Saturday, March 21, 2015

The tremendous impact of sick sick days...

Like every parent, we've previously had our share of sick days: runny nose, flu like syndromes... The type of vague viral infection that doesn't warrant antibiotic therapy. While I always kept an eye on them, just in case, they didn't seem to have a significant impact on Max's BG levels. If they had one, it was lost in our daily variations.

This time was different. Max came back from a school trip in the UK with a runny nose and a slightly sore throat. For the first three days, nothing seemed different from the past innocuous events. Then, on the morning of the fourth day, the symptoms worsened. There was a bit of fever, the cough became rougher and wetter. Most tellingly, perhaps, Max lost interest in his smartphone and started dozing through the day. The time had come for a visit to the family MD and some amoxicillin.

Rough day -1

Here's his BG profile for that day. Slightly higher than our typical profile, but nothing alarming. There is a dawn phenomenon starting a bit earlier than usual, corrected at 3:45 (blood check and re-calibration visible), wake up at 7:00 (the correction was perfect) the breakfast spike, the lunch spike and the dinner spike. The total insulin dose for the day was around 34 units, including the correction, or about 0.64 U/kg.

Rough day

The next day was "interesting", to say the least. Max woke up, decided he was unable to eat and immediately headed for the couch where he basically slept through the day. The slope of the (very early) dawn phenomenon was steeper. The more aggressive correction at 3 AM only had a very short term effect. At 8:00 AM, BG confirmed value was at 250 mg/dL. This is why we decided to dose exactly as if he had eaten even though he had not (5U novo). As soon as that dose had run its course, his BG started to rise again at 11:00. At that point, having seen how resistant the 'high' was, I decided to take a strict control stance. We corrected at 11:15 and then again at 13:30 when it became clear that the BG was staying high despite, again, zero food intake at noon. Around 4:15, Max woke up a bit and moved around, immediately triggering a fall. A single 2 grams glucose tablet was sufficient to prevent what looked like an unavoidable low. And then, BG started its slow rise again, which we corrected again, and again and again.

Total food intake for the day: 2 grams of carbs and lots of water.
Total insulin for the day: 50 units (close to 1 U/kg), a 50% increase.

I thought I knew what to expect...

but to be honest, I was a bit surprised. I expected I would have to keep giving insulin even if food intake was low or non existent. Max is on MDI and I expected the Levemir to keep things relatively under control during what was, after all, a relatively benign infection.

If we had applied a stupid rule such as "no food, no fast acting insulin", my model predicts Max would have shot above 400 mg/dL before noon. I can't be sure it would actually have happened (and had no desire to test) but my model is very often quite accurate, as shown by the perfect correction the day before. Notes: no paracetamol was given at any point, around 1gr of aspirin was used per day for headaches, pre-correction and general levels were checked by BG meter to exclude any eventual sensor inaccuracy.

If I had applied a slightly less stupid rule, such as "keep giving the normal doses even if he does not eat", we would probably have spent the whole day in the 300 mg/dL range and that would have been a problem. Why?

The problem

Well, we have previously established that Max renal threshold is around a fairly 250 mg/dL. Any extended period above that range and glucose starts leaking in his urine. We have also established that somewhere between 250 mg/dL and 300 mg/dL his BG becomes much harder to control, especially if he starts to exercise a bit. Whereas aerobic exercise under 250 mg/dL always leads to a correction, it does not work well between 250 mg/dL and 300 mg/dL and not at all above 300 mg/dL. In that case, we have seen extremely steep increases rather than controlled decreases. This is again expected and fairly typical as acidosis and possibly ketosis show up.

We did test for ketones and had none (thanks to the strict control) but I can't help thinking that, had we taken another course of action or had the information a CGM provides, we would have experienced additional difficulties. The "no food, no bolus" rule, even if we are supposed to be somewhat covered by the morning Levemir, would have almost certainly brought us to the edge of DKA.

I won't dive into the DKA mechanisms in this post but, generally speaking, the vicious circles of acidosis and ionic disturbances are what makes the initial phase of DKA dangerous (correcting too quickly is what makes the correction phase dangerous). Those disturbances can lead to arrhythmias and, at the extreme, sudden death or death in bed.

And that is understandable. Here is a ECG I took from Max, as he was sleeping on the couch. He has a typical rest rate of 60 bpms. At the end of that no-food day, with a body temperature around 38C (fever, but not severe), he was pumping around 125 bpm. It seems to me that, had the fever and acido ketosis been severe, a very real cardiac risk exists even in relatively benign situations.  

It is one thing to read about such risk from a theoretical point of view. But seeing how hard it was to keep BGs in a relatively correct range, despite no food intake and the relative protection of MDI based therapy, real life was an eye opener.

Conclusion - rough day +1, rough day +2

Max has now recovered. Here are the profiles of the two following days. Amoxicillin was effective. Average BG normalized itself progressively Food intake resumed. And we were immediately able to lower the insulin doses back to our normal schedule.

And finally, Friday afternoon, after 72 hours of total or relative inactivity, Max started moving again and we started to see some hypos again. 

To be honest, the whole week that followed the sick week was a bit peculiar. Nothing dramatic, but a recurrent tendency to go in hypo faster than usual. A plausible reason for this would be that
  • the glucose that was mobilized during the sick period had to come from somewhere since it did not come from food.
  • those mobilized reserves had to be rebuild during the recovery period.

Plausible, but speculative at this point as I have no hard science references to explain the phenomenon. Lastly, it seems that Max needed more time to recover fully from that infection than he needed before he became T1D. It is highly subjective: I just noticed his tennis and endurance had taken a big hit when he resumed training.

