More commercial AID systems. What do the trials tell us?

As many will be aware, at ADA Scientific Sessions this week, Medtronic announced the the outcomes of a number of trials of the 780G, including two that compared the 780G with either the 640G or the 670G, alongside the European approval for the device, while Omnipod also announced the outcomes of one of their pre-pivotal trials on the Omnipod 5 (née “Horizon”) system.

I’ve previously looked at the 670G and Control-IQ side-by-side, so felt that it’s worth taking the data that’s been released this week and seeing how it looks in comparison to those trials. Be warned though. What this highlights is that trial design and participant choice makes a big difference to what comes out at the end and makes it really difficult to to make comparisons across different companies’ datasets, as Medtronic have managed to prove yet again.

First up, what’s the news from this week?

Medtronic 780G

For those who didn’t know, the 780G is the next in line in Medtronic’s Automated Insulin Delivery platform, boasting bluetooth connectivity (at last) and the ability for the data from the pump to be read on your phone, and also for that data to be shared with up to five others, That’s right. Medtronic finally catching up with #WeAreNotWaiting!

What’s perhaps more important is the algorithm at the heart of the system, which is adapted from DreaMed’s Glucositter, a more than five year old CE marked AID algorithm, that Medtronic obtained exclusive rights to use back in 2015. One of the things that’s very notable about how Medtronic describe the 780G is using the term “AHCL” or “Advanced Hybrid Closed Loop”, to give the impression this is perhaps more advanced than some other offerings on the market.

Medtronic undertook three trials:

1. A 90-day at home U.S. pivotal trial, studying adults and adolescents aged 14-75 years old;
2. The Fuzzy Logic Automated Insulin Regulation (FLAIR) trial which directly compared the 670G and the 780G in a randomized crossover design, with a study population ranging from people using multiple daily injections and not using continuous glucose monitoring to people already using the MiniMed 670G, and with a wide range of starting Hba1C; 
3.  A randomized cross-over clinical trial comparing the 780G with a Predictive Low Glucose Suspend system, based in New Zealand studied a more challenging patient group, including those with less-controlled diabetes and a younger patient population of children as young as seven years old.

We’re going to look at the first two of these here.

The characteristics of the first trial were a single arm, 16-site trial that enrolled 157 participants: 118 adults (ages 22-75 years) and 39 adolescents (ages 14-21 years). Inclusion criteria were that all participants were previously on pump therapy, with or without a CGM for at least 6 months and a mean baseline Hba1C of 7.5%. In this case, a very traditional clinical trial population.

It ran with two different target levels and variable insulin action times. The two targets were 120mg/dl and 100 mg/dl

The published results are:

• No severe hypoglycemia and diabetic ketoacidosis
• Average A1C of 7.0%
• Overall Time in Range (defined as 70-180 mg/dL) of 75%, with overall Time Below Range (defined as less than 70 mg/dL) of 1.8%
• Overnight Time in Range of 82%, with overnight Time Below Range of 1.5%
• Autocorrection contributing to 22% of all bolus insulin
• Participants being in SmartGuard™ (closed loop) 95% of the time
• Mean sensor glucose (SG) of 148 mg/dL overall, and 144 mg/dL at the default 100mg/dl target

The pivotal trial also presented the following Time in Range data in relation to adjusting the insulin action time when using the 100mg/dl target:

What’s perhaps unclear here is how Medtronic define Active Insulin Time. As many readers of this website will be aware, the half-life of just about all the available insulins right now is around 55 minutes. For insulin to only be active for 2-3 hours therefore suggests that their algorithm takes a very different view as to what Active time is compared to most of us in both the DIY and real world. It may raise some interesting challenges for those losing access to their pump in the case of a malfunction and being required to go back to traditional methods of delivering insulin. 

The second trial is more interesting. The Fuzzy Logic Automated Insulin Regulation (FLAIR) trial compared the 670G to the 780G using a randomised crossover trial design (potentially linked to the issues raised in regard to the 670G’s pivotal trial). The study contained a population that was considered more challenging because of the following aspects:

• The population had a range of users from people using multiple daily injections (20%) and not using continuous glucose monitoring (CGM; 38%) to people already using the MiniMed 670G (13%);
• The average glycated hemoglobin (HbA1c) was 7.9% (63 mmol/mol), but there was a wide range, with 27% of people having levels below 7.5% (58 mmol/mol) and 14% having levels of 8.6–11% (70–97 mmol/mol);
• The age range of the population was 14-29

The trial results can be viewed in many ways, but the key point for me was the increase in time in range from baseline, with particular focus on those who were using MDI to start, who went from 45% on MDI to 65% on the 780G (AHCL). Overall results are shown in the below image):

Perhaps the most important point to come out of FLAIR was that those that had previously never used diabetes technology had equally good results to those that had previous experience.  

Comparing to the 670G pivotal trial

As a reminder, in the initial pivotal trial of the 670G saw a TIR of 72% and an increase from baseline of 67%, with 87% of participants continuing to use the system during the trial, within a starting population with Hba1C of 7.4%.

