It’s that time of year again when folks start whispering of the next “holy grail” solution for non-invasive glucose monitoring. With continued inbound requests coming into Valencell asking for this snake oil, I feel it important to summarize some of the key questions one must ask before placing a modicum of trust in any non-invasive glucose monitoring claims.
Phrased another way, you need to ask yourself a few questions about the proposed solution first before proclaiming or promoting that someone has killed the “deceitful turkey”. I’ve summarized a few of the most important questions below:
1) What’s the use case?
In the case of glucose monitoring, the only application that makes money today is monitoring glucose for the purpose of dosing insulin. Because most researchers in the field will at least admit to themselves – but sadly not to their investors — that achieving this is impossible, the only way they can live with themselves is by thinking, “Well, at least this method can possibly measure glucose spikes! And that would be valuable.” Yet while some of these proposed methods can indeed assess glucose changes, and while there is some value to that, the dirty little secret is that one can assess glucose changes using advanced PPG analytics and activity context data alone. There’s no need for a fancy new sensor. Moreover, while assessing the existence of a glucose spike can indeed be quite useful to those managing diabetes (or even prediabetes), it is not nearly as marketable as displacing the ancient finger-prick method for dosing insulin. Indeed, although the global market for finger pricking glucose monitors used to dose insulin is double-digit billions per year, I am not aware of a single device in this marketplace that markets itself as a “glucose spike” detector!
2) What’s the sensing modality?
Or rather, how is the glucose being sensed? This is a critical question, as the sensing modality places a fundamental limit on the accuracy, precision, repeatability, and universality of glucose monitoring.
An example of the importance of sensing modality was widely publicized throughout the wearables community in April 2018: “Non-invasive, transdermal, path-selective and specific glucose monitoring via a graphene-based platform”. The paper published in Nature proposed a sensing modality comprising a hair follicle-based path guiding mechanism, using electroosmosis, to estimate glucose in interstitial fluid. It’s important to note that it is not blood glucose that is directly measured in this case but rather interstitial glucose. It’s well-known that estimating blood glucose from interstitial glucose requires a routine calibration, as the transfer function between blood and interstitial glucose is not only personalized (unique to the individual) but also time-varying. Thus, the sensor modality is particularly important in this case because there is a fundamental limit on how accurate this modality can be for measuring blood glucose levels well enough to dose insulin.
3) How do you define “non-invasive”?
It is unfortunate that this Nature paper referred to the sensing modality as “non-invasive”. An important question to ask is: What does “non-invasive” truly mean? In my professional view, “non-invasive” should imply that nothing material is unnaturally transferred in or out of the body for the testing. For example, a sweat-based approach would be non-invasive if it were measuring sweat naturally generated by the body, but forcing sweat out with pumps or electrostatics would not be, as this would represent an unnatural, invasive transfer. The same goes for sampling interstitial fluid.
4) What is the quantification mechanism?
Quantification is the always-overlooked factor where, 9-times-out-of-9, non-invasive glucose proponents inevitably fall flat. Namely, even if one could directly sense blood glucose non-invasively, with no material item unnaturally going in/out of the skin, how would one gauge the sampling volume? Because volume changes as x3, where “x” represents one of 3 dimensions, even a small under-/over-estimate in just one dimension can lead to dramatically different results for glucose concentration… thereby killin’ any chance one can measure it well enough to dose insulin. Indeed, how would one teleport blood out of the blood vessels into an external volumetric glucose sensor? I present to you that if one were to invent such a teleportation device, they would make more money teleporting gold from Mars to Earth rather than teleporting blood from inside the body to an external wearable device.
5) What is the required measurement accuracy?
Of course, the accuracy of sensing is just a means-to-an-end for the accuracy of the measurement. The measurement accuracy has to be “good enough” for the use case. And in the case of dosing insulin, “good enough” is a tough target to hit. Demonstrating a linear relationship between blood glucose and sensor outputs is not enough. Ultimately there needs to be a measurement of glucose concentration, requiring a reliable, universal physiological model to be developed and validated.
6) How reliable are the results?
One-time success doesn’t cut it! Not only must one show accuracy in enough people to satisfy the statistical power requirements, but one must show that the measurement method is repeatable. In fact, I wouldn’t trust any biomedical sensor claim, no matter how large the human subject size, unless the experiment had been repeated at least twice, on the same subjects, and had been shown to be repeatable.
7) Is it universally applicable?
It’s essentially a given that any non-invasive sample quantification mechanism will require individual calibration to achieve the highest accuracy. Calibration is required to accommodate for individual differences and time-varying changes in the relationship between the sensed species and true blood glucose levels. The key questions are: How is this calibration done, and how often? Phrased another way, in the case of non-invasive glucose monitoring, how complicated and involved would individual calibration be? As an example, for the case of Dexcom’s continuous glucose monitoring “G5” device, a calibration is required multiple times using blood samples as the benchmark (Note: Dexcom’s G6 claims no calibration required, but the vast majority of CGM’s on the market still require calibration). Thus, while existing solutions reduce the number of finger pricks needed in a day – a major benefit for those with type-II diabetes — you still need multiple pricks a day. By definition, I view any such solution as minimally invasive. A truly non-invasive solution can perhaps get away with a one-time calibration session, but anything requiring continued pricks is minimally invasive at best.
8) Is the user experience better than what’s available today?
The insatiable drive for a non-invasive glucose monitoring technology is fueled by 2 painful facts: a) drawing blood hurts and b) drawing-blood is time-consuming and distracting (from one’s everyday living). Both of these pain-points have discomfort at their center. Indeed, one can argue that the whole “roux of the gumbo” with respect to non-invasive blood glucose monitoring is to make it much more comfortable for those with diabetes. Yet many of the proposed solutions for non-invasive glucose monitoring, by the time they’ve adapted their protocols to get something with close to “good enough” accuracy, are just as uncomfortable (if not worse) than a finger prick. I recall on occlusion spectroscopy approach which, in addition to simply not being accurate enough, required a ~5-minute process to measure a person. All of you fellow ADDers out there will agree that waiting still for 5 minutes to take a measurement is much more painful than a quick finger prick! This means that non-invasive glucose developers should always remember that it isn’t just about avoiding the prick, it’s about the overall user experience.
Hopefully my thoughts help provide some valuable perspective when the next hyped-up non-invasive glucose monitoring solution hits the blogosphere. I’d love to see an honest solution for truly non-invasive glucose monitoring, good enough to dose insulin, enter the marketplace with smashing success. And frankly, I’d even love to see a truly non-invasive blood glucose screening technology in the marketplace, which would passively monitor blood glucose “good enough” to recommend that the user take a blood draw or other invasive/minimally invasive measurement to dose insulin. But with so many garbage headlines in the general media — even within peer-reviewed journals — it’s an uphill battle to convince me that we’re close to either of these 2 compelling use cases.
For those interested in more detail, here are two great resources for further reading:
- The Pursuit of Noninvasive Glucose: Hunting the Deceitful Turkey by John Smith
- Just Beyond our Fingertips by Kate Bechtel