Automated algorithms: A useful tool to optimise insulin pump therapy?

Automated control of insulin delivery is perceived as an increasingly important development in helping optimise glycaemic control for people with type 1 diabetes using pump and sensor technology to manage their condition. After initial suspicion, the US Food and Drug Administration has approved low glucose suspend systems, paving the way for commercialisation of treat-to-target systems and, in due course, closed-loop systems. These are sophisticated and are likely to be expensive, even in comparison to current sensor-augmented pump therapies. Can insulin adjustment algorithms be harnessed in a simpler and more widely applicable way to help optimise existing insulin delivery systems? 

The University of Melbourne diabetes team have previously reported the utility of the ALGOS algorithm in assisting people using sensor-augmented insulin pump therapy with adjustments to their insulin delivery settings (Jenkins et al, 2010), and other units have described similar algorithms. These all rely on the user to make the therapy adjustments themselves.

Now, in the article summarised alongside, the team from the University of Toronto have attempted to develop an automated system to adjust insulin pump settings, using an algorithm accessible through a web-based interface to adjust overnight basal insulin delivery on the basis of continuous glucose monitoring (CGM) data obtained during a controlled basal rate assessment. The basal insulin infusion settings were adjusted gradually on the basis of five assessments performed over a period of 2–8 weeks. Glycaemic outcomes were assessed in terms of the change in a 3-day CGM profile performed pre- and post-intervention. The change in overnight basal insulin delivery following these automated adjustments resulted in significantly less time spent in hypoglycaemia and an average HbA_1c reduction of 2 mmol/mol (0.2%) despite the short duration of the study. However, perhaps surprisingly in view of these changes, there was no improvement in glycaemic variability overnight.

Does this relatively modest benefit mean that such automated algorithms are unlikely to be a useful adjunct to optimising insulin delivery via a pump? It is worth noting that this was a pilot study and the adjustments made to the infusion rates following each assessment were deliberately conservative. Learning from these pilot data should allow more aggressive adjustments to be made, which would result in greater improvements in blood glucose variability in a shorter period of time. Furthermore, basal rate testing is dependent on a lack of interfering factors, such as active insulin from prior boluses and prolonged glycaemic excursions following fat- and carbohydrate-rich meals, alcohol and exercise. Repeated iterations of this assessment and adjustment process could minimise the impact of such interference, making this automated method of basal rate adjustment more reliable and effective than current systems of basal rate testing, which are usually done once every 3 months at best!

Current guidelines suggest sensor-augmented pump therapy has a limited role over and above stand-alone pump therapy (NICE, 2015), and most pump users are not that keen to add further invasive technology to their existing pump therapy. However, intermittent use of CGM for automated optimisation of insulin pump settings could prove particularly effective, so further development of such systems would be a welcome advance.

To read the article summaries, please download the PDF