My son, David, was diagnosed with type 1 diabetes in March 2013. In full-blown DKA, we spent several days in the hospital recovering and learning as much as we could about diabetes—something about which my wife and I knew absolutely nothing. After this very brief introduction, our pediatric endocrinologist told us that while for most patients he recommended a blood sugar between 100 and 180, he thought we might be able to achieve a target of between 100 and 150…hope!
We returned home with our emaciated (he had lost 12 of his 65 pounds to DKA) son, prescriptions for insulin, a glucose meter, and a standard American Diabetes Association carbohydrate counting meal plan. I took time off from work those first few weeks and tried to get a handle on this new diagnosis and care for my son. A typical day: we would test, bolus, eat the recommended meal from the meal plan, retest an hour later, and so on. Here is what we saw on the meter:
This result from the first month threw me into a bit of a panic—I had been reading about the catastrophic consequences of high blood sugars and was terrified. I had also been reading about potential cures and did not see anything promising on the near-term horizon. Trying to figure out our next move, two things stuck with me. First, when our doctor discussed with us the typical blood sugar profile (the actual chart from the hospital talk is below!), because of my work as a scientist, I remember thinking, “Aha! The blood sugar concentration is a time series signal—maybe it can be manipulated to stay in range.” At that same exact time, while in the hospital bed, my son asked the doctor, “Then why eat carbs?” in response to the explanation that carbs raise blood sugar and high blood sugar is the source of complications (the answer given: “for energy”).
Thirty days following Dave’s diagnosis, we purchased Dr. Bernstein’s book, Diabetes Solution. Bernstein had the answer to our problem of out of control sugars: minimize carbohydrate intake (replacing with fat and protein) and normalize blood sugar in order to avoid complications. My wife and I wondered, “But don’t folks need carbs for energy?” (We also thought my son needed insulin as a growth hormone and that carbs necessarily should be used to cover insulin! Not so!) But as Bernstein explains, “energy” can be had through protein glycogenesis, the liver makes all the glucose the body needs. Moreover, Dr. Bernstein explained, “There are essential fatty acids, essential amino acids, but no essential carbohydrates.” What Dr. Bernstein wrote made sense—why eat grains or sugars when they raise blood sugar? What health benefit would outweigh the negatives? Our family made a unified decision to try and follow a low carb (LC) diet. All carbs were removed from the house. My wife began scouring the internet for low carb recipes. Our favorites were from Maria Emmerich. As it turns out, cookies, pizza, ice cream—even Cheetos(!) can be made low carb. Moreover, these new foods we ate seemed very healthy! Nothing processed, etc. The food “problem” actually became a positive family bonding experience. Our family had a lot of fun cooking pre-diagnosis, and so this new way of eating was another culinary challenge! It was stunning to see that we could indulge in all of our favorite foods, but without the blood sugar rise—for the non-diabetics in the house as well.
At this time, Dave was also getting outfitted for an insulin pump and CGM—the truly great Dexcom G4 system. We now had all of the supportive technological elements in place to start focusing on assisting Dave in controlling his blood sugar. The initial reports from the Dexcom CGM were positive. The lower carb methods allowed us predictable control. By the six-month mark, Dave started to have A1cs in the mid-5s and our CGM traces looked like the following:
During this time, I talked to a gentleman at my gym that I see frequently in the mornings. He was overweight and struggling to resume an exercise plan. After we became friends, the topic of his A1c came up. His was 5.9. Seemed okay, and yet, he had cardiovascular complications. I did some research and read a paper that showed the correlation between coronary bypass surgery patients and A1c. Not one patient in the study had an A1c below 5 and very few in the low 5s, but as the A1c increased the probability of surgery increased dramatically. Moreover, I learned that the death rate of cardiovascular disease is 4 times higher for diabetics than non-diabetics, for whom it is already the leading cause of death. Going further, even with Dave’s lower A1c, he was frequently (daily), undergoing excursions above 140, due to “empty” carbs…just a few chips here and there, and the blood sugar would skyrocket. On her blog, Jenny Ruhl documents the damage these excursions above 140 mg/dL do to our internal organs. And of course, there was Dr. Bernstein, advocating blood sugars at 83 mg/dL before, during, and after meals which corresponds to an A1c between 4.2 and 4.6. More reading—we learned that the entire “lipid hypothesis” correlating total cholesterol to heart disease was based on faulty epistemological studies performed by Ancel Keys and instead it was glycation and inflammation that are the source of chronic diseases. We were convinced that we needed to further help Dave to normalize his blood sugars. What to do?
