Failed sensor and some new info
I am happy to report that Dexcom will be sending me a new sensor to replace the failed one. This procedure certainly serves to explain the high cost of the sensors; if the failure rate is relatively high, then it would drive costs higher overall.
To look this up I searched the performance information which is available at the back of the user manual. Indeed, it appears that 13% of sensors fail within the first 3 days of use (page 133, bottom right). That's a bit less than one in seven.
In fact, the chart indicates a 25% total failure rate on the first six days, and 19% on the first five. I don't know what the policy is at Dexcom for sending replacement sensors, but let me assume for a second they will send one if the current one fails during the first five days. Also, because of the cost of these things, some folks will try to take advantage of the system and drive the replacement rate slightly higher; let's say an additional 2-3% percentage points.
This means that practically, they are sending a new sensor out to replace a failed one once for each box of sensors. If the policy is six days, it gets even worse.
That's a terrible way to run a business.
My suggestion to Dexcom: if this is more or less right, and if the distribution of failures is more or less even, save yourself a lot of trouble and tons of admin and support costs and send an additional freebie sensor with each box by default, at least until you get a handle on the production side. Make sure it is clearly communicated that this freebie sensor IS the "first replacement" for a failed sensor, and track that internally per receiver. The communication side is going to be really critical here, you will need to make sure customers understand the change and what it means for them (while they are getting 5 sensors for the price of 4, it also means that when one fails, they don't get a new one shipped unless they have already used the freebie as a replacement).
While this will not entirely eliminate sending replacement, it will most likely reduce it by 80% or so; my guess is that the savings on admin and shipping costs would more than offset the cost of the extra sensor added to each box.
One thing for other users to note: if you tossed your packaging when you put in the sensor, and then it fails, and you are for the lot and serial number, remember that (1) the lot number is identical across batches (so all four sensors in a box have the same lot number); and (2) the lot and serial number are printed on the sensor itself. When you take it out, look on the inside and you'll see them in tiny black letters.
Something else that turned up as I was going through this issue. Apparently, recalibrating the sensor more often when it has trouble reading the signals is just going to confuse the system more. It's not only not helpful, it's a contributing factor to faster deterioration of its performance.
In practice, this means that when your sensor starts acting weird, for whatever reason, showing scattered reading that are less accurate, adding a few calibrations for good measure is in fact a bad measure. Instead, reset the sensor and see if things get smoother. The reason for this is that the system uses the last six sensor readings to compare against the meter calibration test, and when there is trouble reading the sensor (showing "???"), there are less data points and therefore diminished accuracy. Since I presume there is a statistical model hidden in there somewhere, the impact would be exponential, not linear. Thank you Keri from Dexcom for this piece of advice.
Thinking about this, I will also venture to add that calibrating during a strong trend (rising or falling) may also not as useful as during a stable period, as the trend in itself introduces increased variance (inherent in the rate of change in readings). This may be flushed out if the model is strong enough, but since I don't know the model I can't really say.
To keep each post more or less on topic, I'll end here and move on the next post, in which I will discuss a couple things that I haven't seen talked about much elsewhere.
To look this up I searched the performance information which is available at the back of the user manual. Indeed, it appears that 13% of sensors fail within the first 3 days of use (page 133, bottom right). That's a bit less than one in seven.
In fact, the chart indicates a 25% total failure rate on the first six days, and 19% on the first five. I don't know what the policy is at Dexcom for sending replacement sensors, but let me assume for a second they will send one if the current one fails during the first five days. Also, because of the cost of these things, some folks will try to take advantage of the system and drive the replacement rate slightly higher; let's say an additional 2-3% percentage points.
This means that practically, they are sending a new sensor out to replace a failed one once for each box of sensors. If the policy is six days, it gets even worse.
That's a terrible way to run a business.
My suggestion to Dexcom: if this is more or less right, and if the distribution of failures is more or less even, save yourself a lot of trouble and tons of admin and support costs and send an additional freebie sensor with each box by default, at least until you get a handle on the production side. Make sure it is clearly communicated that this freebie sensor IS the "first replacement" for a failed sensor, and track that internally per receiver. The communication side is going to be really critical here, you will need to make sure customers understand the change and what it means for them (while they are getting 5 sensors for the price of 4, it also means that when one fails, they don't get a new one shipped unless they have already used the freebie as a replacement).
While this will not entirely eliminate sending replacement, it will most likely reduce it by 80% or so; my guess is that the savings on admin and shipping costs would more than offset the cost of the extra sensor added to each box.
One thing for other users to note: if you tossed your packaging when you put in the sensor, and then it fails, and you are for the lot and serial number, remember that (1) the lot number is identical across batches (so all four sensors in a box have the same lot number); and (2) the lot and serial number are printed on the sensor itself. When you take it out, look on the inside and you'll see them in tiny black letters.
Something else that turned up as I was going through this issue. Apparently, recalibrating the sensor more often when it has trouble reading the signals is just going to confuse the system more. It's not only not helpful, it's a contributing factor to faster deterioration of its performance.
In practice, this means that when your sensor starts acting weird, for whatever reason, showing scattered reading that are less accurate, adding a few calibrations for good measure is in fact a bad measure. Instead, reset the sensor and see if things get smoother. The reason for this is that the system uses the last six sensor readings to compare against the meter calibration test, and when there is trouble reading the sensor (showing "???"), there are less data points and therefore diminished accuracy. Since I presume there is a statistical model hidden in there somewhere, the impact would be exponential, not linear. Thank you Keri from Dexcom for this piece of advice.
Thinking about this, I will also venture to add that calibrating during a strong trend (rising or falling) may also not as useful as during a stable period, as the trend in itself introduces increased variance (inherent in the rate of change in readings). This may be flushed out if the model is strong enough, but since I don't know the model I can't really say.
To keep each post more or less on topic, I'll end here and move on the next post, in which I will discuss a couple things that I haven't seen talked about much elsewhere.
posted by Blinkered at 12:09 PM
0 Comments:
Post a Comment
<< Home