The G5 was lower than the G6 a whopping 82.5% of the time!
Thanks. Got it.
Over the 126 readings the G5 was -7.5mg/dl relative to your meter and the G6 was 18mgl/dl relative to your meter. On strict (arithmetic) average the G5 was at 92% of the meter and the G6 was at 121% of the meter (that is an average of ratios).
So, on the percentages, that is a spread of .29 across a meter of unknown accuracy which, anyway, is reading some number of seconds ahead of both CGMs. Difficult for me to draw a conclusion.
Ok, here is a simplistic but statistically correct analysis of those 126 readings:
- Averages:
G5) The G5 read, on average, 98% of the Contour Next One.
G6) The G6 read, on average, 114% of the Contour Next One. - Standard Deviation (nothing to do with sex, this is a statistics thing):
G5) 16.74%
G6) 17.34%
Executive summary (most certainly required for those numbers): the two sets of sensors were self consistent; they agreed even when G5/G6 they disagreed. There are no significant differences between the two
Eh? I hear you say; the G5 was consistently [aka on average] 16.04% lower than the G6. Yes, but the standard deviation in either result was greater than that; this happens about 33% of the time (i.e. one out of three) just for a single figure.
So we don’t even know that @Eric’s G5 was reading low, or that his G6 was reading high.
We actually know nothing, but we have data and with data we can learn.
Just to add, to make it clear; @Eric has proved to me that there is no significant difference between the G5 and G6 readings.
So my very personal concerns about the 24 hour fluctuations are more than offset by the fact the darned G6 saves me from 2 hours of no readings for three days. Dang, my Omnipod expired 10 minutes ago and my G6 expired 45 minutes ago. I hate doing major body mods while I am trying to communicate.
1 Like
Same for my analysis. The G5 was less than Xdrip (in all three tests) 80%-90% of the time. This is astounding to me. And what’s even more astounding is looking at your daily data, There are enormous differences between the G5 and G6 from the same blood at the same time. I am about to get a Tandem Control-IQ. I wouldn’t want it to adjust my insulin based on these G6 numbers that I cannot prove were as accurate as the G5. Of course I can always use the Control-IQ as a dumb pump.
Did you do your analysis on Excel? Could you message me the data in this post in excel and I will drop it into my analysis which is slightly different than yours. Of all the conclusions I have reached on your data I guess the most important is that the G5 and G6 have given you substantially different readings from the same blood stick.
2 Likes
I could be mistaken, but shouldn’t you at least run some kind of statistical test other than comparing standard deviations before drawing that conclusion?
I am not sure I can PM you an excel file. But you can copy the data I posted above and save it as a *.csv file (comma separated values). Like a file named Results.csv or whatever you want to call it.
Then you can open the csv fle with excel and it will populate cells with those numbers.
1 Like
The whole point of the comparison was to compare the CGM’s to the meter at that particular moment.
I put a helluva lot more trust in what the Contour is telling me than any CGM.
Like when my meter said I was 101, but the CGM’s were telling me 151 and 137…I know when I am 150 vs 100. There is no question to me on the meter tests vs the CGM’s.
Eric,
A CSV format is fine. Thanks.
I am now going to look at the numbers and decide if I would want the basal IQ to make adjustments to my insulin based on the BG and the G5 readings and on the BG and the G6 readings. That is aT1D common sense way of determining if the differences are significant. They may not be but at first glance they appear will be to me.
1 Like
Lemme know what you determine!
I will. Have you sent the CSV file yet?
BTW my G5 was lower than Xdrip drip readings 71/29, 85/15 and 90/10% in the three comparison I did from Dec though Jan. That is significant to me especially as yours was 82.5/17.5 versus G6.
We are not able to send files. You will have to create it yourself.
Just highlight the values I put in above, in post # 40. Paste them into a blank Notepad file and save it. And then rename it to a .csv extension, instead of the default .txt that Notepad uses.
