2.1.

NCGM Device

The NCGM system, developed by Taiwan Biophotonic Corporation in Hsinchu, Taiwan, incorporates an advanced optical measurement setup comprising a light source, beam splitters, photodetectors, and a processing camera unit, as shown in Fig. 1. Utilizing a 1650 nm Fabry–Perot laser diode for its light source, the system adheres to the International Electrotechnical Commission’s safety standards for medical laser devices. The initial beam splitter focuses the laser light into the eyeball. The light that reflects off the eye’s anterior chamber is then redirected through the beam splitters to the photodiode detectors. The system is also equipped with a zeroth light receiving module, responsible for assessing the feedback light beam’s power intensity and estimating the emitted collimated light beam’s strength from the laser source, ensuring both safety and the prevention of analyte damage.

Fig. 1

Diagram of the NCGM system including (1) laser diode, (2) to (4) beam splitters, (5) detector for optical feedback control, (6) detector for polarization of light, (7) detector for absorption of light, (8) monitoring camera, and (9) testing human eye.

The light beam, originally emitted, is converted into a collimated beam by a collimator, proceeding through the first beam splitter to the analyte. A part of this beam is redirected as a feedback light beam to the zeroth receiving module, where it is converted into electrical signals. Post interaction with the analyte, the measurement light beam is split by the second beam splitter into two detection beams. These beams are then captured by the first and second light receiving modules. The crucial integration and collaboration between the optical measurement module, microprocessor, and light source underpin the functionality of this sophisticated glucose monitoring system.

These modules gather data on absorption energy and polarized optical rotatory distribution, aiding in the differentiation between emitted and reflected light. The device’s developer has identified optical angular differences and absorption energy differences across various biological molecules, including glucose, cholesterol, and uric acid. The processing unit leverages this information, employing simultaneous equations based on absorption and polarization values to calculate the biological molecules’ concentrations. Polynomial equations have been formulated to accurately depict the relationship between these concentrations and the measured optical and energy differences, taking into account both the target and interfering molecules. This analytical method enables the accurate determination of target molecule concentrations, even in the presence of interfering substances, by utilizing optical angular difference and absorption energy difference values. Ultimately, this system is capable of calculating glucose levels in the aqueous humor, showcasing its precision in monitoring biological molecules.

The general time frame to obtain a measurement is between 30 and 60 s. Further details about the device and glucose calculation can be found in the previously published literature²⁴ and the US patents (Nos. US20160299058A1 and US9662004B2). The blood glucose level, which correlates with the aqueous humor glucose level, can then be determined. The device must be calibrated for each patient because each eye has a unique background optical signal. A blood glucose sample is taken simultaneously with the NCGM measurement to calibrate the device.

2.2.

In Vitro Repeatability Tests of the NCGM

In vitro and repeatability experiments utilized custom-designed glass test tubes featuring convex bottoms that closely resembled the cornea’s curvature radius ($r=6.4\mathrm{mm}$) and a thickness of $500\mu \mathrm{m}$. Seven glucose concentrations (50, 100, 200, 300, 400, 500, and $600\mathrm{mg}/\mathrm{dL}$) were prepared in a balanced salt solution (Alcon Laboratories Inc., Fort Worth, Texas) and added to the test tubes to simulate the eye’s anterior segment. The NCGM was then calibrated using distilled water. Following this, the absorption and polarization of the seven glucose solutions were measured with the NCGM. These measurements were recorded and graphed. Adhering to the ISO 15197:2013 standard, which standardizes in vitro glucose measurement, the measurements need to be within 15% of the target glucose value (when the target $\text{glucose value}\ge 100\mathrm{mg}/\mathrm{dL}$) or $15\mathrm{mg}/\mathrm{dL}$ (when the target $\text{glucose value}<100\mathrm{mg}/\mathrm{dL}$).

2.3.

Participants

This prospective study was conducted at a single center according to the Declaration of Helsinki and with the approval of the Institutional Review Boards at Taiwan Food and Drug Administration and Chang Gung Memorial Hospital (IRB No.: 201301153A0). This study was also registered with ClinicalTrials.gov (Identifier No.: NCT02430441). Patients planning to receive intraocular surgery, such as cataract surgery, intraocular injection, or vitrectomy, were invited to participate in the study. The study participants provided written patient informed consent after a detailed explanation, and informed consent was obtained from all subjects involved in the study. Participants were aged $\ge 20$ years at the time of the study, with DM or not. Eligibility inclusion and exclusion criteria are provided in Table 1.

Table 1

Participant eligibility inclusion and exclusion criteria.

2.4.

Study Protocol

For each enrolled participant, one eye was assessed (the planned surgery eye or one eye if receiving two-eye surgery). The study’s outcomes were eye aqueous glucose level, preoperative blood glucose level, intraoperative blood glucose level, and NCGM reading. Patients fasted for 8 h before the experiment. The experimental protocol is described in Table 2. All adverse events were reported and coded based on the Medical Dictionary for Regulatory Activities. Patient hospital medical records were examined.

Table 2

Experimental protocol.

2.5.

Aqueous Tapping

An aqueous sample was collected prior to the planned surgical procedures. The preoperative sampling of aqueous fluid involved topical or periocular anesthesia and 5% povidone-iodine using a sterile lid speculum. A 30-gauge 1-in. needle was used for anterior chamber paracentesis. Approximately 0.1 mL of aqueous humor was collected during sampling. Samples were transferred to microcentrifuge tubes and were assayed for glucose levels immediately.

2.6.

