Correlation Between Sweat Glucose and Blood Glucose in Subjects with Diabetes

Introduction

D iabetes affects millions of people, and its prevalence is growing rapidly. ¹ Results from the Diabetes Control and Complications Trial show that increased frequency of blood glucose (BG) measurement coupled with intensive therapy leads to substantially improved outcomes. ² However, current methods used for self-monitoring of BG are invasive, painful, and inconvenient. Several groups have attempted to develop minimally invasive or noninvasive methods for self-monitoring of BG. ^3,4 Sweat glucose (SG) has been previously measured in human subjects. ^{5 –12} However, there is no clear demonstration of a correlation between SG and BG in these studies. In the study presented here we have investigated the relationship of BG and SG during sweat stimulation using a collection method to specifically harvest sweat and a detection method with accuracy, precision, and specificity to measure low levels of glucose found in sweat.

Subject and Methods

Perfusion sweat collection

The correlation of SG and BG was evaluated using a perfusion technique. A photograph of the perfusion device is shown in Figure 1. The collection area of the perfusion device was 2 cm². The device rests on the skin's surface after sweat stimulation and is held in place with an elastic band. The sweat is collected in a serpentine collection chamber that is perfused with 150 μL of water every 10 min. The sweat is then swept into a collection vial along with the perfusate. The procedure is repeated every 10 min during the experiment, and each sample is subsequently analyzed for glucose by high-performance anion exchange liquid chromatography with pulsed amperometric detection (HPAE-PAD) (see below). Devices were attached to the left forearm and right forearm for simultaneous measurements from each site.

OPEN IN VIEWER

FIG. 1.

The perfusion, sweat collection device. Color images available online at www.liebertonline.com/dia

To determine the volume of sweat collected the sweat rate is measured in an area immediately adjacent to the sweat collection area. The sweat rate is determined by measuring the relative humidity of a stream of dry nitrogen blown over the skin's surface. From the sweat rate, area, and duration of the measurement the sweat volume can be calculated. The mass of glucose in the sample (determined by HPAE-PAD) and the volume are used to determine the glucose concentration in sweat.

During the initial application of the sweat collection chamber leakage of sample and contamination were often encountered. Hence, the data from the first 10-min collection intervals were omitted from analysis.

Results and Discussion

Following pilocarpine stimulation sweat rates increased to above 1 μL/cm²/min. During the course of the study, however, sweat rates decay continuously to about 0.1 μL/cm²/min, similar to unstimulated values. Sweat samples were collected by the perfusion method and analyzed by HPAE-PAD. The data for one subject are shown in Figure 2. The subject entered the study with BG values near 200 mg/dL, which increased to just over 300 mg/dL over the next 60 min. During this period BG and SG samples were collected every 10 min. As can be seen in Figure 2, the SG values “lag” the corresponding BG values. The “lag” time is determined by moving all SG values along the time axis until maximal correlation with BG values is obtained for all the data in an individual study. The results of this analysis, obtained across a large group of subjects, have shown that the SG values “lag” BG values by about 8 min. All linear regression results presented here utilize a constant 8-min time lag. The linear regression analysis of the data in Figure 2 yields a correlation coefficient (R) value of 0.92.

OPEN IN VIEWER

FIG. 2.

Sweat glucose (diamonds) and blood glucose (squares) data for one subject. Color images available online at www.liebertonline.com/dia

The same linear regression analysis was performed for all studies. The individual R obtained for all 23 studies yielded a mean value of 0.83 and a median value of 0.94.

A linear correlation between SG and BG is equivalent to a constant SG/BG ratio during each study. The dependence of SG/BG on sweat rate is shown in Figure 3. These results show a small but statistically significant (P<0.001) increase in SG/BG with increasing sweat rate. Note, however, that at more physiologically relevant sweat rates below about 0.9 μL/cm²/min there is no effect of sweat rate on SG/BG (R=0.065, P<0.45). In addition, these results show that the correlation maintains even at unstimulated sweat rates of 0.1 μL/cm²/min.

OPEN IN VIEWER

FIG. 3.

