With the prevalence of mobile computing, self-monitoring and Big Data technologies, new approaches to the management of type 1 diabetes management might be possible.
Take for example the administration of insulin in type 1 diabetes when eating pizza. It is often reported that rather than taking one single bolus insulin dose when eating pizza, the insulin dose should be spread over an hour or two – this is commonly referred to as a dual-wave or a combination (“combo”) bolus. The theory is that the high fat content of the pizza delays the ingestion of carbohydrate and so the duration of the insulin should also be extended by splitting the bolus into two smaller doses (e.g., half the insulin when the pizza was eaten, and the other half after 60 minutes). Exactly how much insulin should be taken, and when, is very specific to the individual and the pizza.
Compare the different approaches to better understand this phenomenon (the method and data is fictional and is used to illustrate the approach):
The method of a clinical study of this effect may look something like this: 30 people were given a cheese and tomato pizza to eat (Domino’s Margarita 10″). 15 took a single bolus which was appropriate for the amount of carbohydrates in the pizza, in line with their normal dosing regimen. The other 15 took 50% of the bolus immediately and 50% after 60 minutes. The rate of post prandial hypoglycaemia was recorded.
This approach has the benefit of being very controlled, but the downside is that while the results are very applicable to the particular population and to the particular pizza, the real-life situation may be different. A person’s particular physiological response may be different and it might be a different pizza. Of course the number of people and types of pizza could be increased, but this would be expensive and time-consuming.
Imagine an application which logged food, insulin and blood glucose levels and was used by 100m people. The application would have a large amount of data from which to provide suggestions based on the actual pizza eaten and the actual parameters of the bolus (i.e., a single or a combo bolus) taken at the time, and the resultant glucose levels. For instance, if you were about to eat a Papa John’s pizza 9″ Meat Feast pizza you could type this into the app and it might come back with: “85% of people (135,198 people) who took a single bolus went hypoglycaemic within 4 hours of eating the pizza. 23% of people (892,762 people) who took a combo bolus went hypoglycaemic within 4 hours. The combo which produced the lowest rates of hypoglycaemia was to take 60% of your usual dose now, and 40% after 60 minutes”.
The benefit of this approach is that the sample size is very large – a lot of data, including the outcomes, is logged and can be mined. It can provide aggregate data which is, for instance, specific to the actual pizza being eaten. However, the physiological response is an average and may not be applicable to the user, and the data might not be “clean”. For instance, someone who went hypo after eating a pizza may have done so because they went for a run, not because they took a combo bolus.
An extension of the aggregate approach is to provide suggestions based on the actual individual’s experiences. This is what the team at Tidepool are looking at. If I type into the app that I’m about to eat that 9″ Papa John’s Meat Feast, it could pop up and say, “the last time you ate that pizza you took 1.65u using a 60:40 1 hour combo but you went hypoglycaemic 55 minutes later.”. Furthermore, the data could be analysed to provide a suggestion of better bolus parameters. “Based on similar carbohydrate:fat ratio in the pizza to other food you have eaten, a 50:50 90 minute combo might be better”.
Of course there are significant perils (both legal and health-related) associated with letting software determine the level of medication, but it is a very exciting area which might, at the very least, provide new insights into the management of diabetes.