International Society for Pediatric and Adolescent Diabetes (ISPAD)

October 30-November 2, 2019; Boston, MA; Day #2

Executive Highlights

  • We are very happy to be in Boston’s Back Bay, host of the 45th Annual ISPAD Congress! The learnings are also very strong in day #2 – check out our top highlights below and check our preview for a look at days #3 and 4.

  • Dr. Mark Clements’ presentation on a machine learning model, utilized at Children’s Mercy Kansas City to identify higher-risk patients, left the audience even more excited about tech-enabled insights to improve clinical care. Notably, the model was able to forecast future A1c levels, using the A1c at last visit and EMR data. With patients at-risk for significant A1c rises identified, clinicians at Children’s Mercy can take preventive actions.  It was fascinating to see that the number of fingersticks/day and insulin boluses/day each had a significant weight and could be targets for interventions. Dr. Clements did acknowledge that statistical models are time-consuming and difficult to run each now (right now!).

  • The promise of CGM and automated insulin delivery (AID) was a huge excitement generator for ISPAD attendees. Tandem presented very strong Basal-IQ real-world data, showing a remarkable 0.9% of time <70 mg/dl – just 13 minutes/day! Dr. Bruce Buckingham shared a few new details about the hotel study of Omnipod Horizon (hybrid closed loop) in young adults (ages 2-6), while Dr. Roy Beck pushed glycemia metrics forward beyond A1c, asking his audience, “Would you rather have [an ambulatory glucose profile] or would you rather have one number that reflects someone’s average over the last three months, and may not even be an accurate average?”

  • At a Lilly symposium on Baqsimi, Jeff Hitchcock, founder of Children with Diabetes and Friends for Life, shared powerful, personal stories about his experiences with severe hypoglycemia. The stories are well worth reading and emphasize how important it is for people with diabetes and their care partners to be prepared to treat severe hypoglycemia.

Greetings from the City of Champions, home of the Red Sox, Celtics, Bruins, Patriots, and of course, ISPAD 2019! Day #2 was incredibly busy and full of excitement – see our top highlights below!

ISPAD Day #1 Highlights - Lessons and strong results from telehealth-driven CoYoT1 care model; Kelly Close looks back and ahead; KOLs galore: Ms. Laurel Messer, Prof. Chantal Mathieu, and Dr. Simeon Taylor

Top Seven Highlights

1. Machine Learning Model Can Predict Future A1c: For 3-Month A1c Rise of 0.6%, Positive Predictive Value of ~50% Using EMR Data Alone, ~60% With EMR Data + Diabetes Device Data

During a rapid-fire afternoon session on ways to improve clinical care delivery, Dr. Mark Clements (Glooko / Children’s Mercy Kansas City) presented promising data on a machine learning-based model for predicting A1c at the next visit, based on the last A1c value, discrete EMR data, and available diabetes device data. The model, which is a random forest (i.e., a “forest” of many decision trees), was fitted and tested using data from 1,743 youth (≥9 years old) who recorded 9,643 visits. Discrete EMR data (vs. free-form text fields), patient reported outcomes (PROs) from an electronic intake form, and diabetes device data were used to feed the model, along with the last recorded A1c value. The machine-learning (ML) model outperformed both a null prediction model and a linear prediction model (i.e., future A1c rise will be the same as the most recent A1c rise) at predicting A1c rise at the next clinic visit. Using EMR and PRO data alone to predict patients whose A1c would rise 0.6%, the model’s PPV was ~50%, meaning about half of the cases that the model identified actually saw an A1c rise of ≥0.6%. In comparison, the PPV for an A1c rise of 0.6% was just 30% using a linear model, i.e., 30% of the cases identified by the linear model actually had an A1c rise of ≥0.6%. For an A1c rise of 0.3% over 3-months, the PPV was 56%, with a sensitivity of 21% (i.e., the model was able to identify one-fifth of all participants who actually saw a 0.3% A1c increase). When 2-weeks of diabetes device data were added into the model, PPV was improved to ~60% for an A1c rise of 0.6%. Unsurprisingly, the model performed better with device data that was more recent. Using the model, a clinic is able to identify at-risk youth to clinicians so that preventive action can be taken. Children’s Mercy is using the information to perform video “micro-visits” every 2-3 weeks and encourage data sharing and peer mentoring for identified patients.