Glucose Measurement Methods

The glucose levels in aqueous samples from the tapping solution were determined using the hexokinase/glucose-6-phosphate dehydrogenase method with deproteinization, which is widely acknowledged as the standard reference method for measuring glucose (Fisher Diagnostics, Middletown, Virginia).²⁵ Serum glucose levels were assessed using two different assays: one based on venous blood samples and the other through SMBG measurements. For the venous blood samples, 2 mL of blood was drawn through phlebotomy and immediately sent for analysis. The serum glucose concentration in these samples was also measured using the hexokinase/glucose-6-phosphate dehydrogenase method with deproteinization, adhering to the aforementioned reference method for glucose measurement (Fisher Diagnostics, Middletown, Virginia). For the SMBG measurement, finger strips were used with SMBG devices according to the manufacturer’s instructions (Roche Diagnostics Corporation, Indianapolis, Indiana).

2.7.

Statistical Analysis

In this study, two approaches were employed to evaluate the agreement between NCGM measurements and aqueous glucose, blood glucose, and SMBG measurements. The first approach involved calculating the intraclass correlation (ICC) of consistency between the two methods. The ICC value was determined using a two-way mixed-methods model, with observations as a random effect and the raters (the two measurement methods) as a fixed effect. An ICC value of $>0.7$ signified good agreement between the two measurement methods. To assess the acceptability of individual differences between the methods, the acceptable boundaries were established at the 95% confidence interval (CI) of the mean difference between them. A two-sided $P$-value of $<0.05$ was considered statistically significant. The second approach involved generating a Bland–Altman plot of the mean glucose levels from NCGM in comparison with other measurements against their difference. The Bland–Altman plot and ICC analysis were conducted using MedCalc Statistical Software (version 13.1.2.0; MedCalc Software, Ostend, Belgium). These methods were essential in ensuring the precision and dependability of the measurements.

3.1.

Study Participants

This study enrolled 28 participants between April of 2015, and December of 2017. The participants’ mean age was 66.38 years; 15 (53.6%) were female, and all were Taiwanese inhabitants. Eight (28.6%) participants had known DM (all were type 2). Patients’ fasting blood glucose levels were as follows: 9 (32.1%) $<100\mathrm{mg}/\mathrm{dL}$, 11 (39.3%) 100 to $150\mathrm{mg}/\mathrm{dL}$, and 8 (28.6%) $>150\mathrm{mg}/\mathrm{dL}$ (maximum $320\mathrm{mg}/\mathrm{dL}$). All participants completed a 6-month clinical follow-up. There were no subjective or objective findings related to optical damage to the external eye, lens, or fundus due to the measurements or systemic adverse events.

3.2.

In Vitro Repeatability Test

Prior to applying the NCGM to human subjects, the device was tested with standardized glucose concentrations to ensure its repeatability and accuracy in vitro. Employing the methods outlined in the previous section, each glucose solution underwent 12,000 measurements (10 times per second for 20 min). Subsequently, consecutive sets of 1000 measurements were averaged, and the results are shown in Fig. 2. The red lines in Fig. 2 indicate the error limits set by ISO standards: $\pm 15\mathrm{mg}/\mathrm{dL}$ for a glucose concentration of $50\mathrm{mg}/\mathrm{dL}$ and an error margin limited to within $\pm 15\%$ for concentrations ranging from 100 to $600\mathrm{mg}/\mathrm{dL}$. The black dots represent the measurements obtained from our device, which consistently fell within these prescribed error margins across the range of standard glucose level solutions tested.

Fig. 2

Repeatability test of the NCGM system using seven glucose concentrations (50, 100, 200, 300, 400, 500, and $600\mathrm{mg}/\mathrm{dL}$).

3.3.

Correlation Between the Different Measurements

The ICC results are shown in Table 3. The ICCs of NCGM to SMBG or blood glucose levels were all greater than 0.9 preoperatively and intraoperatively, indicating a statistically significant correlation (all $P<0.001$). Furthermore, the ICCs of aqueous humor to SMBG or blood glucose levels were $\sim 0.8$. In addition, the ICC of NCGM to aqueous humor glucose levels was 0.76. The agreement of NCGM to other measurements and aqueous humor glucose was also evident in the Bland–Altman plots (Fig. 3). Most of the observations in these plots did not exceed the upper and lower 95% CI boundaries, indicating that the individual differences between the measurements were within the acceptable range and did not deviate significantly from the mean difference between measurements. We also utilized Pearson’s linear correlation coefficient to assess the relationship between different measurements. Our analysis revealed a significant linear correlation between NCGM reading and blood glucose ($R=0.968$, $p<0.001$). Moreover, there was a notable correlation between aqueous glucose and blood glucose ($R=0.954$, $p<0.001$). These results indicate a strong correlation between NCGM readout from aqueous glucose and blood glucose levels, suggesting a reliable prediction of blood glucose levels by aqueous glucose levels. Furthermore, a Clarke error grid analysis was conducted and is shown in Fig. 4 to assess the accuracy of the NCGM system. SMBG values served as the reference for this analysis. All data points fall within zone A, demonstrating a minimal risk of error in the measurements.²⁶ These findings affirm the validity and reliability of the measurements and instill confidence in the accuracy of the NCGM results.

Table 3

The results of intra-class correlation between different measurement methods.

Fig. 3

Bland–Altman plot demonstrating the agreement of NCGM with (a) 30-min preoperative SMBG, (b) preoperative blood glucose, (c) 10-min preoperative SMBG, (d) intraoperative SMBG, (e) intraoperative blood glucose, and (f) aqueous humor glucose.

Fig. 4

Clarke’s error grid analysis comparing measurements from the NCGM with SMBG values.