The rise in obesity has become a worldwide concern, with an increase from 30.5% in 1999–2000 to affecting approximately 41.9% of adults in the US from 2017–2022, according to the Centers for Disease Control and Prevention (CDC) 1,2 . Obesity is a complex disease related to the imbalance between caloric intake and calories expended, which is driven by an intricate interplay between genetics, environmental, nutritional, and socioeconomic factors 3 . Obesity is related to several health conditions including cardiovascular disease, type 2 diabetes, sleep disorders, metabolic disorders, respiratory abnormalities, hypertension, and cancer due to the secondary effects on human physiology. The rise in obesity has also caused growing financial strain on United States (US) adults, with an annual estimated medical cost of $ 173 billion for the US in 2019.
Physical inactivity is one of the leading causes of obesity. Sedentary lifestyles are estimated to double the risk of obesity and related disease, resulting in approximately 2 million deaths per year. With 60–85% of people worldwide leading sedentary lifestyles, the increase in lifestyles has greatly attributed to the increase in obesity-related mortality in the US4.
In obese individuals, glucose intolerance is more common, which can be an early diagnosis of diabetes mellitus5. Glucose intolerance is one of the markers of metabolic syndrome6,7. People with high blood glucose who are not yet diagnosed with diabetes remain at risk for diabetes complications. Appropriate management is necessary to prevent this complication.8 One of them is to recognize the various risk factors associated with glucose intolerance. Glucose intolerance is related to several risk factors, including percentage body fat, BMI, higher social economy9, and dietary pattern10,11. Cortisol is a hormone released during stress and associated with gluconeogenesis.12 Increased cortisol hormone in people who have diabetes mellitus has been widely studied13, but the study in people who have glucose intolerance is still lacking. However, as cortisol and glucose have a dynamic interplay within human physiology19, what is evident is that there is an immediate need to monitor glucose and cortisol dynamically in a concurrent manner.
Glucose (6-carbon structure) is the main source of energy for cells in the human body. Hyperglycemia occurs when the blood glucose level in the blood is too high, this is driven by factors such as stress, illness, high carbohydrate intake, obesity as well as insulin resistance. Chronic hyperglycemia may lead to type-2 diabetes.
Cortisol is a glucocorticoid that plays an important role in the regulation of the circadian cycle, activity, stress management as well as nutrition, and glucose regulation. The circadian cycle or the sleep-wake cycle is influenced by sleep schedules. Dysregulation of the circadian rhythm has been linked to several health risks14, including an increased risk of breast and prostate cancer by the national toxicology program (NTP)15,16. Chronic stress results in the disruption of the homeostasis of the HPA-axis due to a continuous increase in cortisol levels. Chronic high levels of cortisol may lead to an increased risk of health problems such as anxiety, depression, weight gain, as well as Cushing syndrome. 17,18
Sweat is a biological fluid — like blood, saliva, and urine — that contains metabolites, electrolytes, proteins, and hormones. The levels of these vary depending on a person’s health. Wearable sweat sensors have been developed to track users’ health conditions and monitor the levels of these substances (known as analytes) in sweat20. The developed wearable sweat-based biosensor, utilizes permanent direct skin contact with the sensing area, to accurately analyze the biocomponents of the sweat in a continuous, dynamic, and noninvasive manner. The resulting information will be readily converted into readable electrical/optical signals and transmitted to the user using a wireless transmission device, ultimately providing the dynamic health information of the wearer. The biggest challenge with sweat wearables is the difficulty of collecting and routing sweat. There are various methods for sampling sweat for sensing. One of the most representative approaches for collecting sweat is using microfluidic systems with channels that deliver sweat. However, such systems are unreliable for real-time monitoring of the dynamic metabolites glucose and cortisol. These systems require the accumulation of sweat up to a minimum addressable volume, which results in an accumulation of concentration of the metabolite. Hence the measure is not a true representation of the metabolite.
Electrochemical methods are used to generate electrical signals by the specific redox reaction of the analytes on the electrode, thus enabling the conversion of biochemical information into readable electrical signals, offering high sensitivity, high selectivity, and fast response time21,22. In recent years, different electrochemical methods have been implemented for wearable sweat biosensors, including potentiometric, amperometric, voltammetric, and impedance methods 23,24.
We have pioneered the development of electrochemical passive sweat wearables for monitoring and tracking metabolites, ions, small molecules, and proteins in a real-time continuous manner20,25−28. Additionally, we have demonstrated the correlation of these expression profiles with systemic expression20. We have previously demonstrated correlating macronutrient consumption with the dynamics of glucose and cortisol expression profiles using our sweat sensor20. We have demonstrated the fidelity of sweat measurements over time for these two metabolites. In this work, we present for the very first time a human observational study to track the relationship between glucose and cortisol over the 24-hour period i.e. the day and night cycle, to observe the interrelationship between these metabolites, due to human performance over the day.
The second aspect of tracking metabolites dynamically is the application of machine learning towards rapidly interpreting the concentrations of the metabolites in a real-time continuous manner based on the electrochemical sensor data.
Machine learning has various types of training based on the final objective of the problem to be achieved. Based on the availability of data labels supervised or unsupervised methods can be implemented. Apart from these two, there are semi-supervised learning and reinforcement learning algorithms. Data used as the training set is collected from past experience and it can be numerical or categorical, in some cases even text data can be used. Machine learning tasks are either classification in the categorical target output variable or regression in the continuous target variable. To date machine learning research for wearable devices belongs to the classification tasks, some are for clustering29, and few can be tackled as regression problems29. In this work, we have evaluated a regression-based approach for glucose and cortisol. The prediction algorithm for glucose was published in our previous publication30. The cortisol algorithm was built on a similar algorithm and details are presented in the results section.
Here we present for the very first time an observational study of tracking glucose and cortisol dynamics in 4 human subjects over a normal 24-hour period. Using a novel data-driven approach to sweat glucose and cortisol tracking through a passive electrochemical skin sensor. This observational study focuses on the effect of sedentary lifestyles in comparison with an active lifestyle, on glucose and cortisol levels in sweat.