Friday, January 29, 2016

Visualizing two months of diabetes...

How am I doing long term? I switched insulin, did it change something? Has the change of composition in my breakfast really had an impact? Typical questions we are probably all asking ourselves.

They can be answered by looking at period averages, number of hypo or hyperglycemic excursions. SD evolution, or whatever variability marker we fancy. Most people don't bother and not everyone knows what SD is (or actually needs to know to lead a happy life). Our anecdotal memory doesn't help much either: we tend to find spurious correlations everywhere we look or give too much importance to memorable incidents.

As far as I a concerned, I find that visualizations can greatly help getting the message through. Here's a plot of 60 days of Max Dexcom's readings. The first 20 days are school days: control is obviously not optimal. Exams do not help, but the "hypo safety margin" is necessary. Days 20 to 40 are holidays, at home, where Dad can influence the control a bit more. Five minutes on an ergometer help tremendously. The New Year's Eve meal is a lonely peak right of center. Days 40 to 60 are again school days. It seems some of the lessons of the Holidays have been remembered. Full size plot available here.





Monday, January 25, 2016

Dexcom G4 (505): likely trauma drop and recovery


A quick post, just because I found this example particularly clear and interesting.

Context


Tennis Saturday, late wake up, sensor calibration a bit before 10:00 AM. That sensor has behaved extremely well, with a running MARD around, or even below, 10%. There's a breakfast spike, a correction that is a tad high, but that we let go. The late low carb lunch comes at 13:30 (trying to time our small Novorapid injection optimally for the tennis). We do a small warmup on a bike at 14:50 to avoid an early BG climb. We start carb loading at 15:30, with 10 grams every 15 minutes from that point (typically: 15:30, 15:45, 16:00, 16:15, 16:30...)

Issue


Tennis starts at 16:00 – we see a huge drop (50 mg/dl in one reading) at 16:25. We stop and double BG test at 16:30: 95 mg/dl. The sensor goes to LOW and starts beeping. A new BG still tests around 90 mg/dl while the sensor is now in LOW. At 18:30, tennis is over and, since the sensor is stable, we recalibrate. We are still around 100 mg/dl. The algorithm picks up the new calibration factor almost at once. That would have been very different with the G4 non AP: a new significantly different value would have had a variable impact depending on the previous values the calibration table held (or even the number of calibrations in the table if the sensor had been recent)

Since we’ve not seen an evening meal rise (frequently absent after intense exercise), I suspect a delayed hypo is in the book, which is why I double check and re-enter a calibration just in case. The guesstimate was spot on: we go down the slide. The sensor then accurately tracks the real delayed hypoglycemic even, which we correct (duh!).

Follow up


Sensor has now resumed its accurate behavior, initially with a clearly decreased conversion factor that seems to be creeping up a bit.

traumadrop


Friday, January 22, 2016

Article reviews 1: DIABETES TECHNOLOGY & THERAPEUTICS Volume 18


‘DTT’ has made its full special CGM issue available for free here. If you are a typical CGM user, this is a great opportunity to see how professionals think about your favorite diabetes management tool. I plan to review, summarize and comment the articles I have read in full.

The Future of Glucose Monitoring by Satish K. Garg, MD – is an introduction in which the author states that he expects CGM to replace BG Meters at some point during the next decade. That is obviously my wish as well but a couple of major obstacles remain: cost, of course – which is why we need more competition in the field sooner than later – and the remaining odd behavior/calibration issues I have covered on this blog.

The meaty part begins with in Continuous Glucose Monitoring: A Review of Successes, Challenges, and Opportunities, where David Rodbard, MD addresses those issues and more. A few words on some of the points he raises

Lack of approval by the FDA for dosing – is definitely an issue for the “establishment”, less so for most users who usually develop their own reasonable strategies. On that point, I have mixed feelings. On one hand, the recommendation NOT to dose insulin based on CGM readings is baseless as a well working CGM trend provides more information than a single blood measure. But, on the other hand, a blanket dosing approval remains dangerous if it is followed blindly as long as we have calibration, compression or temperature compensation issues.

Cost is a no brainer. The technology has to get cheaper before every T1D or social security system can afford it. Cost will go down if there is real competition in the market. Real competition will come with more players, a more open architecture, non captive data, less patent obstruction or agreements not to step on other people’s toes… Many conditions. I am not too hopeful in the short term on that side…

Need for recalibrations: absolutely! Calibration MUST DIE (even if that means I have nothing left to write about). Chained measurements are a strong as their weakest link - although oversampling can solve some of those issues -and the user itself and his environment is a great source of errors.