The sweat glucose/blood glucose (SG/BG) ratio versus sweat rate for all subjects studied. Color images available online at www.liebertonline.com/dia

To prospectively “predict” BG values from SG, a calibration is required. The calibration utilized the first analyzed data point assuming a linear relationship between SG and BG and an intercept of 0. This is equivalent to using the BG/SG ratio of the first point to “predict” all following points in time. This calculation is shown in Eq. 1 where BG at time t (BG _t ) is calculated from SG at time t (SG _t ) and the calibration constant, which is the ratio of BG/SG at the initial time point (BG/SG _t=0): \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} \begin{align*}{\rm BG}_t = {\rm BG} / {\rm SG}_{t = 0} \times {\rm SG}_t \tag{1} \end{align*} \end{document}

The data for all 23 studies analyzed using the first point calibration are shown in Figure 4. Note that the calibration point is not included in these data. Hence, there are 115 total data points (23 studies×5 points per study). The mean relative difference (MRD) is defined as the mean value of ([SG – BG]/BG) over all data pairs. The mean absolute relative difference (MARD) is the mean of the absolute value ([SG – BG]/BG) over all data pairs. The MRD reflects accuracy, and MARD reflects spread of the data, or precision. For the data presented in Figure 4 the MRD is −6±32%, and the MARD is 23±23%. The correlation coefficient for these data is 0.75, which is statistically significant (P<0.0001). None of these metrics includes the calibration point.

OPEN IN VIEWER

FIG. 4.

Predicted (measured) blood glucose values from sweat glucose versus a corresponding reference blood glucose value. The upper and lower lines represent +20% and −20%, respectively, about the middle line. Color images available online at www.liebertonline.com/dia

A summary of the data for all subjects is shown in Table 1.

The results presented here show that there is a statistically significant correlation between SG and BG in a group of subjects with diabetes. Moreover, the correlation was obtained as the sweat rate varied over a 10-fold range. This is equivalent to sweat rates obtained from ambient conditions to temperature near 50°C. The SG values lag the reference BG value by an average of 8 min. Much of that time delay is due to the time required (10 min) to collect sufficient sweat to measure glucose by HPAE-PAD. Using a single-point, prospective analysis, BG values obtained from SG were obtained. These results show that for all subjects the average deviation (MARD) of the measured value relative to the reference value is about 23%. Note that a commercially available BG meter and strip designed for self-monitoring were used as a reference in these studies so that some of the deviation can be accounted for by BG values. The error in these measurements is as much as 20% and is included in the total value (23%) cited here. (The International Organization for Standardization standards for BG monitoring devices [ISO 15197:2003] state that 95% of the individual glucose results shall fall within±15 mg/dL of the results of the manufacturer's measurement procedure at glucose concentrations of <75 mg/dL and within±20% at glucose concentrations ≥75 mg/dL.) Thus, the actual error associated with SG measurements is comparable to the error obtained in BG measurements. The results presented here show that glucose in sweat is at a concentration that is about 1–2% of the corresponding blood value. This is lower than the value reported in previous studies. ^{6 –9} The difference most likely reflects the care taken in this study to collect sweat samples uncontaminated with glucose from other sources on the skin (see below). In addition, the studies presented here used a specific glucose assay, whereas those earlier studies measured “reducing” substances, which likely contain other sugars as well as glucose and other substances.

The data presented here also show a strong correlation between SG and BG over a broad range of subjects, BG levels, and sweat rates. These results are in contrast to earlier studies where no correlation was found. ^6,7,8,10 We believe the failure to show a strong and reproducible correlation in those previous studies stems from the fact that they sampled glucose from multiple compartments in or on the skin's surface.

There are at least three distinct kinetic compartments of glucose on the skin. These compartments are associated with (a) the stratum corneum, (b) the outward migration of glucose from the interstitial fluid, and (c) sweat.

The stratum corneum is the outermost layer of the skin. This outer layer contains glucose that is enzymatically cleaved from a lipid precursor as epidermal cells migrate outward. The second source of glucose is from outward diffusion from the interstitial fluid underlying the dermis. Several groups have shown that glucose from this source can be collected on the skin's surface over several hours period of time. ^16,17 Again, because of the time required to obtain the glucose sample, this source cannot reflect changes occurring rapidly in blood levels. The most rapidly appearing source of glucose is that found in sweat. As sweat response occurs rapidly and the sweat gland is highly vascularized, this source can reflect glucose levels within the body.

Previous researchers have not consistently found a correlation between SG and BG. We believe the reason for this difference lies in the manner in which sweat is collected. We have taken care to clean the skin to remove residual glucose and occlude the skin prior to sweat collection to mask other sources of glucose on the skin. The occlusive barrier is such, however, that active sweat can penetrate it, forming beads of liquid on the surface. This sweat can then be collected without contamination from other sources of glucose on the skin's surface. We also collect sweat samples at short time intervals to minimize mixing with other sources of glucose.

In summary, we have demonstrated in subjects with diabetes that SG and BG are highly correlated over a broad range of sweat rates and BG values. These studies form a proof-of-principle for sweat collection and glucose measurement. Although the device described here is not commercially viable, a substantially simpler device has recently been disclosed. ¹⁸