Thanks
That is the right statistical test; @Eric’s hypothesis was that the G6 read higher than the G5. This reflects the averages he quoted and when I repeated those calculations I got the same numbers. However that isn’t enough for any statistical statement; the question is whether the result was random or not. The statistical test for that - remember I am only talking about @Eric’s conclusions using his methodology - is to look at the standard deviations. Those deviations are both bigger than the percentage difference. I actually looked at the absolute difference too and the results are the same:
G5: -3.78mg/dl, deviation 15.99mg/dl
G6: +11.24mg/dl, deviation 15.42mg/dl
The null hypothesis is that the G5 and G6 both read the same as the meter. That hypothesis cannot be disproved any more than @Eric’s can using the normal levels of certainty statisticians look for - 90%, or 95%. @Eric’s hypothesis has an 80% chance of being right, the null hypothesis has around a 60% chance of being right.
I am looking at some other statistical treatments of the data. Unfortunately I am missing some information; in particular there are 3 G5 sensors and 2 G6 sensors, but I don’t know which readings correspond to which sensor. This is obviously important given that calibrations are per-sensor.
Eric, I did an analysis of your data which I believe would only be relevant for you. I am going to do the same analysis for me with data that I compiled. The following is not a statistical analysis. I do look at std dev, etc. all the time but I would base my decision on what I believe is best for you and me (G5 or G6 ) as a T1D.
This analysis assumes you tested your BG with a CGM and based on the CGM you put in a bolus to adjust your BG to 100. Although we do know what you meter BG readings were, assume we didn’t know them at the time. And although there are certain safeguards to Control-IQ, assume Control IQ put in a bolus that would adjust you to 100 based on your CGM readings. This is what I found.
You presented us with 126 readings. It assumes that every time you looked at your CGM readings you would put in a bolus if the reading was above 100 to get you back to 100. I was only concerned if your bolus adjustment would have caused you to not be at 100 but rather be below 70 (in the 50s or 60s).
G5. 2.4% of the time, or 3 readings out of 126, your bolus, which was intended to take you to 100, would have instead dropped you BG into the 50s and 60s. (1 50s, 2 60s).
G6. 7.1% of the time, or 9 readings out of 126, your bolus, which was intended to take you to 100, would have instead dropped you BG into the 50s and 60s. (4 50s, 5 60s).
Interesting observations. Despite your G5 showing lower reading than your G6, 82.5% of the time, you only had 3 instances (versus 9 for the G6) of the bolus adj causing you to fall below 70.
As mentioned previously, I would do the analysis again to see if you got similar results before making a definitive decision.
As soon as I get my G6 and Control-IQ, I am going to do the same comparison before deciding on using all the features in the Control-IQ. As I mentioned, this situation would never happen as there are safeguards to the Control-IQ and my assumption were illustrative only, But your analysis was fascinating and I agree that my analysis was not scientific and statistical.
2 Likes
For me the simple takeaway is this:
I compared the readings of the G5 and the G6 to the most accurate BG meter by almost any reputable report - the Contour Next/Next One. And about 62% of the time, the G5 was closer to the meter reported number than the G6. In a head-to-head comparison of readings, the G5 won 62% of the time.
Since I am on the G5 right now, there is absolutely nothing from these numbers that would make me switch to the G6.
1 Like
I would come to the same conclusion,
1 Like
Anxious to do my own g5 G6 test in a few weeks.
1 Like
As far as I know (not much, I admit), that’s not sufficient to draw conclusions about the statistical significance of the difference. I think you need to perform something like a t-test to find out.
1 Like
It doesn’t really seem necessary to perform a statistical test like that.
Eric’s study was simply for himself, right? Any statistical test you performed would also be specific to Eric and wouldn’t say anything about the accuracy of either sensor for everyone else.
The sample size would need to be much bigger in order for a statistical test to say anything like that.
For Eric’s purposes, it seems reasonable to me to analyze it the way he did.
3 Likes