Making no apologies for ripping off Yazz songs, that somes up what I’ve found over the last week of data. But first things first.
I took off the old sensor yesterday once it had expired, fully expecting to see some form of welt on my arm. Fortunately, there was nothing showing, and the removal process was as difficult as it had been previously, suggesting that the lifting around the edge of the sensor really made no difference to the stickiness of the module itself.
But what of the data from the sensors themselves? How did that all work out? First up, here is a comparison of the data that each of the two sensors read between 10/01/2015 and 15/01/2015.
This doesn’t look disastrous at first glance, but there are some noticeable gaps and inconsistencies between the two data sets with regard to the trends they are displaying, most noticeably towards the end of the experiment where it is possible to see the new sensor getting completely confused.
If we look at the differences between the two sensors plotted on a graph throughout the period of the test, we can see that there is variation between the two, although it’s not a huge order of magnitude. The peak difference was roughly 3mmol/l, although there was a clear trend towards around a 1.5mmol/l variance.
There does seem to be some correlation between higher or lower readings and the variation between the sensors, but the context of the higher/lower readings needs to be observed. These are typically not particularly high or low absolute glucose levels, just higher or lower in the overall trend.
But this is comparing two sensors with each other. It’s clear from the graphs that sensors reading Interstitial Fluid on different arms may not be seeing the same thing, or there may be manufacturing differences between these two sensors. The other point of interest is that during the data collection period, the two devices’ clocks seemed to run at slightly different rates meaning that the data became misaligned. This has resulted in the gaps shown in the above graph.
The other big question is how do these two compare to the blood glucose tests?
This is perhaps a more relevant question, and the data set is also significantly smaller. Over two days I did a number of fingerprick tests to gauge what the variance of IFG was from BG. This is what it looks like:
The associated sensor readings are taken 20 mins after the equivalent bg reading. What’s very noticeable is the distance that the sensors are reading from the blood glucose levels, and the variation of variance if you like. Looking at some statistical indicators (and Standard Deviation is not a good measure for blood glucose under normal circumstances, but gives an indication of the consistency of the two data sets), we get the below table:
What’s interesting in this view is that we can clearly see that over the small sample used, the new sensor was typically reading high and the old one low. Based on mean alone, the new one was 8% higher and the old one was 15% lower.
On the other hand, the standard deviation of the new sensor at 2.2 matches the blood tests, whereas the old one seems to have been closer, suggesting that the old sensor was perhaps understating level changes when compared to bloods.
But these two deal with absolutes. What’s interesting is the percentage variation, as Abbott considers a reading that is “More than 15% away from a group of blood tests taken at the appropriate time” to be erroneous. Let’s take a look…
Over the testing period, neither sensor really shone out as a beacon of accuracy. The older sensor clearly has a significant number of readings below the 15% variation, and the newer one has its fair share, but on the upper side. What might be of more concern is that the wide variation appears not to have occurred during periods where the blood glucose levels were moving rapidly. One might expect it in those circumstances. In periods of moderate or flat movement, it strikes me as more of an issue.
What is worth noting for users of the Libre was that it was rare for either sensor to be within 5% of the blood glucose tests during the two day test period.
So what has the experiment shown? I think the key really is the comparison of sensor data against blood glucose data. While it is based on a small sample, these two sensors often produce results that are not terribly close to the real blood glucose levels in percentage terms. The key question to ask, though, is why?
During the course of yesterday evening, after a High Intensity Interval Training session, I noted that the glucose levels in my interstitial fluid increased at a far greater velocity than those in my blood. I’m sure there will be a good reason for this.
I’m not sure that this has provided any clear answers. For me it shows that there is a clear variance between blood glucose and interstitial fluid glucose levels that is not apparent from the Abbott marketing material. I still think the trend data is useful, but for making decisions on insulin or glucose, the blood test meter should always be used.
I understand that there are some competitive products in the pipeline from other manufacturers, so I’m looking forward to seeing what they claim, what they do, and how they compare.