Periodic replacement of sensors: again, yes. Cost, of course. And not so much for the inconvenience itself – T1Ds are tough -, but mostly because the hours following sensor insertions remain a large source of inaccuracies… which can take huge proportions (essentially, pick a random slope) when combined with user calibrations.

Day-to-day variability in glycemic patterns: are cited as an obstacle because they limit the predictability of findings in masked professional use. This is the only weak/sub-optimally argued point in the article I think: day-to-day variability may be an obstacle in up-sell of service by professionals or companies but day-to-day variability and unpredictability of patterns is the strongest argument for CGM use as far as patients are concerned. Patients, let me apologize for them, do not live in the fairy land world of nice meal curves, slow/fast carbs and optimal basals where  non-CGM aware endos imagine they are. 

Time, implicit costs and inconvenience of uploads: a small issue in my opinion but that can become very annoying for practices that have to upload data from multiple devices in multiple proprietary software. That’s a VERY easy issue to solve though, and here is my recipe
  • regulation agencies: please do not allow proprietary and closed formats!
  • do not turn a blind eye or let covert data theft fly under your nose.
It so simple that it should not need to be restated, but let me do it once more. Commercial entities are all in love with Apple’s business strategy: the closed and walled eco-system in which subscriptions, planned obsolescence and mandatory upgrades bring tons of dollars. That model is absolutely outrageous in terms of healthcare but they want to replicate that. Your own interstitial glucose data stream is now licensed back to you (as I feared earlier) and you become a captive subscriber to your own data. Your doctor then also soon becomes the captive of the system and, unless he wants to spend significant time developing his often sorely lacking IT skills, becomes the one that chooses your prison. Granted, the walled garden has already been somewhat present in the field of drug prescription where the MD’s choice is often oriented by marketing, deals with social security or insurances. But if the drug does not work, or if there is another better option, a good MD will move his ass to make sure his patients received the best care. In this case, moving data around, when the “owner”, who is not the patient, does not collaborate or actively prevents it is going to be almost impossible.

Reimbursements for physician time, inexperience and lack of training go hand in hand. However, I do not see that as a huge obstacle. An obstacle to the acceptance of the technology by the medical establishment, yes. But, even though I was trained as a MD and I fully understand that MDs have to take a central (and possibly authoritarian) role in some circumstances, management of a chronic condition such as Type 1 Diabetes is not one of them. You are the one in charge, you are the one who needs to develop skills and the occasional quarterly MD advice should be seen as coaching. This being said, the author makes a very good point about how automated and standardized analysis of data would solve a lot of issues. Once again, open access and open standards are sorely needed. Rodbard uses the ECG analogy but does not push it far enough... A good question could be
Where would ECG technology be today if we had started the field by “licensing real time feeds of patient heart rates”, patenting infarction and arrythmia detection algorithms or “methods to transcribe electrical heart signals on paper by means of a moving pen” ?
Lack of standardization of software methods for analysis of CGM data, is mildly annoying. Again, it is an issue that can be solved by open standards and an issue that will certainly not be solved if proprietary standards and systems are allowed to proliferate.

Clinical guidelines won’t hurt but don’t matter much from a patient point of view. A caring and concerned physician will remain a caring and concerned physician. The indirect beneficial impact of clinical guidelines might be to push technophobic doctors into the right direction.

A few additional comments on the article content.
 

In “Lag time of interstitial fluid glucose relative to blood glucose”, the author rightly notes that the situation is improving through better algorithms (a possible allusion to the Libre improved response time). That is, by the way, an area where Dexcom is sorely in need of improvements and probably lagging the competition. The real life benefits of a higher sampling frequency (as long as the signal is good enough) combined with possibly predictive algorithm is tremendous.