Table 1. ML-model performance using EMR-data to predict 0.3% A1c rise over 90 days

Parameter

Estimate (%)

95% CI

Sensitivity (what percentage of all true positives was the model able to flagged)

21%

18%, 25%

Specificity (what percentage of all true negatives was correctly unflagged)

86%

83%, 89%

Positive predictive value (what percentage of flagged cases was true positive)

56%

49%, 62%

Negative predictive value (what percentage of unflagged cases was true negative)

57%

54%, 60%



  • Using data from diabetes registry data, Dr. Clements identified five typical tracks for population mean A1c from ages 8 to 18. For ~20% of the registry population, A1c rose greatly from ages 8 to 18, while mean A1c stayed flat for the remaining ~80% of the groups. Referencing the infamous T1D Exchange graph of population mean A1c over time, Mr. Clements noted that the A1c peak for teenagers and young adults was driven primarily by this ~20% of the population. Thus, the ability to identify these higher-risk individuals could be the key to flattening the A1c rise seen in adolescents and young adults.

Selected Question and Answer

Q: What are some of the modifiable risk factors identified by the model?

A: Well, there are some risk factors that aren’t modifiable, like ZIP code, but there were modifiable risk factors. From the device data, we were able to parse out the number of fingersticks/day, insulin boluses/day - each of these had a significant weight and could be the targets for interventions.

2. Dr. Roy Beck: “The Only Value of HbA1c if CGM is Available is for Historical Purposes”

A leading voice in the beyond A1c movement, Dr. Roy Beck (Jaeb Center), made a strong case for time-in-range (TIR) over A1c, summarizing his argument by showing an ambulatory glucose profile and asking, “Would you rather have this or would you rather have one number that reflects someone’s average over the last three months, and may not even be an accurate average?” Dr. Beck’s talk touched on many facets of the beyond A1c movement, including consensus targets for CGM metrics, measures of glycemic variability, and the relationship between TIR and complication rates.

  • Dr. Beck discussed the international consensus on CGM metric targets, which was presented as a late-breaking poster at ADA 2019. Interestingly, Dr. Beck noted that the consensus group chose to use time above 180 mg/dl and 250 mg/dl as the measures of hyperglycemia, rather than area under the curve or high blood glucose index, which might be more intuitive measures given that they are weighted by severity of hyperglycemia. Ultimately, Dr. Beck stated that it was not important, as all these metrics correlate extremely highly with each other (0.95 to 0.99).

  • On measures of glycemic variability, Dr. Beck recommended using coefficient of variation (CV) over standard deviation (SD) and mean amplitude of glucose excursions (MAGE). Because SD and MAGE are positively correlated with mean glucose (r=0.82 and 0.76, respectively), patients and providers can be misled into thinking their variability is being reduced when only their mean glucose is changing. In contrast, CV is not correlated with mean glucose. Perhaps unintuitively, CV is also not correlated with hyperglycemia, though it was significantly correlated with hypoglycemia (r=0.68). Thus, by reducing time spent in hypoglycemia, patients are likely to improve their glycemic variability.

  • Referencing his 2018 paper in Diabetes Care, Dr. Beck showed graphs demonstrating the relationship between decreased TIR and increased rates of complications using data from DCCT. TIR was calculated using the quarterly seven fingersticks/day data that was collected in DCCT. For every 10% decrease in TIR, hazard ratios for retinopathy and microalbuminuria increased by 64% and 40%, respectively. With actual CGM data (up to 288 measurements/day), the association between TIR and complication rates could be even stronger.



Selected Question and Answer

Q: If you had to design a trial, would you go for the seven fingersticks/day (from the DCCT trial), A1c, or CGM? How much time would you need for the CGM to be used?

A: CGM is easier and now, you get more than seven points a day – I’d rather have 288 points/day. We’ve shown you need about 10-14 days during a pretty stable time to get a good estimate of glucose metrics, so in studies, we try to get two weeks of data, either with real-time CGM or blinded, and then we gather data in 10-14 day intervals … In some closed loop studies, like this week’s NEJM paper about Control-IQ that the NIH-funded, they use CGM metrics as the primary outcomes. The FDA is not willing to accept that yet as part of an effectiveness claim, so there’s a lot of work going on to get the FDA to come along. One challenge with CGM, compared to A1c, is that you have to actually wear it. It becomes a challenge if it’s going to be a metric to make sure you get a high level of adherence.

