Answers are grounded in this paper and cite their sources.
The importance of continuous glucose monitoring (CGM) usage in managing diabetes is clearly established. Barriers to greater global adoption of CGMs are economic costs, discomfort, and medical waste. Development of a highly accurate, non-invasive CGM that doesn't require disposables and that is economical for both patients and society would thus be of clear benefit. In this study we examine the efficacy of the novel Know Labs radiofrequency (RF) sensor for non-invasive blood glucose (BG) monitoring via a study design using venous blood as a comparative reference in participants with prediabetes and T2 diabetes. The RF sensor exploits the fact that the electrical properties of human tissue and blood change based on glucose levels. By scanning through several thousand radio frequencies in mere seconds, the sensor precisely captures this electrical signature. Ten study participants' forearms were continuously scanned during a 75g Oral Glucose Tolerance Test as venous blood was collected every five minutes using a peripheral intravenous catheter. We trained a machine learning (ML) model to estimate these reference BG values from the sensor data on 80% of the data (the 520 values in the "training dataset"), and then applied the model to the remaining, unseen 20% of the data (the "held-out test dataset") to assess its accuracy. Overall, the model produced 130 BG estimations on the held-out test dataset, allowing us to assess the efficacy of these methods. BG was estimated on the small held-out test dataset with a Mean Absolute Relative Difference (MARD) of 11.1 ± 2.1% relative to venous blood. There was no significant difference in accuracy between normoglycemic (11.0 ± 2.7%) and hyperglycemic ranges (11.5 ± 3.1%). A Surveillance Error Grid analysis of model accuracy showed 82.3% of estimations in Risk Grade A and 17.7% in Risk Grade B, indicating that no estimations fell in the higher Risk Grades. These results suggest that this novel sensor, paired with ML techniques, holds considerable promise for the non-invasive measurement of BG. If made widely available, the sensor could afford several population-level advantages, allowing patients to get real-time BG readings pain/needle-free and at a lesser cost – leading to more readings available to more patients with potential concomitant health benefits. Ongoing work will examine the ability of the sensor to be further miniaturized for wearable convenience, and to take accurate spot (in addition to continuous) readings. Outcomes of these efforts will inform the path toward clinical trials in pursuit of FDA clearance for this device.
No takes yet. Share an insight, caveat, or question.
Klyve et al. (2024) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: