Results! - The G5-G6 Challenge - An actual comparison

Tips, Tech & Gear C/GM
21

Sensor 1 died early towards the end of Day 9; Dexcom tech support listened to my troubles and agreed to send a free replacement sensor. So my second sensor ever on G6 was inserted yesterday evening and woke up to give readings just as I went to bed. All looked good, sugars were fine.

But, as happens occasionally (Okay, more like once every three days or so…) my sugars crashed just about midnight. Dex woke me up with “Falling” and also “Low Sugars” and I was in the 60’s. I did not worry much; day one after all; I happened to be wearing a 14-day Libre and my Libre Link app showed 139 and flat. But better safe than sorry. So I toddled down stairs, fingerstick etc and had a snack just-in-case. Contour Next said 136, in almost perfect agreement with LibreLink. But Dex continued stock show drops and now Libre was agreeing, with the fall rate.

As it turned out my sugars went as low as 82 per the Contour Next fingers tick; dropped to 63 very briefly on the Libre; and things started to rise and I went to bed.

One of the interesting take-away data sets is the comparison of Dexcom reading real-time vs the comparison of the ‘smooth’ data that carries forward (the data that shows up on the Clarity reports). Below is both the raw data listing, and a plot of the DEXCOM G6 data as it is published, and also as it is presented later, smoothed and pretty…

G6 Sensor 2 Low PNG
G6 Sensor 2 Low PNG885×1146 40.6 KB

4 Likes
22

Hi Eric, when will you post your G5/G6 comparison?

1 Like
23

I will soon! I gotta process all the numbers! Will take a bit of time to get them all put together.
:dizzy_face:

24

I have all the data from my comparison. I can do it by BG range too, or by particular days of the sensor. But here are the overall comparisons to start with. See my notes above to know what data I was using.

There was no cherry-picking. If I did multiple tests - like a 15 minutes apart or 30 minutes apart - I did not want to over-emphasize a BG value by counting it multiple times. I would always declare before the test and before looking at either CGM whether or not I would include it in my data. So I was trying to be very fair with it.

On the first G6 sensor I wore, I did not calibrate it. It seemed pretty bad. So for the second G6 sensor, I tried calibrating it. And it seemed a little better.

Overall, the G5 beat the G6 in pretty much every category.

I can pick certain days and BG ranges where the G6 was better (it was better in higher ranges, worse in lower ranges), but if I am looking at all days and all BG ranges, the G5 won.

If you want certain ranges or days, let me know.

image

image

image

The G6 seemed to be “over” my actual BG a lot, and the G5 was under a lot. Here are the higher/lower comparisons.

image

image

image

The G6 seemed to pick up rising or falling trends faster. But it just never settled on the right number as much as the G5. I did see up or down arrows sooner on the G6.

At this point I can’t see a reason to switch to the G6. I know eventually I will have to, but I won’t do it on my own at this point.

Anyway, lemme know if there are ranges or days for which you want to see comparisons.

10 Likes
25

Eric, thanks for the data comparison. Your analysis was not what many people have stated, that the G6 is more accurate.

So I have a couple of questions.

Average % difference. I assume you compared both to your Contour Next.

Did you look at the % difference for each finger stick and them sum and average the differences?

or

Did you sum the differences from the CN and then average the sum based on the number of readings?

% of readings better than the other sensor.

Is this a simple average of the number of times one sensor was better than the other?

or

Was this a weighted average, somehow.

If the sensor was better by 2 points or .3% did that count as one just as one sensor was better by 16 points an 15% would count as one or did you try to adjust by the magnitude of the difference?

Thanks

2 Likes
26

That is a good question, I should have clarified it better.

As an example, you can imagine a 50 point difference between meter and CGM when you are at 300 is not nearly as significant as a 50 point difference would be if you were 100!

So I did not want to use just the difference in BG points between meter and CGM. Instead I used a percentage.

