Dermal Glucose Sensing has a Shorter Time Lag Relative to Blood Glucose: Implications for Hypoglycemia Detection and Time in Range

Abstract

Introduction:

Current continuous glucose monitors (CGM) sensing glucose in the subcutaneous tissue have a significant time lag (τ). This delay could result in severe hypo/hyperglycemia and lower time in range (TIR). Dermal sensing can greatly reduce time lag.

Methods:

In a clinical study conducted at two US-based clinical centers, subjects with type 1 diabetes mellitus (DM) wore a novel dermal CGM + Abbott-Libre 3 or Dexcom-G7. All were compared to a YSI-glucose analyzer. Time lag kinetics for all sensors were modeled using the two-compartment model and compared to published data. Time lag data and its potential effect on TIR were also analyzed.

Results:

Data from 55 subjects showed fast kinetics for the dermal CGM. In total, 93% of the Laxmi sensors had a τ of 0-2 minutes, whereas commercial CGMs had a varying distribution of τ (−10 to 10+ minutes). This reduction in τ by 10 minutes has profound effects on errors in insulin administration in both open-loop and in a proportional-integral-derivative (PID) model of automated insulin delivery (AID). To evaluate the effect of tau on TIR, we used an in silico PID controller in a well-accepted model (UVA type 1 diabetes simulator) over a variety of conditions. We observed that tau greatly affects TIR and the distribution of the time out of range parameters.

Conclusion:

Dermal sensing has a time lag close to 0. Individuals with DM can have lower glucose targets with a system that eliminates fear of hypoglycemia, resulting in higher TIR and better control of DM.

Introduction

According to the International Diabetes Federation (IDF), nearly 600 million people worldwide had diabetes as of 2024. ¹ Diabetes causes abnormal fluctuation in glucose levels, leading to acute and long-term complications. ² To avoid such complications, timely monitoring of glucose levels is crucial to the timely management of glucose fluctuations.

Continuous glucose monitoring (CGM) systems have transformed the management of diabetes over the past two decades by providing close to real-time monitoring of glucose, and since the Food and Drug Administration (FDA) approval of the first commercial CGM in 1999, CGMs have improved overall. Moreover, the enhanced insurance coverage has contributed to the widespread use of CGMs across individuals with type 1 diabetes and intensive insulin therapy in patients with type 2 diabetes.

Despite these advancements in the world of CGMs, commercial CGMs have a major limitation related to time lag that could be as long as 10 to 15 minutes. ³ Commercial CGMs rely on sensing glucose in the subcutaneous tissue for glucose monitoring, where the presence of lipocytes/adipocytes delays the transfer of glucose relative to blood. The subcutaneous tissue, also known as the hypodermis, is located beneath the dermis; it represents a deeper layer below the skin surface.

Dermal sensing, by contrast, offers the promise of far faster kinetics due to the unique physiology of the dermis, which is characterized by a dense capillary network making the dermis richly vascularized and the minimal to no presence of adipocytes. The lack of insulating adipose tissue and the high vascularity of the dermis facilitate more rapid diffusion and exchange of glucose between the interstitial fluid (ISF) and the underlying capillary bed, resulting in faster time lag kinetics overall.

In this clinical study that we are reporting about, the prototype version of the Laxmi dermal CGM showed that a sensor in the dermis yields lag times of 0 to 2 minutes. By capturing glucose dynamics more quickly with a minimal time lag, dermal sensing offers the opportunity for the timely management of glucose fluctuations, especially when levels are rapidly changing. This in turn can also enhance time in range (TIR) and the overall management of diabetes and the associated outcomes. A CGM like this may overcome the current challenges related to time lag, which can subsequently result in better diabetes management and the overall quality of life of individuals with diabetes.

This article presents detailed analyses of clinical data pertaining to a proof-of-concept dermal CGM, its minimal time lag, the expected improvement in TIR achievable through sensing glucose in the dermis, and the potential role of dermal CGM when integrated with automated insulin delivery (AID) systems to enhance diabetes outcomes.

Methods

The Continuous Glucose Monitoring System

The Laxmi dermal CGM technology utilizes Glucose Oxidase (GOx)-based electrochemical sensing to measure glucose. The system is composed of a sensor and a Bluetooth-based transmitter that is connected to an app. A patented inserter designed by Laxmi therapeutic devices was used to reliably and securely place the sensor in the dermis, about 1.8 to 2.0 mm below the skin. Unlike other proposed dermal CGMs, the Laxmi CGM uses a single sensor and does not rely on a microneedle-based system.

