The Seattle-based company published its peer-reviewed study in Diabetes Technology & Therapeutics Journal. The published results demonstrated that Know Labs’ proprietary non-invasive radiofrequency (RF) dielectric sensor and trade-secret machine learning (ML) algorithms correctly classified an individual’s glycemic status as hyperglycemic, normoglycemic, or hypoglycemic with 93.37% accuracy compared to venous blood glucose values–serving as an early proof-of-concept for a novel, non-invasive diabetes screening device.
Know Labs first unveiled KnowU, its wearable CGM set for FDA submission, in February. KnowU’s sensor uses spectroscopy to direct electromagnetic energy through a substance or material. Through this, it can capture a unique molecular signature. The technology integrates into wearable, mobile or bench-top form factors.
The company says it hopes to expand the potential application of KnowU beyond non-invasive continuous blood glucose monitoring. It believes that the non-invasive screening device could support underserved populations. Know Labs said it could facilitate early identification and intervention, potentially reducing diabetes-related hospitalizations.
“Early diagnosis and intervention for diabetes are critical to both improving patient outcomes and enabling healthcare systems to allocate resources more economically and efficiently,” said Ron Erickson, CEO and chair at Know Labs. “This proof-of-concept for the use of our novel RF sensor as a glycemic status screening tool indicates the device’s potential to help funnel previously undiagnosed patients more effectively into the healthcare system.”
A look at the Know Labs study
The study evaluated 31 participants aged between 18 and 65 with prediabetes or type 2 diabetes. Know Labs’ RF sensor continuously scanned participants’ forearms for up to two three-hour sessions each during a 75g Oral Glucose Tolerance Test. A third session saw water given instead of liquid glucose to act as a control.
Concurrently, patients received venous blood draws every five minutes, measured with an FDA-cleared glucose hospital meter system. The company trained a machine learning model to estimate reference venous blood glucose values on 80% of the data, consisting of 2,109 paired RF device and venous blood glucose values selected randomly.
From a total dataset of 528 paired values, the model correctly classified glycemic status 93.37% of the time as hyperglycemic, normoglycemic, or hypoglycemic. The model achieved sensitivities of 96.63% and 85.51% for normoglycemic and hyperglycemic cases, respectively. Specificities came in at 84.51% and 96.92%. Know Labs says it requires more data in the hypoglycemic range to evaluate sensitivity and specificity.
The company says its results support the accuracy of its proprietary non-invasive RF dielectric sensor and ML techniques. Know Labs plans to expand the application beyond proof-of-concept alongside potential strategic partners. It still plans to bring its non-invasive CGM to the market as well.