Now, also at this time, I was working hard to understand the current state of diabetes research. As a chief scientist at a major DoD contractor, might there be a role for me to play? I learned of the work Professor Damiano on the artificial pancreas (AP), and naturally became very excited. Sensor signal processing is precisely in my wheelhouse and I have been implementing such real time sensor systems all the way back to the 1990s. My son and I read stories about Damiano and his AP trials—folks being turned loose in the neighborhoods of Boston free to eat what they want, sleep, and still wake up in range! But the excursions in blood sugar were still there. I wondered, “Shouldn’t the artificial pancreas be able to combat this issue?” Nevertheless, we were excited. With the LC diet in place, the biggest obstacle in place for Dave’s diabetes management is the cognitive load and worry that diabetes puts on Dave and our family. The AP attacks this very problem! And clinical trials are underway!
Knowing that the chance of working on one of these research teams was small, I began to write my own code to simulate the effects of food glycation and insulin action as well as an “expert system” based in fuzzy logic to control the simulated pancreas. The technical results of these simulations and N=1 instantiation will be published on the arXiv (a service owned and operated by Cornell University) soon, but the conclusions are now firm. In particular, our input was different than the AP research, where subjects ate pancreas-stressing meals, e.g. pizza, to really push the system. We wondered—how would the AP perform in response to a LC way of eating? We decided to find out. We built our “fuzzy logic” rule set we developed in simulation and first tuned it until we could get a nearly flat blood sugar:
We then applied it, albeit it manually, for 90 days. My son would implement the expert system at school, and while at home, I would watch the CGM and operate the “automation.” Now, briefly, what is fuzzy logic? It is nothing more than using operator-based rules, rather than say, mathematical functions, to control a dynamic system. Diabetics do fuzzy logic every day! For example, if Dave tests at 120 mg/dL, he might take a full unit rather than a half unit of insulin correction if his blood sugar is rising rather than stable. In addition to having to manually operate the fuzzy logic control system for insulin dosage, we also wanted to mimic Professor Damiano’s bihormonal system using glucagon, but of course we don’t have a glucagon pump, so we used glucose tablets and berries instead. A few examples of some of the system rules we used are given below. These rules are anchored in Dr. Bernstein’s concept of The Law of Small Numbers.
- Meals. Low carb meals are chosen from a menu of about two dozen different choices, which all require Dave about one unit of insulin to cover.
- Insulin Corrections. Corrections are given at thresholds. For example, 0.5 units given when blood sugar rises above 95 mg/dL, another 0.5 given when blood sugar rises above 105 mg/dL. No more than 2 units of insulin on board at one time, or the action is too fast for the CGM.
- Glucose Corrections. Glucose is given depending on insulin on board, slope and value. For example, a rule would be not to correct an 83 mg/dL which is “flat,” while a dropping 90 mg/dL with .8 units on board would receive a dedicated amount of insulin.
These rules are implemented in a “fuzzy” way. There is some latitude and judgment used which is inherent in the so-called fuzzy “membership functions.”