3. Real-World Pediatric Data from Basal-IQ (n=2,696): 0.9% of Time <70 mg/dl, Mean ~5 Insulin Delivery Suspensions/Day

A poster demonstrated the effectiveness of Tandem’s low glucose suspend algorithm, Basal-IQ, in 2,696 real-world pediatric (ages 6-17) t:slim X2 users. The data, uploaded to t:connect between August 2018 and March 2019, showed 0.9% of time spent under 70 mg/dl, or just 13 minutes per day! On average, the Basal-IQ algorithm suspended insulin delivery 4.9 times/day for an average of 15.5 minutes per suspension. On average, users’ glucose values were at 111 mg/dl when suspension was initiated and 100 mg/dl when insulin delivery resumed. The 0.9% of time spent in hypoglycemia was even lower than the 1.2% reported in Tandem’s first round of real-world data from Basal-IQ shared at ATTD 2019.

  • Before and after Basal-IQ initiation data was available from 491 users. That subgroup showed a reduction in time <70 mg/dl from 1.6% to 1.1% (~7 minutes/day; 31% relative reduction, p<0.001). The ability to significantly reduce hypoglycemia from an already-low baseline is certainly a major testament to Basal-IQ’s effectiveness. In just 5% of insulin suspensions, the suspension was manually overridden by users. Time-in-range before and after Basal-IQ remained approximately the same, at 53%-54%, though the percentage of time >300 mg/dl decreased slightly by ~9 minutes/day (p=0.007). These nine minutes total an hour per week – we’re sure patients will take this!

  • This real-world pre-post relative reduction in hypoglycemia exactly matches Basal-IQ’s pivotal trial result, presented at ATTD 2018. Both data sets show a 31% relative reduction in time <70 mg/dl, from a baseline of 4.5% in the pivotal and 1.6% in the real-world subgroup. As a reminder, Basal-IQ looks ahead 30 minutes and suspends insulin when glucose is predicted to drop below 80 mg/dl or if glucose is currently below 70 mg/dl and falling. The system resumes insulin once glucose values start to rise.

4. Children with Diabetes’ Jeff Hitchcock Shares Powerful, Personal Stories on Severe Hypoglycemia

At a Lilly-sponsored symposium on Baqsimi (nasal glucagon), Mr. Jeff Hitchcock (Children with Diabetes) shared multiple personal stories about his experiences with severe hypoglycemia. The stories were powerful and moving, and all of them emphasized how important it is for people with diabetes and their caretakers to be prepared to treat severe hypoglycemia. Baqsimi, launched just a few months ago, has already seen strong uptake in the US, holding one-third of all new brand prescriptions. Given the challenges with using traditional glucagon, we’re surprised this number is not even higher although we know drug adoption for anything new can be slow. We hope to see more investment in marketing.

  • “It was 1989, my daughter was 2 years old and she would beg us for water – we didn’t realize at that time that those were classic symptoms of type 1 diabetes. We had taken her to the pediatrician, where she was diagnosed with oral thrush, later diagnosed with type 1 diabetes, and taken to Washington DC Children’s Hospital. We met with a pediatric endocrinologist, and he basically said ‘don’t worry, everything’s going to be okay.’ In the end it turns out that he was correct – this year, my daughter celebrates her 33rd birthday living with type 1 diabetes. But, as we all know, living with type 1 diabetes on a daily basis has some challenges. One of those challenges is matching food to insulin for a toddler and young child. This is a picture of my daughter at the age of six, recovering from the lowest blood sugar she ever experienced, it was 17 mg/dl. At the time, the best instrument for measuring blood glucose took 120 seconds. We began feeding her cake frosting frantically until we got enough in her for her blood sugar to rise. At the time, glucose tablets were orange and chalky, and she would only take cake frosting – a little bit of advice, don’t use red because when they throw up, it stains everything. Had we had [glucagon] with us, we would have definitely used it. She was fortunate enough to have never needed to use a glucagon rescue, she was always able to rescue herself with oral carbohydrates, but we have been fortunate.