The generation-one Laxmi CGM that was utilized in this study was designed for wear on the back of the arm, for a duration of seven days. The Laxmi CGM measures a signal that is later (retrospectively) turned into a glucose value, at the sampling interval of 30 seconds (2880 measurements per day).

Study Design

We evaluated the performance of dermal sensing with the Laxmi CGM through a prospective, multi-center clinical study conducted at two US-based clinical centers.

The study was approved by the Institutional Review Board (WCG, IRB0000053); written informed consent was obtained from all subjects and documented. In total, 55 participants with type 1 diabetes mellitus (T1DM) were enrolled and wore the Laxmi CGM on the back of the arm and simultaneously wore either the Abbott FreeStyle Libre 3 or the Dexcom G7 CGM on the back of the other arm. The total study duration was seven days, with a total of three clinic sessions of up to 10 hours each. Data from both the Laxmi and the comparator CGMs were compared to a total of 40 real-time, plasma glucose measurements made by the YSI benchtop glucose analyzer in each clinic session, totaling 120 measurements per subject for the whole study. Participants were advised to continue insulin administration as routinely instructed by their physicians. Participants arrived at the study sites in a fasting state and were given a standardized and controlled meal to consume, which had approximately 85 g of carbohydrates, similar to a typical meal. Subjects were informed of the carbohydrate content of the administered meal, and they self-administered insulin accordingly.

The three clinic sessions covered days 1 or 2, 3 or 4, and day 7 of sensor wear. At the beginning of each clinic session, an intravenous (IV) line was placed with a heating pad applied to the same arm as the IV line. Study participants had their venous blood glucose drawn and tested using the benchtop glucose analyzer YSI 2300 or YSI 2500. Samples were collected once every 10 minutes for the first 30 samples and once every 15 minutes after that. Samples were centrifuged and tested within 3 minutes of draw time; draw and measurement times of the YSI samples were captured and documented in real time. Participants were blinded to data from the CGMs for the whole duration of the study. At the end of each clinic session and on the final day of the study, data were downloaded from all sensors and uploaded into the Laxmi database for analysis.

Figure 1 shows an example of clinic data from Laxmi and the comparator CGMs as well as the YSI analyzer. The trace in red represents the YSI data; the one in purple represents continuous data from the Laxmi CGM, while the one in blue represents data from Libre 3, all over the same duration of time. Note that the Laxmi data are shown as a continuous curve since data are available every 30 seconds.

Figure 1.

Example of data from a clinic visit.

Data Analysis

Analysis of time lag between the sensor signals and results from the YSI analyzer was done using the two-compartment model ⁴ (Figure 2).

Figure 2.

The two-compartment model.

In this model, blood glucose (concentration = C₁) diffuses into the tissue and yields a tissue glucose (Concentration = C₂) which lags the blood. The time constant tau (τ) characterizes the delay. Longer τ means more delay. Literature reports of τ range from 5 to 20 minutes. We have developed validated software to calculate tau using sensor output signals compared to YSI measurements as truth.

The Laxmi dermal sensor, like all CGM sensors, measures ISF glucose except that it does so in the dermis. However, blood glucose is more physiologic for insulin dosing, so all CGM technologies use an algorithmic approach to predicting blood glucose based on ISF measurements. Commercial CGMs compensate for the Blood-Tissue time lag in various manners including look-forward algorithms ⁵ or using a fixed time lag. Both approaches use population metrics to attempt to convert sensor readings to appropriate blood glucose.

Since there is closed-form solution to evaluate tau in the two-compartment model, we developed an algorithm based on SciPy brute optimization. Inputs to the calculation are the sensor signal with corresponding time stamps and the YSI values with their times. The method then calculates which tau value (0-15 minutes in 1-minute increments) minimizes the mean square error between the measured and predicted fits. The model was validated by creating a series of known glucose and signals with known tau and with the addition of non-Gaussian noise to the sensor signal. Overall performance is excellent with calculated MARDs for control systems near 0%. Note that we do not optimize tau for MARD but rather for minimized mean square error.

Results

Time Lag

Data from 55 subjects (see Table 1 for characteristics) demonstrated fast kinetics, as most of the Laxmi dermal sensors (93%) had time lag (τ) that ranged between 0 and 2 minutes (Figure 3). Commercial CGMs had a varying tau distribution, ⁶ of up to plus or minus 10 minutes in this study. The negative time lag that commercial CGMs may have does not indicate that the sensor was physiologically faster than blood but rather reflects the combination of filtering and predictive algorithmic signal processing used in such CGMs, whereas the generation-one Laxmi CGM utilized in this study did not utilize any look-forward algorithms.

Table 1.

Demographics.