We implemented these rules on a homemade display (developed in part by John Costik et al who can be found on the Facebook group CGM in the Cloud) which takes output from the Dexcom system and alerts to given appropriate action rule sets, see below. These rules, combined with the meal plan, as well as techniques to combat growth spurts at night (checking frequently and understanding patterns is key) kept Dave in range over a period of ninety days and he obtained normal blood sugars like those seen below. The A1c obtained during this period of ninety days was 4.4, which is right in the middle of what is considered to be the normal, healthy non-diabetic range of 4.2 to 4.6. You can imagine our enthusiasm regarding the artificial pancreas after Dave achieved this result! With an AP, Dave’s burden of diabetes would then come down to brief infusion set changes/checks as well as glances at the CGM. Nightly wake-ups/checks would be reduced as well!
A main issue remains, however: the perceived restricted nature of the LC diet. For us, this is now not a big deal, as the whole family is enjoying the health benefits of LC eating. However, the LC way of eating is clearly not appealing to most folks.
So, the million dollar question is, “What are the benefits of the AP for the diabetic eating the modern high carb American diet?” The fundamental issue can be understood by examining the key time scales which can be used to characterize the performance of an automated diabetes management problem:
1. Τ CGM, which is the delay time from the true blood glucose level to that measured in the interstitial fluid by the CGM—about 15 to 25 minutes.
2. T INS, which is the delay time between injecting insulin and witnessing the blood sugar change. This is about 50 to 60 minutes.
3. T FOOD, which is the time between ingesting the food and seeing the blood sugar rise. Depending on the food, this can be a matter of minutes to a matter of hours.
The chart above identifies how these time scales work together in order to determine the capability of the artificial pancreas to obtain truly normalized blood sugars—a constant 83 mg/dL. When fast acting rapid carbohydrate—fruit, flours, grains, sugars—are eaten, the ratios are much greater than one. When these large ratios exist, the AP cannot keep up. Consequently, blood sugar rises. However, according to our simulations and in our N=1 experiment, when these ratios are both kept at around 1 (or lower), there is minimal blood sugar variation. This requires a diet consisting of small amounts of fibrous carbs—say salads, nuts, etc., as well as protein (slow glycation), and healthy fats (for example, read The Big Fat Surprise by Nina Teicholz). In this case, normalized blood sugars can be obtained. Therefore, my view of the AP, given our experience with a controlled diet and the manual implementation of the AP, is one of extreme enthusiasm. There is a known “way out” of diabetes complications—a modified diet, which minimizes fast carbohydrates. With this way of eating, the cognitive load of diabetes management is minimized, but still remains. The AP will greatly minimize the remaining minute-to-minute attention needed for diabetes management and provide very good tools to help diabetics who are careful with their diet, to eliminate the worry of complications of high blood sugar. For non-LC eaters, the results are mixed. A1cs will be lowered by a point or so—this is an important step. Still, however, the impact of high carbohydrate foods on blood sugar will remain, not to mention the extreme episodes of hypoglycemia experienced when large doses of insulin are used to cover meals.
In conclusion, we believe that a (safely functioning!) AP system could be an ideal tool for normalizing blood sugars given a LC diet and removing cognitive load. For the “modern American diet” or high processed/refined carb diet, our research directly points to the required advances for an AP:
- The need for faster acting insulin to combat glycemic load from these meals.
- A highly accurate CGM that does not, for example, suffer from pressure lows (during sleep for example when the wearer might sleep on his sensor and erroneously force a bihormonal injection). This is a requirement in all cases, but is particularly acute when wide glycemic variations are in play.
In the case of the high carb diet, given the above advances, and assurances on safety, an AP would not be forced to run at high target blood sugars during post meal periods. As for my son, the 90-day exercise was proof enough for us to stay the course. We have found that counter to our initial thoughts, a low carb, high fat diet is a rather luxurious, satisfying way to eat. The cognitive load of diabetes management is dramatically lessened. My son’s growth and athletic performance has been astounding. And most importantly, there is no longer the feeling that type one diabetes is a hopeless state where complications will inevitably creep in. Indeed, we feel like we are winning.