A signal processing application for evaluating self-monitoring blood glucose strategies in a software agent model
Introduction
Diabetes is fast becoming an epidemic in today's society. The International Diabetes Federation estimates that, in 2012, over 371 million people suffered from diabetes and 4.8 million people died [2]. In addition, diabetes causes a significant financial burden to individuals as well as to society, e.g., diabetes healthcare costs were estimated around 471 billion US dollars in 2012 [2].
Self-monitoring of blood glucose (SMBG) proved to be very effective in terms of blood glucose (BG) control in Type 1 and Type 2 diabetes, alongside insulin treatment [3], [4]. However, whether SMBG improves BG control for Type 2 diabetic patients in the absence of insulin remains controversial [5]. Some researchers believe that SMBG does assist these patients in managing their BG levels [6], [7]. Diabetic patients are better informed of their BG levels through regularly sampling their BG, which leads to better lifestyle decisions including diet control, exercise, and medication [8]. The likelihood of complications is reduced with good BG balance, which can be obtained through frequent SMBG. Furthermore, some of the discomfort associated with hyper- and hypo-glycemia, as well as of repeated finger-prick testing, can be avoided with frequent SMBG [9]. Some other researchers are of the opposite opinion, believing that SMBG is a waste of money [10]. Nonetheless, SMBG has significant potential benefits and is commonly recommended to diabetic patients of both types [6], [7]. According to the recommendations of the American Diabetes Association (ADA) [11], SMBG should be utilized to achieve the optimal postprandial glycemic target. For instance, it is recommended that patients on intensive insulin therapy sample their BG at least 3 times per day [11].
There are several methods of BG monitoring available to diabetic patients today. In certain situations, patients are required to use an invasive continuous glucose monitor which reports BG every 5 min, while most other, patients are asked to use an external blood test using a finger prick and a BG sensitive strip. For the medical community as well as society in general, monitoring costs are an important issue; additionally, they vary tremendously depending on monitoring approaches and monitoring frequencies. Continuous glucose monitors can only be put in place for a limited number of days, and though BG strips are much less expensive, even their costs mount with many samples per day over many months. For instance, the National Health Service (NHS) spends approximately 40% more on materials used for testing BG in diabetic patients than on oral hypoglycemic drugs [12]. In addition, the necessary frequency of SMBG can vary according to the individual patient. Given the cost and discomfort associated with SMBG, it seems appropriate that effort be directed toward the determination of an appropriate SMBG rate for patients. The conventional approach of a Randomized Controlled Trial (RCT) in vivo has obvious disadvantages for determining this appropriate sampling rate, as it is time-consuming, has a significant monetary cost, and is limited by inherent difficulties in the study of unique events.
The BG signal is a complex signal whose values continuously change as a result of numerous patient factors, including health condition, age, exercise, medication, time of day, food intake, lifestyle, etc. The BG signal can be characterized as non-stationary [13] because its fundamental statistics change as a function of time based on conditions experienced by the patients, for example, the consumption of a glucose bolus as a result of eating carbohydrate-rich foods. As a result of its non-stationary quality, periodic sampling of the BG signal is less effective than event-driven sampling. Typically, the patient is advised to record BG levels before and sometime after consumption of a meal. This allows the physician to assess the change in BG values as a result of fasting between meals and due to the meal itself. This variation of BG levels is a useful indicator of the extent of the illness of the patient.
We propose a signal analysis technique of calculating the cross-correlation function and the average deviation (AVD) between continuous blood glucose (CBG) and the interpolation of samples to evaluate BG monitoring frequency (MF) using the self-aware patient agent (SPA) model [14]. This model has been developed to study various aspects of the treatment of diabetic patients. The SPA is a 24-h circadian, self-aware, stochastic model of diabetes, which produces the CBG signal in a mobile agent environment [1]. We extend the original seminal work of Ackerman et al. [15] in creating a mathematical model of human BG levels in three aspects: (1) We incorporate the stochastic and unpredictable effects of daily living; (2) we extend the Ackerman model into the period of night-time; (3) patients’ awareness of their own conditions is incorporated. The SPA has a distinct advantage over human testing, as the ethical issues of patient well-being do not need to be considered.