So % difference is (absolute value of | Meter value - CGM value |) / Meter value

If your meter BG is 300 and your CGM says 350, that would be |300-350| / 300 = 50/300 = 16.7%

If your meter BG is 100 and your CGM says 150, that would be |100-150| / 100 = 50/100 = 50.0%

Same difference in BG points, but a much bigger percent difference in that example.

Then I averaged all the percentages.

So in the two examples above, the total percent difference would be
(16.7% + 50.0%) / 2 = 33.3%

Does that answer your question?

Here is a very small and simple set of data to show this:

Meter G5 G6
106 93 122
80 74 111
81 84 125
119 105 159
Average % difference
G5 G6
8.8% 35.4%
3 Likes
27

Hi Eric,

I couldn’t get to your numbers. Assume G5 had 2 differences from Contour Next of of 5% and -5%. If you combined them you would say G5 was perfect, the average of the differences 0%. But some would look at that and say the G5 averages a 5% difference but one was 5% over and one was 5% under. Did you look at the differences with Contour Next as the denominator. Just trying to understand what you did.

28

I used absolute value. So a 5% and a -5% would average out to 5%, not 0%.

The denominator was the BG meter value.

If my BG meter said 90, and the CGM said 115, I would take the absolute value of the difference and divide by the BG meter value.

So the difference between 90 and 115 is 25. That percentage would be
25 / 90 = 27.8%

For comparison, if my BG meter said 205 and the CGM said 180, the absolute value of the difference is also 25. An when I divide by the BG meter value I get

25 / 205 = 12.2%

So you can see a 25 point difference is bigger when your BG is 90 compared to when it is 205.

29

Great, that’s the way I would have done it so let me give you a few comments.

If you recall, I did an analysis of G5 and Xdrip versus meter and used absolute values and compared differences to the meter. So we are in sync.

In my analysis, G5 readings were lower than Xdrip readings 80%-85% of the time. You probably have the comparison of G5 to G6. What % of the time were they lower/higher compared to each other? You may have already done that as you compared them both to the meter. My conclusion was I’d rather have a CGM that had readings lower than the blood stick than higher. That would be more conservative in terms of not hitting a hypo. I would not want CGM to be higher than blood stick and find out I was actually very low.

I actually did the analysis of G5 to Xdrip three times. The first time Xdrip was substantially more accurate. The second time the G5 was more accurate and the third time they were about the same.
My view thus is that a single comparison even if was for three weeks is not conclusive. I also noticed that on average your readings were 13%-18% different than the meter. My tests were on average 6%-10% different than the meter. My point is not a criticism, It’s simply that everyone will be different and everyone should try it themselves rather than looking at an analysis you did on yourself. I’m sure you’d agree that if you were doing a comprehensive study you would do it on 100s or 1000s of people.

I think your analysis is great and as you said you could even share more data and conclusions based on others preferences.

I will do exactly the same as you when I move to the G6. It gives us some confidence on which is more accurate for us but I think we shouldn’t necessarily extrapolate that to others especially with the many comments we have seen that the G6 is more accurate for many other users.

Thanks for sharing. Was it fun? Or at least enjoyable?

1 Like
30

Did you download to Clarity? I would be more interested in general stats comparison instead of specific meter comparison. I have compared G4 and G6 downloads of every 5 minute reading, and found them quite similar when checking trends, or 15 min period averages. But I realize your goal was to compare to meter for accuracy, and I’m more interested in trend accuracy.

31

I have a percentage above showing how often it was lower/higher than the meter. You want to see how often they were lower/higher than each other?

I can pull that out if you want to see it.

Yes, I agree that my test was very small. And a longer test with more sensors and 1000’s of participants would be great.

But…I was really most interested in how it works for me. If it works better for 999 people, and I am the 1 it does not work better for, that is what is most important to me.

And everyone doing this for themselves would be great - except for the fact that it is hard to get G6 and keep your prescription for G5. Or go back to a G5 script after trying the G6.

So doing this type of comparison is not easy for a lot of people.

Shout out to @LarissaW for making this happen! Thanks!