Demographic characteristic	Category	Number of subjects (N = 55)
Age	Mean	48.2
	Median	46
Gender	Male	34 (61.8%)
	Female	21 (38.2%)
Ethnicity	Hispanic	7
	Not Hispanic	48
Race	White	44
	African American	5
	Asian	1
	Middle Eastern	2
	Other	3
HbA1c Level	Mean	7.1%

Figure 3.

Percentage of sensors within tau ranges, by CGM type.

We used the two-compartment model to evaluate τ for both Libre 3 and G7, and despite their use of algorithmic approaches to real-time results, we show that many sensors had large time lags. This is consistent with data reported by Zhang et al. ⁶ , which showed significant variability in time constants.

To evaluate the consequences of long lag times, we used a proportional-integral-derivative (PID)-controlled loop series of in silico evaluations of 30 subjects in the UVA diabetes simulator. ⁷ The PID controller is similar to the Medtronic controller described in the literature ⁸ and in US Patent 7,354,420. Using the controller, the PID coefficients were manually optimized for TIR for each patient. Each simulation ran for 72 hours with common meals on each day. The target for the PID controller was 140 mg/dL.

In general, for a reasonably optimized adult subject, TIR (70-180) was reduced from 86% to 71% by increasing tau from 0 to 8 minutes. For adolescent subjects, TIR was reduced from 73% to 65% by implementing an 8-minute tau. Figure 4 shows the continuous relationship between Tau and TIR. Changes in TIR were predominately an increase in blood glucose (BG) > 180 mg/dL coupled to a smaller increase in values <70 mg/dL for adults. For pediatric subjects with high insulin sensitivity, however, an increase in time below 70 mg/dL was the predominant pattern.

Figure 4.

Time lag vs TIR.

It seems likely that fast time constants should allow patients to be more aggressive with feedback control targets. To test this, we initially hand-optimized the PID controller for a given subject (child 7). Using optimized parameters, we ran four simulations, Target 140 with zero τ, Target 140 with 10-minute τ, Target 120 with 0 τ, and Target 120 with 10-minute τ. The results are shown in Table 2.

Table 2.

Effect of Tau on TIR.

TIR metrics	Target 140 mg/dL tau = 0 min	Target 140 mg/dL tau = 10 min	Target 120 mg/dL tau = 0 min	Target 120 mg/dL tau = 10 min
70-180 mg/dL	69.7%	54.5%	77.5%	62.6%
>180 mg/dL	23.7%	26.4%	13.0%	15.1%
< 70 mg/dL	6.6%	19.0%	9.5%	22.3%
>250 mg/dL	0.0%	1.3%	0.0%	0.0%
<50 mg/dL	0.0%	1.9%	0.0%	6.56%

As seen in the data, for a given target, a 10-minute tau leads to about an absolute 15% decrease in TIR with a small increase in the >180 mg/dL range and a large change in time below 70 mg/dL. As expected, a sensor with a short time constant allows the patient to safely change targets from 140 to 120 mg/dL. The TIR increases by 8%, time above 180 significantly decreases, while time below 70 increases by a small amount. However, a target of 120 with a 10-minute tau leads to a dramatic increase in time below 50 mg/dL. Clinicians now have the opportunity to increase TIR by allowing patients to safely be more aggressive in closed-loop therapy.

Discussion

In this study, we demonstrated that our dermal CGM exhibits a minimal time lag relative to blood glucose when compared to conventional CGMs sensing in the subcutaneous tissue. The in silico PID-controlled model shows a meaningful increase in TIR.

Time in range is a critical metric of glycemic control, especially in individuals with T1DM, and it correlates with long-term outcomes of diabetes.^9,10 A key challenge in optimizing TIR with CGM use is the time lag between the real-time blood glucose measurements and measurement reported by the CGM, since a longer time lag delays the detection of abnormal glucose fluctuations, especially during rapid rates of change, and ultimately leads to the delayed management of hyper- or hypoglycemia events, reducing the clinical utility of the CGMs and limiting its ability to prevent severe hyper- or hypoglycemia.

Time lag is a direct result of both physiological and technical factors. The physiological delay results from the diffusion dynamics of glucose from blood into the interstitial fluid. Technical delays arise from factors like membrane mass transfer and analog and digital signal processing.

The currently available commercial CGMs sensing in the subcutaneous tissue have long time lags, as demonstrated in several studies, including the study we are reporting in this article. Recent data published by Dexcom on the time lag of the G7 system showed a varying distribution of G7 sensors’ time lag between −10 and +10 minutes. ⁶ Another study ¹¹ reported a median time lag of 9 minutes and a maximum of 20 minutes using commercial CGM systems, underscoring the magnitude of delay that can occur even under optimal calibration conditions. Results from our clinical data, as described in the results section above, also report a varying distribution of time lag for both G7 and Libre 3 CGM systems, which was up to 10 minutes (Figure 3).