Ackerman et al. proposed a linearized mathematical model to evaluate glucose-tolerance and BG regulation in the 1960s [15], [16]. It was originally used to predict a combination of sine and exponential wave response to oral glucose consumption in order to characterize the human glucose regulatory system. While there have been significantly more complex (glucose–insulin reactive) models developed since the advent of the Ackerman model, these models rely on more in-depth knowledge and measurement of the metabolic process, which is, in general, not available to the typical diabetic patient on an on-going basis. The Ackerman model is defined by a set of differential equations, and the solution has the form as follows:where x is the BG level as a function of time t, G0 is the fasting BG, ω is the natural frequency of the system, β is measured based on the intensity of exercise and medication, and F is a measure of food intake. Their aim was to improve the distinction between normal and abnormal glucose regulation. Ackerman's work has led to significant follow-up research, including the study of Jansson et al. [17] in which Ackerman's model was used to analyze the BG curves obtained during the oral glucose tolerance test in 378 cases. An oral glucose tolerance test measures the body's ability to use glucose. The intestinal glucose re-absorption levels were observed using Ackerman's model to improve the distinction of a diabetic condition from a normal state. Following that, Wu [18] used Ackerman's model to evaluate the degree of diabetes in a particular subject and this subject's response to medications. Wu attempted to define the effect of the medication in terms of parameter fitting of BG measured from a diabetic subject with or without medication. He attempted to assess the impact of the medication based on the values of the parameters in Ackerman's model. Most recently, the Ackerman model was used by Shiang [19] in 2010; their study aimed to develop methods to interpret laboratory glucose and insulin data from glucose tolerance tests, as well as to enhance the Ackerman model.
Section snippets
Methods
In this work, the SPA model generates both CBG and BG samples. The event driven samples of the continuous BG are converted into a continuous signal through interpolation. The cross-correlation function is applied to a normalized signal and is calculated for the zero lag term. It is identified as the normalized cross-correlation (NCC), and is a measure of the similarity in shape of two waveforms. The AVD evaluates the average difference between the CBG and the interpolation of the samples.
Experimental configurations
A number of experiments were conducted to simulate nine categories of SPA with different MFs. In these experiments, we only intended to observe thirty days; however, for the SPA with the MF of 1 time per week, forty-two days were required to acquire the six critical BG measurements. Therefore, we extended the simulation duration to two months for this case. In each patient category, there was one SPA. The simulation in a two-month period allows the model to represent many different conditions
Simulation results
The mean value and variance of NCCs and AVDs were calculated to quantitatively characterize to what extent the patients can have a good understanding of the daily pattern of CBG from the BG samples.
Discussion
NCCs illustrate the similarity between the interpolation and the CBG, while AVDs show the difference of the average values between them. To emphasize the similarity, both average values of the interpolation and the CBG are set to be zero by subtracting their average values. The observations of both NCCs and AVDs show how accurate the samples can reveal dynamic BG levels. Furthermore, the mean and variance of the two factors facilitate a general understanding of the relationship between the
Conclusions and future work
SMBG is a highly recommended protocol for diabetic patients; however, the cost of self-monitoring is significant for both the healthcare system and individuals. Discomfort when patients’ fingers are pricked is also a factor in SMBG. In addition, the frequency for using the process of SMBG can vary according to the individual patient. Given these facts, it is increasingly important to investigate individually adjusted sampling frequencies. The conventional approach of the Randomized Controlled
Conflict of interest
None declared.
References (22)
- et al.
The self-aware diabetic patient software agent model
Comput. Biol. Med.
(2013) - et al.
Self-monitoring of blood glucose levels and glycemic control: the Northern California Kaiser Permanente Diabetes registry
Am. J. Med.
(2001) - et al.
Self-monitoring of blood glucose significantly improves metabolic control in patients with type 2 diabetes mellitus: the Auto-Surveillance Intervention Active (ASIA) study
Diabetes Metab.
(2003) - et al.
Diagnostic value of the oral glucose tolerance test evaluated with a mathematical model
Comput. Biomed. Res.
(1980) - et al.
A computational model of the human glucose–insulin regulatory system
J. Biomed. Res.
(2010) IDF Diabetes Atlas Update 2012
(2013)- et al.
Frequency of blood glucose monitoring in relation to glycaemic control: observational study with diabetes database
BMJ
(1999) - et al.
Self-monitoring of blood glucose in non-insulin treated patients with type 2 diabetes: a systematic review and meta-analysis
Curr. Med. Res. Opin.
(2009) - et al.
Impact of self monitoring of blood glucose in the management of patients with non-insulin treated diabetes: open parallel group randomised trial
BMJ
(2007) - et al.
The efficacy of self-monitoring of blood glucose in NIDDM subjects: a criteria-based literature review
Diabetes Care
(1997)
Economic costs of diabetes in the US in 2012
Diabetes Care
Cited by (6)
Modeling the SMBG measurement error
2020, Glucose Monitoring Devices: Measuring Blood Glucose to Manage and Control DiabetesHealthcare professionals' experiences and perceptions of artificial intelligence in regional and rural health districts in Australia
2023, Australian Journal of Rural HealthSecure agent-based diagnosis and classification using optimal kernel support vector machine
2021, International Journal of Biomedical Engineering and TechnologyA Model of Self-Monitoring Blood Glucose Measurement Error
2017, Journal of Diabetes Science and TechnologyOptimal Residential Demand Response for Multiple Heterogeneous Homes with Real-Time Price Prediction in a Multiagent Framework
2017, IEEE Transactions on Smart Grid