  • “I just dropped my daughter and wife off at the airport and I got this on my phone: ‘Dad, don’t panic. My blood sugar is 33 [mg/dl]. If you don’t hear from me in five minutes, please call.’ Nothing good starts with ‘Dad, don’t panic.’ I’m sitting in my car, outside, on the free-way near the airport and I thought, what is her address? How do I call emergency services in Tampa? Can they get to her before she loses consciousness? I called her back, told her to sit down, put her phone on the ground, and put it on speaker – ‘we’re just going to talk.’ Her sensor had caught the lowering blood sugar, and she had treated it, but it was still going down and she was scared. A friend of mine said I should be very proud that at 23 she felt confident enough to call me, and I am, but it was terrifying. She had called her boyfriend, now husband, but phone service did not work in the hospital. We were talking, she was starting to feel better, and I’m also thinking, well, if she had a glucagon kit and had tried to flag someone down, would they have known how to use it? Probably not. It took about 40 minutes for her blood sugar to finally get up to 70 [mg/dL], and she said she felt good enough to go to work. It took me a little bit longer before I could drive home.

  • “[It] was breakfast on Thursday night, this young man with type 1 diabetes had bolused for his breakfast. A friend came by and asked him for help to carry some fun stuff in his room back. He got up, started walking across this resort, he’s a thousand feet away from this giant swimming pool and his blood sugar crashes. He ends up falling to the ground. The people around him that worked for a diabetes care company had no idea what to do. By chance, a mom attending the conference had a glucagon emergency kit in the back of her child’s stroller. Also, a CDE with type 1 diabetes was running by on his morning jog. He stopped, got the glucagon kit, mixed it up, injected it, and rescued him. That’s a lot of things that had to happen all at once to save that young man.”

  • “Later that afternoon, there was a young teenager who had gotten the stomach virus. We watched that teenager’s blood sugar slowly decline. They couldn’t eat anything. We had time to take a glucagon kit, mix it up, inject glucagon, and prevent severe hypoglycemia. The teenager was taken by their parents to a local hospital for an IV and emergency room visit for the stomach virus.”

5. Hotel Study of Omnipod Horizon Algorithm in Children Ages 2-6 (n=14): One Severe Hypoglycemia Event (Unrelated to System), Time-in-Range Improved From 55% to 73% (+4 Hours/Day)

In yet another demonstration of his brilliance and poise, Dr. Bruce Buckingham (Stanford University) powered through a presentation of positive results (+4 hours/day time-in-range) from Omnipod Horizon in young children, even as the presenter’s laptop malfunctioned making his slides disappear (“Photography is permitted at this presentation,” he quipped.). While Dr. Buckingham first read out the study (n=14) at ADA 2019, he shared information about an adverse event, which was deemed to be unrelated to the system itself, for the first time. The study evaluated the safety and efficacy of Omnipod’s Horizon automated insulin delivery (AID) algorithm in young children, ages 2-6. The study involved seven days of standard therapy baseline data collection followed by 48-72 hours in a supervised hotel setting (a fun-looking “hotel” we’d note – see below). Without the algorithm integrated on the pump, participants had to carry a tablet with MATLAB that held the algorithm and handled communication between the CGM (Dexcom G4 or G5) and Omnipod. The participants, mean age 4.8, had fairly tight glycemic control at baseline with an A1c of 7.4% and mean glucose “around 174 mg/dl.” Notably, the study included both MDI and CSII participants, with MDI participants going straight to the closed loop system. The kids were allowed to “free-range” their meals and averaged exercise time of ~1 hour/day (e.g., trampoline, laser tag, etc.).

  • During the study period, time-in-range (70-180 mg/dl) was 73%, compared to 56% in standard therapy (p=0.0002) – a remarkable four additional hours in range! Both time in hyperglycemia and hypoglycemia were reduced, though the time in hypoglycemia difference was not statistically due to small sample size. Participants spent 3.6 fewer hours/day above 180 mg/dl (40% vs. 25%; p=0.005) and even 2.6 fewer hours/day above 250 mg/dl (17% vs. 6%; p=0.002). During overnight hours (11 PM – 7 AM), time-in-range was 85%, compared to 58% during baseline, a massive difference. Lastly, coefficient of variation during the closed-loop period was 36%, compared to 40% under standard therapy (p=0.02).

