Written and clinically reviewed by Marissa Town, RN, BSN, CDCES
If you wear an insulin pump, you were probably told, “When in doubt, change it out.” That means that if your blood sugar is high for a seemingly inexplicable reason, it’s likely the infusion set not working properly. Some researchers are trying to create infusion sets that last longer in the body, and others are working on algorithms to detect infusion set problems earlier.
As discussed in this article from the Journal of Diabetes Science and Technology, ways to identify infusion set failures automatically have been investigated, but not tested on real data.1 The authors acknowledge the challenges in creating an algorithm that will allow for the differences in persons with diabetes, but are hopeful that they have found a valid possible solution.1
Researchers tested three anomaly detection algorithms that find faults in data based on the person’s insulin pump and continuous glucose monitor (CGM) history.1 The algorithms look at multiple things, including the increase in blood glucose levels, increase in insulin administration, and length of time the infusion set has been worn to help determine the prediction of a failure.1
The researchers used the algorithms on data from a study from 2013, where the participants wore infusion sets on areas with lipohypertrophy (abnormal fatty tissue from frequent insulin injection).1 They used this data because the purpose of the research was to wear the infusion sets in areas where they were likely to fail, and to wear them for longer time than is typically recommended (7 days vs. 2-3 days).1
In the best circumstances, the algorithm was able to detect 15 out of 20 failures, which makes it better than the previously tested algorithms.1 This could be beneficial, as more and more people with diabetes gain access to insulin pumps and closed loop systems. Since the study was completed on computers using data that was already collected, the algorithm needs to be studied in clinical trials to see if it is effective when used in real-time.
Either way, this approach sounds more accurate than me, as a person with diabetes, asking people to smell my infusion sets to see if they can smell an insulin leak! Here’s to making progress with detection algorithms and finding ways to make diabetes devices work better.