To overcome the time lag challenges, commercial CGMs utilize signal-processing algorithms that convert raw electrochemical systems into ISF glucose estimates. These look-forward algorithms work by analyzing past and present glucose trends to predict future values, and they are incorporated to help issue early alerts or adjust insulin delivery proactively in AID systems. However, in our study, these look-forward algorithms were not as effective as dermal sensing, as described in Figure 5, which illustrates a case in which one of the study participants administered a higher than necessary insulin dose based on readings reported from their own personal CGM and subsequently experienced hypoglycemia, demonstrated by two values of 69 mg/dL reported by the YSI analyzer. Notably, the participant was blinded to the comparator CGM utilized in this study (Libre 3, in this example), which was concurrently overestimating glucose values by approximately 50 mg/dL relative to the reference YSI. Although no insulin dosing decisions were made based on this comparator CGM due to blinding, this case highlights a critical real-world risk. Had this subject been using a personal CGM of comparable performance (which was likely the case), this misinterpretation of glucose levels—possibly due in part to the time lag—could have directly influenced their insulin dosing and contributed to the hypoglycemic event.

Figure 5.

Real-world example of CGM overestimation.

Furthermore, an important question arises regarding how CGM validation protocols have historically been designed overall and whether they may have systematically underestimated the impact of sensor lag by using carbohydrate loads smaller than a typical full meal?

In many in-clinic validation protocols, carbohydrate challenges used to stimulate post-prandial glucose excursions may have involved amounts lower than those typically consumed in a full meal. Such sub-maximal glucose challenges would induce relatively gradual changes in blood glucose at slow rates, during which the physiological gradient between blood and ISF remains small, thereby minimizing the observable effect of sensor lag. In contrast, full meals with high glycemic loads, such as the standardized meal we utilized in this clinical study (approximately 85 g of net carbs), produce normal glucose excursions, resulting in higher rates of change of glucose. During a full meal, the disconnect between the blood and ISF glucose (ie, lag) becomes more evident, unlike smaller amounts of glucose that induce gradual glucose increases, during which the blood-to-ISF gradient remains small and the true lag impact of CGM devices may be obscured in the process.

During periods of rapid glucose fluctuations such as post-meals, exercise, or insulin administration, time lag (despite the existing look-forward algorithms) becomes more observable, as mentioned above, and results in misinterpretation of glucose trends, delayed therapy adjustments, and undershooting or overshooting of insulin administration. This leads to a decrease in the time spent in range (TIR). We studied the performance of the dermal CGM under conditions that involved “real-world” glucose rates of change to enable an accurate evaluation of the time lag, and as the results indicated above, the time lag was minimal (0-2 minutes). We expected the time lag to be minimal based on the physiology of the dermis ¹² with capillary beds above and below it, and our study results supported this concept.

Using in silico simulations, we showed that the integration of our dermal CGM into PID algorithms led to improved glycemic metrics. Particularly, the dermal sensor resulted in an 11% absolute increase in TIR and reduced time spent in hyper- and hypoglycemia. These findings suggest that reductions in CGM time lag can allow insulin delivery systems to respond more quickly to glycemic changes, particularly when glucose is changing rapidly, such as in scenarios involving meals, exercise, or insulin administration.

The clinical relevance of time lag is well-established. Delayed glucose readings can lead to over/under correction of glucose levels and sub-optimal adaptation to rapidly changing glycemic trends, which leads to a reduced TIR and compromises effective management of diabetes. By reducing time lag, the in silico results show that our device will allow patients and AID systems to make faster and more reliable decisions, leading to safer and more precise insulin delivery. Ultimately, this will increase TIR and result in better short- and long-term outcomes of diabetes overall.

Conclusion

Overall, the results of this study demonstrate that dermal sensing using the novel, first-generation Laxmi CGM has a minimal time lag that offers near instantaneous readings of glucose, which outperforms that of currently available commercial CGMs.

With dermal sensing, individuals with diabetes can have lower glucose targets when using a system that eliminates fear of hypoglycemia, especially when integrated with feedback and control closed-loop systems and eventually could achieve a higher TIR.

This translates to improved diabetes management, much lower risk of dangerous glucose fluctuations, greater patient confidence and adherence, and an overall better quality of life for individuals with diabetes.

Laxmi Therapeutic Devices is actively pursuing a second-generation, fully disposable dermal CGM that will be commercializable. The second-generation product will have a transmitter form factor that is smaller than the G7.