Friday, March 13, 2015

A mixed bag...

Abbott Libre Update

I keep receiving private questions about the technical aspects of the Abbott Freestyle Libre. Yes, I know a bit more than what I have posted publicly, but what I know isn't really earth shattering. If I come up with something that is new, genuinely useful and can be shared publicly, I'll be sure to report it here. But right now, our family in France seems to be experiencing... difficulties in obtaining sensors. I am not reading too much into this (update 14/03/2015 - apparently resolved - could have been an ordering snafu as a few other users have experienced). A couple of the blog readers have offered to send me sensors so I could go on in my investigations and I thank them for that: I would have accepted if I thought I could make significant progress that could be of use to the T1D community but, at this point, this is unlikely.

As far as the performance of the sensor is concerned, most of the sensors we have used were well behaved and performed in line with our previous experiments, in other words, much better than the Dexcom G4 (non AP) sensor.

Here is one trace from a recent tennis game where, again, you can clearly see both sensors in action and the Dexcom's algorithm delay is clearly visible.

We also had a weirdly defective sensor that was always too low between a certain threshold but behaved perfectly in the ranges above that threshold. In a way, that was one of our most interesting sensors as far as providing technical insights was concerned. This "lowish" behavior is quite visible in the "official" chart below.

Acute DKA and death.

This being said, Abbott's PR department seems to currently have bigger worries in terms of image and possible financial damage with this unfortunate story.

From what we know of the technical issue, it seems that someone somehow either forgot to enter or or failed recognize the proper calibration curve of the strips used in a third party device (Omnipod). At first sight, this looks like a software/firmware bug that had fatal and extremely unfortunate consequences... I won't comment in details about this story but regardless of the outcome of the complaint, the main lesson for us is quite clear: do not blindly trust the technology. A lot of people seem to assume, possibly somewhat misled by excessively optimistic advertisements, that devices should behave flawlessly. That's not the case: technology fails, especially in a cutting edge field where it meets biological constraints and human psychology. Abundant information doesn't necessarily means abundant truth. Keep thinking folks. The diabetes technology isn't nearly as accurate as pocket calculators have been for 40 years. The time to forget mental arithmetic has not come yet.

I was also surprised by the discussions of that case that I saw in the Internet forums: a lot of PWDs seemed to be unaware that DKA can kill and kill quickly (another example here). This is probably something I'll get back to as well as my son is just recovering from a severe pharyngitis that had, for a few days, a spectacular impact on his blood glucose levels and treatment. An unfortunate situation but a perfect illustration of why and how, if left unchecked, a fairly standard situation can become dangerous.

Dexcom sensor update

The current lack of Libre sensors drives us back to Dexcom... Unfortunately, the news haven't been too good on that front either. We either have been extremely unlucky or Dexcom is experiencing some difficulties in keeping up with the demand. Our last sensors set contained one sensor that, despite a correct insertion, kept reporting "0" as raw values (and of course failed at once from the point of view of the Dexcom receiver), another that kept reporting an impossibly high and stable raw value. That one was also reported as failed by the receiver but it took a bit longer.

And finally, the icing on the cake, if I dare say so, was this inserter... It apparently came without a needle and a sensing wire, its guide tube floating almost freely in the upper part of the device

This was a bit too much, even for a careless guy such as myself, and I did request a replacement this time. As I understand it, the replacement policy applied by Dexcom's distributors isn't as tolerant as Dexcom's own policy and I had not even bothered complaining until now. I don't blame them as I suppose they might be a bit more constrained by Dexcom's business terms. In my case, the exchange went without a hitch and a new sensor was sent as soon as I complained. Our local distributor has, so far, done a perfect job in terms of deliveries.

But there are more problems on the horizon. I was told a few weeks ago that we would not get the 505 algorithm update in Europe because Dexcom would soon be releasing exciting new products. Well, the exciting new products were indeed released but... we are unlikely to get them either (at least officially). This probably boils down partly to the time and effort required to get approval (eventhough Dexcom did get pediatric approval in Europe before US) and to the possibly inherently illegal strings attached to their data sharing plan.

Caught between a rock and a hard place: xDrip to the rescue.

This leaves us caught between a rock (slow approval/availability for Dexcom's new products) and a hard place (apparent obstacles in ordering sensors from Abbott)... To put it bluntly, once you have used well behaved Libre sensors, it is extremely hard to remain satisfied with the performance of the G4 non-AP system. (in a way, given some of the issues I have with Abbott's behavior, it hurts a bit to report that, but I strive to remain objective).

Are we in a hopeless situation? Fortunately not. xDrip is a promising open source project that offers, at least partly, a solution to our problem. Our initial tests seem to show that its performance is closer to a good Libre sensor (and to the 505 AP Dexcom algorithm should be) than the standard G4 non AP system. It also offers all the advantages of a full CGM setup (constant monitoring, alerts, remote uploads). I suspect it will be somewhat constrained in terms of speed by the sparse 5 min Dexcom sampling, but we'll see. That project probably isn't for everyone and is not, for those who live on the other side of the ocean, FDA approved. Unlike the standard version of Nightscout, it goes a bit further than the simple replication of the Dexcom data by using its own calibration curve. If that scares you, even slightly, stay away from it. It is worth noting however that sensor calibration is a very common technique in analytical chemistry: while it may seem to be rocket science, it definitely is not as it only requires high-school (or possibly 1st year college depending on where you live) mathematical techniques. That simplicity means that it easy to understand and analyze.

That's all for today. Stay tuned for more.