1 Like
32

I did not upload it. But I looked at the graphs a lot, and they had the same general shape to them. The trends were pretty much the same. I never saw one going up and the other going down - except when it might first turn. Like one of them would turn before the other. But they were usually within minutes.

1 Like
33

Are you prepared to make the raw data available? (I.e. sensor times and the three measurements.)

34

If agree with everything. Hope you get lots of value out of this analysis.

If you would, it should only take a miniute to compare what % of the time one was lower than the other. Would love to compare it to my 80%.

1 Like
35

I’ll pull that out for you.

36

I did not write down the time. It was pretty much irrelevant to me, like 2pm versus 5pm or whatever. Did not matter to me in my analysis.

But I can tell you it was done over 19 days and was all the times I did a BG check at home, with the exception of multiple tests minutes apart which I did not count. So it was a pretty random assortment of times. I do not follow any pattern like “Oh, it’s time for my noon test…”

37

Thanks for sharing everything about the comparison @Eric, Very interesting.
I am very surprised.

My opinion was that the G6 was more accurate but that wasn’t based on measurements at all. It was also only based on testing under 70 and occasionally over 220 but only really if I was going to aggressively correct.

Your results seem like justification to add a nice honest looking 8.82% onto Time in Range values since starting G6 :smile:

1 Like
38

Ok, I can work without time; just one-two-three; [next]one-G5-G6. Are you prepared to publish that?

39

That’s similar to my analysis of G4 vs G6 running at same time.
Thanks.

1 Like
40

Meter, G5, G6
106, 93, 122
80, 74, 111
81, 84, 125
119, 105, 159
151, 116, 146
84, 88, 133
79, 68, 121
55, 58, 76
127, 112, 152
72, 82, 86
115, 106, 131
95, 102, 112
60, 61, 75
103, 82, 90
94, 82, 100
58, 47, 43
189, 154, 169
141, 137, 149
151, 164, 150
69, 86, 72
100, 93, 105
197, 199, 196
81, 67, 74
138, 118, 148
56, 58, 69
106, 107, 114
149, 122, 147
145, 132, 161
239, 231, 281
68, 80, 86
130, 128, 152
98, 96, 96
51, 68, 54
87, 97, 129
84, 88, 94
75, 79, 95
124, 122, 137
117, 129, 138
154, 147, 182
101, 151, 137
130, 102, 116
133, 139, 150
95, 85, 103
79, 80, 74
42, 58, 66
66, 62, 71
94, 83, 99
240, 206, 208
60, 54, 67
71, 68, 59
116, 112, 106
142, 109, 149
61, 51, 74
111, 98, 131
89, 80, 107
74, 64, 108
106, 100, 143
125, 88, 131
94, 87, 105
139, 156, 137
104, 144, 124
115, 127, 114
61, 92, 79
110, 126, 115
116, 104, 140
101, 80, 130
103, 104, 148
49, 40, 72
50, 60, 75
117, 138, 157
122, 138, 168
81, 84, 97
55, 64, 76
87, 80, 101
76, 82, 99
62, 49, 67
108, 105, 126
54, 43, 65
98, 95, 115
80, 90, 107
155, 126, 157
89, 76, 93
52, 53, 66
100, 92, 123
159, 157, 185
93, 74, 92
68, 76, 90
114, 119, 133
84, 85, 103
79, 78, 82
118, 139, 141
59, 68, 80
85, 81, 91
78, 62, 74
129, 90, 97
142, 126, 128
84, 85, 90
56, 62, 52
98, 76, 101
100, 99, 115
73, 74, 85
105, 98, 116
156, 148, 153
170, 119, 164
79, 84, 95
109, 111, 119
64, 40, 72
70, 72, 78
94, 100, 108
134, 120, 127
142, 126, 143
49, 63, 44
59, 63, 60
51, 40, 56
90, 49, 87
85, 111, 92
102, 119, 111
65, 42, 56
96, 68, 88
104, 103, 96
118, 138, 111
64, 67, 71
39, 40, 54
129, 123, 145
63, 41, 73
73, 71, 93

1 Like