forecasting pre-fatigue using glucose levels thresholds

Greetings

I have long term ECG data (HR,HRV) from helthcare workers recorded during first stage of the current pandemic. my goal is to use ML classifiers to detect their fatigue levels (as they literally confirmed at the workeplace). However, before get into it, i want to generate a pre-fatigue (as a warning) method by reffering to the strong correlation between fatigue and low blood glucose ( low blood glucose plays a role of fatigue warner). Not having glucose data, and considering HR as a time series data, i thought to implement methods that use thresholds, meaning to set up certain thresholds of hypoglycemia values instead of the data (fore example fatigue <69-60mg/dl, no fatigue >69-60mg/dl) but till now i could not find any clue about it nor where to start. kindly help me with a method that can get along with it.

thank you for your support

[Star Strider]

‘Not having glucose data, and considering HR as a time series data ...’

You have just happened to wander into an area of my interest (and to an extent, expertise). If you are attempting to infer plasma glucose levels from the heart rate (or other EKG characteristics), I doubt that is going to be productive. There just aren’t any relkiable ways to do that in the normal physiological glucose range, although hyperkalaemia (from decreased insullin activity) could show up as low P-wave and high T-wave amplitudes, it might be difficult to discern those from normal variations. Certainly hypoglycaemia could produce tachycardia because of elevated epinephrine and norepinephrine, however that would be only in the extreme and would be confounded by other known variables in healthcare workers under stress. I seriously doubt that you could accurately estimate the actual plasma gluose concentration on the basis or other physiological variables without actually measuring it. This has been something similar to the 'Holy Grail' in diabetes research for closed-loop plasma glucose control for several decades. Directly measuring it (now possible with minimally invasive plasma glucose monitors) is the only option.

If you have not already done so, do a PubMed search on this before you go further.

[Tasmia Avouka]

Thank you for your feedback. However, could you suggest me any other method that would fulfill the need without the glucose data?

[Star Strider]

I am not certain how to monitor or detect fatigue since I have never done that sort of study, and I do not know what physiological variables you have other than what appear to be Holter monitor records. Perhaps looking at the EKG intervals themselves, heart rate, heart rate variation, or any ST-T changes could be surrogate information. There could be other information in the EKGs themselves, although repeating the study with minnimally-invasive plasma glucose monitors is an intriguing possibility. Again, I encourage you to do a PubMed search for that information. Also, consider other metrics such as Galvanic skin resistance (GSR) or metrics of circadian rhythm variations, such as body temperature, if you have those data.

[Tasmia Avouka]

ok thank you

[Star Strider]

My pleasure!