In “Confusion regarding interpretation of glycemic variability”, Rodbard again hits the nail on the head. The different variability indexes circus serves no fundamental purpose. It is basically a sterile and alimentary publication generator machine. Of course, the author makes what I interpret to be the same statement in a much more polite way: “Nearly every measure of glycemic variability is very highly correlated with the overall or total SD”
    Facetious parenthesis:
    • “very highly correlated with SD”  is a nice way of saying much if the work done in this field as led to absolutely no significant result.
    • “nearly every measure”  avoids vexing the 50 or so colleagues who published on the topic as each of them is allowed to think his measure is the one that is meaningful.
    Rodbard mentions that CGM accuracy has now reached the 10% MARD threshold that has been demonstrated (on simulated patients) to be sufficient by Kovachev. That’s essentially restating the manufacturer’s claims, claims that have partially been replicated in clinical studies… and that should, in my opinion, be taken with a grain of salt. Rodbard partially addresses this issue in “Confusion regarding reporting of accuracy and precision of CGM sensors” and suggests some standardization here as well. This is indeed necessary but will not, in my opinion, have a major impact as long as external user calibration is required.
    • I saw a lot of “horrible” things in the few years of G4 ‘(non AP) CGM data users kindly sent me. While this analysis suffers from many limitations, it was quite clear that during the first 10 to 24 hours, the CGM results were a calibration timing lottery at best, a hopeless issue if the sensor/sensor wound did not cooperate. The G4 (non AP) was clearly not performing up to specs in the standard scenario.
    • calibration during compressions could cause huge discrepancies as well. Even if that problem can partially solved by black box “silent rejection” of calibrations in those circumstances, it clearly was not implemented at that time. 
    • users don’t have access to the very accurate BG meters used in clinical studies. Calibrating a perfect CGM with a couple of readings taken by an 15197 ISO2003 compliant meter is a very different situation. Here’s a somewhat theoretical view of the relative performance of BG Meters according to standards (and excluding user errors)
    1milliontests

    Here is what 100.000 simulated perfect CGM runs would look like after an initial double calibration with an ISO 15197:2003 compliant meter (red) vs an ISO 15197:2013 compliant one. The central dark line is the perfect measure of the ISIG

    100000
    • failing sensors (which at one point were estimated to be 1 out of 5 on a significant sample of early Libre adopters) never seem to make their mark in studies. It could either be that sensors used in clinical tests are cherry picked by the manufacturer or that the failing ones are somehow masked in the “people who left the study” category.
    NightScout and its influence on pushing the industry and regulating agencies is acknowledged. This is great!

    In the software table, the most excellent Dexcom Studio is still listed as the analysis toold for Dexcom data. Unfortunately, it has now been replaced by “Dexcom’s Jail” for the G5, which might one day evolve into a decent tool, but is essentially Dexcom trump’s card in the establishment of a proprietary walled garden. 

    In the Controversy regarding clinical benefits section, Rodbard perfectly articulates the essential question that should be asked from a medical point of view,

    ‘‘Did the introduction of CGM result in a change in the relationship between risk of hypoglycemia and HbA1c achieved?’’

    From a patient point of view, the question “How does it improve my every day life?” is even more important. From a purely clinical point of view, our HbA1c actually increased from 5.4 to 5.6 when we started to use a CGM but the number of activities we attempted or felt freer to attempt increase. 

    Our quality of life improved: that is what ultimately matters in a disease like T1D.

    This article is, on the whole, a very good overview of the CGM landscape, the remaining challenges and future opportunities. I strongly recommend that you read the full paper if you have the time.

    But let me insist on one point

    Please note that many of the problems Rodbard identifies would be solved by open data formats and open access to that data.

     

    Sunday, January 3, 2016

    Looking back at 2015: the good, the bad and the ugly.

    I haven't posted about our evolution and control in a while... With the start of 2016, here's a quick summary of what 2015 had in store for us.

    The good

    Our control was roughly in line with what is has always been. HbA1c tested at 5.2, 5.4, 5.6 and 5.4... Since Max's puberty is in full swing, I guess this can be seen as good news. For most of the year, our policy has been "free carbs, as long as one acts one way or the other to keeps things in decent range". A rough estimate of our carbs intake (we don't keep accurate counts) should between 200 and 400 grs per day depending on the level of activity. Our hyper/hypo percentages are OK, especially given the horrors one sees in Artificial Pancreas clinical trials. We're still on MDI and we don't see any reason to move to a pump at this point.

    The tennis season started well, with significant wins against stronger players but we decided not to play the Summer season for reasons that will become clear in the next paragraphs.


    The bad

    If I had to single out a single factor of chronic stress, I would - like many other parents, I assume - rate the awful "adolescent/teen" mindset. Previously, Max would not always know which decision to take: small mistakes could send him above 300 mg/dl or below 60 mg/dl and he would correct one way or the other. This year, on paper, he knows most of what he needs to know but he doesn't necessarily act on it, especially if he has other things on his mind.

    And , unfortunately, we saw the emergence of a new situation that complicated matters further. Around April, his insulin sensitivity seemed to increase tremendously. While a 50 to 60% decrease would previously be adequate for sports, we started constantly going low during exercise. At one point, more than 150 grs of uncovered carbs were unable to keep him in a correct range. That's of course a very annoying problem given the fact that carbs absorption is itself rate limited. We reduced insulin  even more, to the point of suppressing it totally before some training sessions. While we could maintain BG levels with zero rapid acting insulin, cramps and muscle aches show up when insulin level gets too low.

