SmartGuard - One Month In

It's been one month since we first hooked Janki up to the 640g.
SmartGuard has been on for all of that time.

So does it prevent 80% hypos, as predicted from the clinical trial work? I don't know: we'd need a randomised controlled clinical trial with a bunch of Janki-sized T1Ds to answer that :-)

Does it "soften" hypos? I think so, but again we'd need a RCT in children to answer it conclusively. We've certainly been more confident at treating with fewer carbs (5g instead of 10g) and slower (not slow) carbs based on an early, predicted alert. We've also been able to adjust meal times, bringing a meal or snack forward once alerted by the predictive suspend.

There have been some surprises and - as ever - some questions that I haven't yet got my head around. Read on if you'd like to know more...

If you're 'new' to SmartGuard, please read here first.


The Data
Big D I'm afraid - I've gathered this from the CareLink CSV output file, and constructed a simple, but huge (and inefficient) MS Excel spreadsheet to extract the info and calculate the data below. Because of the data size (~34k rows) and the number of logic steps built into the spreadsheet it floors my laptop for a few minutes each time I hit Calculate :-) I'll blog some more about extracting more from this file - it's packed with (potentially) useful info that you can dig into and that doesn't appear to be readily available elsewhere on CareLink. I've already put some of the data up on the blog here.

So, we've used 24 days of continuous data.
During that time, SmartGuard triggered 82 times (please remember we had a chaotic first few days in particular whilst basal rates and sensitivity factors were dialed in...) It was active (i.e. Insulin delivery was suspended / cancelled) for 13% of that time, with a mean suspension time of around one hour (the maximum was two hours, as stated in the User Guide). The minimum automatic suspension period is thirty minutes, which we've overridden on a few occasions and you can see the mode (the most frequent) duration is the 30 minute minimum:
[Edit: Apologies, the incorrect SmartGuard Duration graph was initially shown here]

To put that into context, the last two (slightly quieter) weeks have seen 41 suspends, with SmartGuard active for 6% of the time and a mean suspension time of 50 minutes. The graph below shows the times of the day and night when SmartGuard is most active (much work still to be done with rates and factors as you can see):-

I should make it clear that the "chaos" mentioned is not due to the 640g, it's due to having to set Janki's rates without prior CGM and with little long-term stability leading up to the switch to the pump.

Suspension
I was interested in doing a little digging into how the pump decides to go into suspend and activate SmartGuard. Obvious candidates are Sensor Glucose (SG), rate of change (fall) of SG (labeled as Delta SG below) and - maybe - Insulin On Board (IOB).

The mean SG at suspension was 5.9 mmol/l (range 4.1 - 7.8 mmol/l).
The mean rate of change in SG (delta SG) immediately before suspension was -0.1 mmol/l per minute, or a predicted drop of 1 mmol/l in ten minutes.
The maximum delta SG we recorded at suspension was -0.4 (4 mmol/l in ten minutes, which is just the kind of rate that led us to go for the 640g...)

Allowing for the rounding errors in using the SG data (and not the ISIG numbers...), you can still see a clear correlation with these data: as might be expected, a higher rate of change (fall) tends to trigger a suspension at a higher SG value.

According to Medtronic, the SmartGuard feature should give us thirty minutes warning of a hypo*. Amazingly, just based on the immediate preceding SG data (ie at -5 minutes and at suspend), the mean time to hypo is 25 minutes (mode - i.e. most likely - is 29 minutes). However, whilst this suggests a very significant component is played by these two SG readings, it's not the entire story, as the correlation is not perfect.

*According to the System User Guide, SmartGuard will only trigger if the SG data point is at or within 3.9 mmol/L of the low limit and is predicted to reach or fall below a level that is 1.1 mmol/L above that lower limit within approximately 30 minutes.

Resuming Basal Delivery
Although, I guess, most T1D parents would vote hands-down for hypo avoidance / mitigation rather than worrying about hyper events, we were also keen to ensure SmartGuard wasn't going to lead to significantly more hyper events. We've certainly seen a large number of hyper events. SmartGuard is probably not responsible for these, but it has complicated (by suspending basal delivery) unpicking what is delayed carbs and what is - potentially - due to insufficient basal insulin present. We don't have the data to draw any conclusions yet. Having spent time with our clinical team today, we do intend to curtail SmartGuard's action in the run up to bedtime and during the evening (whilst we're still awake), manually resuming basal when we feel it is safe to do so. By doing this, we'll hopefully get a quicker resolution of our overnight basal settings, which will allow us to tackle some of the other parameters with more confidence.

So, left to its own devices, we see SmartGuard automatically resume at a mean SG of 6.1 mmol/l, with a range of 3.8 - 12.3 mmol/l. The "3.8" was due to a two hour time-out overnight and was on a new sensor (i.e. Janki wasn't left hypo for two hours). The higher numbers are due to the 30 minute minimum suspend period (if left alone).

The mean rate of change (rise) in SG was 0.12 mmol/l per minute (i.e. 1.2 mmol in ten minutes), with the maximum rate of increase recorded at 0.5 mmol/l per minute (reflecting an over-eager hypo treatment episode).

Interestingly, I don't see an obvious correlation in rate of increase and SG at resumption of basal delivery. It may be that the algorithm makes use of a longer cycle of SG values (the data above represents just the SG at resumption and the figure 5 minutes before resumption), but more likely it's the constraint of the 30 minute minimum automatic suspension period.

More data and more thinking to follow in the weeks ahead...