Parameter

With Horizon (2-3 days)

Standard Therapy (7 days)

p-value

Mean glucose

148 mg/dl

172 mg/dl

0.017

Time-in-range (70-180 mg/dl)

73%

55%

0.0002

Time >180 mg/dl

25%

40%

0.005

Time >250 mg/dl

6%

17%

0.002

Time <70 mg/dl

3%

5%

0.24

Time <54 mg/dl

0.4%

1.8%

0.15


  • Omnipod Horizon is scheduled for a pivotal trial starting 4Q19, with US launch coming as soon as 2H20. Insulet is committed to “breakthrough ease of use,” smartphone control, and launching with a pediatric indication. Moving forward, the young children’s study will be done in a fully free-range setting, which will likely require the algorithm to become integrated into the Omnipod itself.

6. Could Early Treatment with ACE Inhibitors and Statins Prevent Vascular Complications in Youth with Type 1? Updates from the AdDIT Study

Dr. Loredana Marcovecchio presented findings from the Adolescent Type 1 Diabetes Cardiorenal (AdDIT) study, which sought to determine if early initiation of ACE inhibitors or statins could prevent the onset of kidney, eye, and cardiovascular complications. This update focused on whether treatment with ACE inhibitors and statins prevented CVD onset via urinary albumin excretion as a link between renal health and CVD. Higher albumin-to-creatinine (ACR) ratio, even within normal limits, was associated with progression to worse CV profiles. Thus, Dr. Marcovecchio believes that early screening for ACR may be valuable to identify teens at high risk for vascular complications. The next phase of AdDIT sought to do just so and focused on a group of patients with high ACRs. Treatment with statins reduced lipid levels, but didn’t impact more direct measures of CVD, while treatment with ACE inhibitors reduced progression to microalbuminuria. Both therapies were well tolerated with high adherence rates, and longer-term studies are needed with stricter CVD endpoints to determine the usefulness of these therapies in this younger population. The AdDIT Follow Up study plans to track risk for vascular complications from adolescence to adulthood, and completion is expected in 2021. The hope is that full results from the follow up study can lead to better risk assessment protocols based on the link between the cardiovascular and renal systems, as well as identify new targets for interventions to prevent complications.

  • Even if ACE inhibitors and statins receive approval for use in pediatric populations, the challenge of clinical inertia cannot be forgotten. Dr. Michelle Katz focused on this issue in her presentation during this session, emphasizing that pharmacotherapy can be a viable treatment option for patients that do not have the full support required for lifestyle change or who do not feel confident in their ability to facilitate one. She stated that other challenges include too short of patient visits, insufficient patient education materials, and limited provider training on facilitating lifestyle interventions for patients. While lifestyle change remains the most recommended and viable option for populations as resilient and transitory as youth, we understand that more patient options in treatment options will help many. We believe in particular that options of therapy can help supplement and even encourage lifestyle changes.

7. Sanofi Symposium Focuses on Managing Type 1, the Importance of Language, and the Advent of TIR

Sanofi’s star-studded symposium provided a broad sweep of the state of type 1 diabetes, from diagnosis to management with up and coming technologies. Dr. Lori Laffel gave the opening address, emphasizing that over 40% of teens and young adults with type 1 diabetes had A1cs of 9 or higher in developed countries. High A1c can be especially problematic when it is persistent and turns into complications many years down the line. Dr. Thomas Danne led a discussion on the progression of type 1 diabetes from disease onset to adulthood and centered his discussion on the early prevention of CVD complications. He believes that digitization can lead to improved outcomes, but only along with increased education, CGM, and related technologies to more tightly manage glucose levels. However, we must not forget that language matters just as much as biology. Dr. Barbara Anderson stressed that it is important to consider cultural factors and avoid judgmental language when treating children (or anyone!) with diabetes. Especially in pediatric populations, Motivational Interviewing should be used to when speaking with patients and families following OARS (Open questions, Affirm, Reflect, Summarize) in order to help families understand the condition’s complexities and options. Though she maintains that face-to-face interactions are the gold standard for behavior change, virtual consultations and training programs with relational agents can also help education and guide patients when clinics are harder to access. Dr. Tadej Battelino wrapped up the symposium with a talk on the importance of time in range (TIR) to manage type 1 diabetes. As A1c doesn’t account for individual glucose variation, using CGM to find TIR is another option for gathering reliable data to advise diabetes care and improve quality of life. With this, the accumulation of data that CGM allows can significantly impact disease burden and cost savings when analyzed in exploratory studies to create novel, effective interventions.

 

--by Albert Cai, Ursula Biba, and Kelly Close