggers or loss of service in WBANs impact the quality of the healthcare directly and may become a life-threatening matter. Therefore, it is highly important to detect any emergent situations and provide timely assistance. In order to tackle the threats, the information should be to be transmitted with a minimum delay even under any node failure or malicious modifications especially in emergent situations. Thus, new mechanisms on emergency data prioritization and fault tolerance techniques should be developed to improve the reliability of information.

Moreover, many studies have reported a lag time of human serum (tears, saliva, sweat, urine and breath) glucose relative to blood glucose. Future systems should develop new algorithms for estimating glucose concentration from the raw electrical signals.

5.3 Human-Centric Level

Another future research direction in WBANs is to tackle human-centric issues. Since the WBANs are usually used to target various aged groups, it is essential that both technology-naive and technology-averse patients are able to use the WBANs with ease. Concerning the sensitivity and mobility of sensors, it is equally important that users are seamlessly connected to the wireless network and they can move around freely anywhere and anytime despite of any environmental conditions. Despite many WBANs are under development and clinical trials, several key concerns such as effective on-node processing, power efficient design, mobility and portability, user acceptance and adoption, and distributed interference during transmission need to be dealt with to promote higher usability and innovative WBANs’ features.

The emerging deployment of WBANs is triggering a global pattern shift from reactive to preventative healthcare whereby design considerations play a fundamental role in confronting the challenges outlined. These systems offer a variety of exciting opportunities from a healthcare perspective such as encouraging self-monitoring, assisted-monitoring, supervised-monitoring, and continuous-monitoring [91].

6 Conclusion

CGMs and non-invasive wearable platforms have shown impressive advances and high potential to manage diabetes. However, several challenges and limitations to the use of these systems still persist. Addressing these issues will definitely lead to emerging and successful WBANs. The ideal glucose sensor is still elusive, but many technologies are currently under development to sense glucose, as well as to perform data analysis in real time for better patient monitoring. The use of such systems is bound to rise in the near future since the prevalence of diabetes is increasing very rapidly, and patients will thus be moving towards non-invasive and painless solutions to manage their conditions. Hence, these systems will have to be cost-effective, user friendly, easy to use, produce automated results, and strive for FDA approvals. The data quality and reliability should also be consistently measured using appropriate QoI framework to spur wide adoption, acceptance and usage. Furthermore, early stage diagnosis of diabetes, with appropriate management and data analytics, will reduce health complications and mortality.