https://www.euro-online.org/enog/inoc2007/Papers/mac-slots.html https://www.euro-online.org/enog/inoc2007/Papers/m https://www.euro-online.org/enog/inoc2007/Papers/mac-slots.html

Finding the sweet spot for blood glucose monitoring

People with diabetes need to check their blood sugar levels on regular occasions, as their body is no longer able to produce the insulin needed to regulate these levels automatically and so they need to control them through their diet and insulin injections. At the moment, this checking mainly involves pricking a finger to release a spot of blood for analysis by a handheld device. Although this test can be performed regularly, it would be preferable to monitor blood glucose levels continuously, as they fluctuate all the time.

NIR spectroscopy is one technique that could potentially conduct this kind of continuous monitoring. This would involve implanting a tiny NIR sensor into the body to record glucose levels in blood or urine. The outstanding challenge is determining the optimum range of NIR wavelengths for determining glucose levels in blood or urine, as glucose does not have a strong interaction with a single characteristic NIR wavelength.

Although several research groups have tried to do this, many of them have come up with different ranges. Mohammad Goodarzi and Wouter Saeys at KU Leuven in Belgium wondered whether this lack of consensus was due to the groups using glucose samples with different compositions and then different multivariate statistical techniques to pick out the optimum wavelengths.

So rather than determine the optimum wavelengths by studying a single glucose sample with NIR spectroscopy and then analysing the data with a single statistical technique, they studied three different types of sample and analysed the resultant data with several statistical techniques. Their aim was to identify those wavelengths common to all the samples and techniques. The three types of sample comprised two samples of synthetic urine, which had slightly different compositions and glucose concentrations, and blood serum samples. They analysed the resultant data with seven different statistical techniques, including interval partial least squares and variable importance in projection.

As they report in Talanta, wavelengths of 1600–1700nm and 2100–2200nm were most commonly selected and thus appear optimum for detecting glucose levels. However, each of these two ranges worked best with different samples: the 1600–1700nm range proved optimum for the urine samples, while the 2100–2200nm range proved optimum for the blood samples.

Blog tags: 
Techniques: