Artificial Intelligence (AI), Machine Learning (ML) and Diabetes Research

Although this is not that ground-breaking, it’s interesting to note that the ML model had better accuracy than the logistic regression, which I assume is expert-driven. The part that is promising, is that AI/ML, combined with the troves of data being stored, might find better treatments, or at least minimize harm from unintended side effects.

One wonders what other types of ideas can come out of this, relationships between lifestyle factors, medication, personality, and treatments that are beyond the eye of doctors and researchers, but that an ML model might be able to find and incorporate.


I always laugh at what they consider hypo.

This included laboratory and point-of-care blood glucose (BG) values to identify biochemical and clinically significant hypoglycemic episodes (BG ≤3.9 and ≤2.9 mmol/L, respectively).

Oh no! Sound the clarion call! Stop the presses! I am 70!

1 Like

Well, they set to points, one at ~70 and another at ~50. Although neither seem severe, e.g., needing assistance, still worthwhile to analyze.

1 Like

I know some people with very stable blood glucose and/or who have very nuanced control over their blood glucose, where their blood glucose rarely or never “doesn’t behave” according to what direction they’re trying to steer in, are comfortable staying at those types of numbers for extended periods of time. I’m not one of those people. I think it’s important to remember that most people with diabetes, even those with well-controlled diabetes, are not in that category, especially when in hospital, where they may not even ab able to treat their own diabetes. It doesn’t seem at all unreasonable to me to consider 2.9 mmol/L to be clinically significant hypoglycemia.