    At the same time, his Levemir/Lantus needs started to decrease: from a maximum of 22 units (split 14/8) of Levemir per day in February, we are now down to 5 or 6 units of Lantus in the evening and are still fighting occasional early night lows. We tracked a few days better than we usually do and it turns out that 1 unit of Novorapid currently covers between 30 and 40 grams of carbs. A total daily dose of 10 to 15 units in a 65 kg teen is low, very low. Add a bit of exercise into the mix and we have seen less than 10U TDD. I'll get back to the possible causes later.

    The ugly

    That extreme insulin sensitivity is, in practice, extremely difficult to deal with. If we make a single 1 unit dosing error, we have to correct with 30 to 40 grams of carbs. Don't time perfectly, take a bath, move a bit too much during the fast insulin activity period... and you are headed for trouble. We had a couple of severe hypos, very unusual for us and, as you all know, scary. In both cases Max was able to correct orally: he was conscious, able to swallow without risks and well aware of his situation... but he was unable to move or act by himself. It definitely seems that his left motor cortex (controlling the right muscles) is the first to suffer in case of hypoglycemia.

    Add a pinch of adolescent mind to the mix and you are headed for trouble...

    The day started normally enough, with a night a bit higher than usual (given the insulin sensitivity issue, I am less inclined to correct aggressively), a reduced dose of Novorapid in prevision of the morning PE class that resulted in a couple of hours above 200 mg/dl. So far, not ideal, but nothing dramatic.

    Very quickly, at the start of the PE class, a steep drop. At that point, a reasonable T1D kid would have told the teacher "I am too low" and stopped until BG was in an acceptable range. Max chose to keep running: soccer with friends must be more important than a stupid hypoglycemia, right? I suspect that, for the next couple of hours, he ran on his glycogen reserves and exhausted them. He managed to bring himself up again. It could have been smooth sailing from there if he had remembered to reduce his normal meal insulin dose. But he did not.

    At 3:30 PM, Max left school with a couple of friends and headed home. At that point, he was already falling extremely quickly: it took him 35 minutes to go from 193 mg/dl to 40 mg/dl, a sustained fall of about 5 mg/dl.min. Without taking the CGM delay into account and the slope of the fall (hard to blame him), he decided to correct around 90 with 10 grams of carbs (he ended up needing more than 100 grams to stabilize). But at this point, he was probably already hitting the 40ies, with no slow down in sight. At that exact moment, rapid insulin activity is peaking and, 3 to 4 hours after the exercise that depleted reserves, his muscles and possibly liver are stealing all the carbs they can grab...

    What had to happen happened. He froze and his right side became spastic. His friends hailed the police (that's the "good" side of the terror era, finding a policeman is easy). Informed by his friends that Max was diabetic and needed sugar, one of the policemen sacrificed his coke. A teacher who had noticed the commotion called an ambulance and I received the phone call every T1D parent dreads... "your son his in an ambulance, any hospital preference?".

    Looking back at the incident, it was not that bad. No glucose perfusion, no glucagon. Max was out of the hospital after two hours, a coke, some dextrose tablets and some chocolate. But it was definitely a reminder that we are constantly living on the edge. What would have happened if he had been alone? That question will remain, for a long time, on my mind.

     

    Going forward

    At this point, I am not going to speculate to much on the root cause of our insulin hyper sensitivity. The rough outline of our thought process has been so far

    partial recovery/delayed honeymoon? While I'd love that option, it is excluded by his complete lack of C-Peptide.

    digestive trouble leading to sub optimal absorpion? That is not out of the question since he is celiac, but he does grow and gain weight normally...

    other endocrine issue such as Addison disease? His cortisol and ACTH levels have been tested and are in acceptable range, if lower (for cortisol) than it was before. No specific anti-bodies have been found. Dynamic tests are scheduled tomorrow.

    2016?

    Let's be honest: I am not looking forward to 2016. Not looking forward to another year of constant worries and tedious routine. Not looking forward to any new information that would explain our issue and further my understanding of his problem. I have no need for the "Have a wonderful 2016" style of wishes. 2016 will not be good, no matter how we slice it.  I just hope it will not suck big time.

    Therefore, here are my 2016 wishes for the T1Ds and caregivers reading my blog.

    I simply wish that your 2016 sucks a bit less than your 2015 did...