Currently submitted to: Journal of Medical Internet Research
Date Submitted: Mar 26, 2020
(closed for review but you can still tweet)
Towards Detecting Infection Incidences in People with Type 1 Diabetes Using Self-Recorded Data: A Novel Framework for a Digital Infectious Disease Detection Mechanism
Type 1 diabetes mellitus is a blood glucose (BG) metabolic disorder, which is caused by deficiencies of insulin secretion from pancreatic cells. The relationship between infection incidents and elevated BG levels has been known for a long time. People with diabetes often experience prolonged episodes of elevated BG levels as a result of infection incidences. Despite the fact that patients increasingly gather data about themselves, there are no solid findings on how to use such self-recorded data as a secondary source of information for other purposes, such as self-management related decision support during infection incidences and digital infectious disease detection system.
The aim of the study is to demonstrate how people with type 1 diabetes can assist in detecting infectious diseases outbreak. Furthermore, to shade light upon the possibility of assisting the individual during such an incident. Specifically, we aim to retrospectively analyze the effect of infection incidences, such as influenza (flu), and light and mild common cold without fever, in order to identify key parameters that can effectively be used as potential indicators (events) of infection incidences. Moreover, the paper presents a general framework of a proposed digital infectious disease detection system based on self-recorded data from people with type 1 diabetes.
We retrospectively analyzed high precision self-recorded data of 10 patient years captured within the longitudinal records of 3 people with type 1 diabetes. Getting such a rich and big dataset from large number of participants are extremely expensive and difficult to acquire, if not impossible. The participants were 2 males and 1 female with an average age of 34 (13.2) years. The dataset incorporates BG levels (Self-monitoring of blood glucose (SMBG) and continuous glucose measurement (CGM)), insulin (bolus and basal), diet (carbohydrate in grams) and self-reported events of illness. All the participants were free from any other health complications and other diseases during these years. Five normal patient years without any infection incidences and five patient years each with at least one or more cases of self-reported acute-infection incidences were analyzed and compared. We investigated the temporal evolution and probability distribution of BG levels, injected insulin, carbohydrate intake, and insulin to carbohydrate ratio within a specified timeframe (weekly, daily and hourly). For the daily and hourly timeframes, a moving average filter and non-parametric density estimation techniques, kernel density estimator, were used to analyze the data trend and distribution respectively, before, during, and after the infection incidences. The pre-infection, infection, and post-infection week analysis were carried out on raw dataset based on the week’s daily mean and standard deviation of BG levels, and daily sum and standard deviation of insulin and carbohydrate. A statistical boxplot was used to depict the comparison during pre-infection, infection, and post-infection week. All the experiments were carried out using Matlab 2018a.
Our analysis demonstrated that upon infection incidences, there is a dramatic shift in the operating point of the individual BG dynamics in all the timeframes (weekly, daily and hourly), which clearly violate the usual norm of BG dynamics. During regular/normal situations, BG levels usually lower when there is a significant increase in insulin injection and reduction in carbohydrate consumption. However, in all of the individual’s infection cases as opposed to the regular/normal days, there were prolonged period with elevated BG levels despite injecting higher amounts of insulin and reduced amount of carbohydrate consumption. For instance, in all the infection week on average, BG levels were elevated by 6.1% and 16%, insulin (bolus) were increased by 42% and 39.3%, carbohydrate consumption were reduced by 19% and 28.1%, and insulin to carbohydrate ratio were raised by 108.7% as compared to pre-infection and post-infection week respectively.
We presented a novel approach on how to use self-recorded data from people with type 1 diabetes to develop an infection detection system. The analysis revealed that despite tight BG management regimes, BG levels were still elevated during the infection period, demonstrating the significant effect of infection on BG dynamics. Throughout the infection period, BG levels were elevated despite injecting higher amount of insulin and consuming lower amount of carbohydrate. The changes might be subjected to hormonal changes in the body as a result of infection incidences. However, the magnitude of the impact on BG dynamics could be correlated with different factors such as degree and severity of infection, type of pathogens, associated hormones involved and others. These changes are quite significant anomalies as compared to the regular/normal days, where BG levels lower with increased insulin injection and reduced carbohydrate consumption, and therefore, can be detected with appropriate individualized computational models, i.e., algorithms that span from prediction models to anomalies detection algorithms. Generally, we foresee that these findings can benefit the efforts towards building the next generation digital infectious disease detection systems and provoke further thoughts in this challenging field.