Comparing this to the pivotal trial for the 780G:

So what we can see is that in the standard pivotal trial set up, the change between the 670G trial and the overall value on the 780G trial is very similar to the change between the 670G and the 780G in the FLAIR study. 

What is noticeable is that there is no reporting in the media of continuation of use, compared to the 670G trial. I’d hope that the Guardian 3 CGM system in use here has overcome some of the major issues that caused people to drop out previously. 

Overall then, the data suggests that with the 780G, you’d probably wnt to start with the 100mg/dl target and a 2 hour AIT for best outcomes. 

Omnipod 5 System

Formerly known as the Horizon system, Omnipod’s series 5 introduces their own AID system. 

It’s worth noting that to run Horizon, users had the choice of a locked Android phone as handset, or on a Samsung device. 

The data presented at ADA was from their pre-pivotal trial, and the study had 36 participants, 18 children (ages 6-13) and 18 adults (ages 14-70). Objectives were to assess the safety and effectiveness of the system at higher glucose targets of 130-150 mg/dL as well as giving the patients free choice of target ranges from 110-150 mg/dL.

The trial period was split in two with the first period in the higher range and the second with free choice. Unsurprisingly, the biggest improvements were seen when patients were given the choice to customize their own glucose targets. During this “free choice” testing over 4-9 weeks, TIR increased from 65.6% in adults to 73.8% and from 51% to 70.1% amongst children. While this was only a small population, it looks promising. 

Given that’s the news, the next question is what happens if you put them side by side?

How do all the AID systems compare?

If we look across the various systems that have presented data, while it’s a little rough around the edges, we can see what the Time in Range (70-180mg/dl) is for each of the systems, but perhaps more equally importantly, the improvement in TIR in the population. Some of this is raw from baseline (here we’re looking mostly at Medtronic) while some is compared to the control group, which makes the ability to draw comparisons ridiculously hard. 

Time in Range comparison

 

As we can see from this graph, the time in range data across the different studies produces a variety of different outcomes. I’ve included the CamAPS, Diabeloop, FLAIR and 780G NZ studies as they present a key factor in this. The more challenging the study population is, the lower the TIR outcome is. This is often because the baseline that you’re starting from is considerably lower. 

Time in Range Improvement comparison

As we’ve already mentioned, TIR levels on their own don’t necessarily tell us a lot about how well a system performed, however we can take into account the initial baseline of a trial population by looking at the TIR improvement. 

I’ll admit that I’ve been a bit cheeky here and used the CamAPS data comparing their end outcome with the baseline of the AID trial group, when the comparison to the control was 10.8%, as it gives a direct comparison to the Medtronic pivotal trials. 

From this, we can see that the improvement on the 780G trial was okay, but doesn’t appear to be anything to shout about, however, when we look at the data that’s been provided for the other trials, we start to see something very interesting. 

1. The 780G in its out of the box configuration for “classical” clinical trial users has a marginal improvement in outcome compared to the 670G;
2. The 100mg/dl target and 2 hour AIT is needed to get the best improvements with the “classical” participants;
3. Control-IQ still achieves an improvement in TIR no worse than any other of the “Advanced” systems, as Medtronic likes to call them, and does so with a “classical” trial population;
4. Those using a more varied study population with a more challenging baseline see a much greater improvement in TIR;
5. Omnipod’s Horizon fits into this trend;

It does beg the question as to why anyone would use the 780G in its out of the box configuration if they already fall into the category of “reasonably managed”, or indeed, why Medtronic ship it in this configuration in the first place?

What conclusions can we draw from this?

It’s challenging. More than anything, the answer t the question “What do the trials tell us?” is be careful about the data that you’re looking at in a clinical trial, as it doesn’t necessarily tell you what you thought it did, as the design of the trial can bias the benefits seen.

Aside from that, it’s really hard to choose between the systems. Anyone who manages their diabetes reasonably can continue to do so with any of these systems. As far as we can tell from the non-uniform clinical trial data we have, taking into account baselines, TIR and improvements, you’re probably not going to be disappointed by any of them, and some will perform better than others in different circumstances. One of the factors I haven’t mentioned in this round of trials was the user satisfaction testing that was undertaken, and the outcome on this seems to be pretty high. 

Perhaps more important than anything else is that if these systems can be put in the hands of people who find managing type 1 difficult, it will significantly improve their outcomes.

We can also see that introducing microbolusing into the mix doesn’t immediately mean that the performance of the system will improve significantly. The quality and capability of your underlying algorithm plays an equally large part in that, as can be seen by the performance of CamAPS and Diabeloop compared to the 780G in trials with a wider range of capabilities. . 

What’s clear is that with the 780G, Medtronic appears at last to be catching up with the rest of the AID pack, but it’s also clear that the 800lb gorilla in the medical devices world doesn’t yet have the one AID to rule them all…