In order for insulin to be dosed -precisely -sensor readings need to be accurate as much as possible. and this is subject to extensive researches done by various companies aiming to enhance :
1-sensor performance/ consistency
3-cutting of possible interference through using special mediators of selected characteristics
4-keeping the sensor its place during the whole wearing period by wiring the enzyme by the aid of specific polymers
5- glucose concentration must be kept in certain limit to avoid misleading readings -a special membrane is used to control diffusion backward and forward.
6-CLS (closed loop control) is one way of controlling insulin pumping based on output sensing ,where outcome feedback is considered before any amendment is needed.
ex. dryer machine is nice /clear example where wetness of clothes is considered before any required/extra action :
if clothes are less wet ; less drying time is needed while if more wet ; more drying time is needed ..etc
i.e clothes drying is sensed through a special sensor; the the signal is sent back to the control before any further action (just a simple example,matter is more complex !) however :
1- "feedback signal shroud be evaluated -mathematically- within the processor to avoid frequent/unneeded adjustment which may cause oscillation/breaking of the whole system
2-this process should be on continuous interval to avoid any accidental change -esp.. in dynamic system.
6-Prediction is the most important part of CGM ,
by the way " retrospective data analysis and proper prediction algorithm are two features were added to CGM sensor output; in order to compensate for sub optimal accuracy resulting from lag time ...for example !"
any prediction -in order to be reasonable :
1- it must take into account previous data and history so patterns and trends can be extracted/extrapolated from them by careful mining and systemic analysis. (machine learning is a must , advanced insulin pumps where a partial closed loop system is implemented , selected time must elapse before turning in auto mode "CLS" so the machine could learn better about each individual) ;
"Remember that each person reacts differently to each triggering action ,even the same person differs in his response from time to time "
2-if prediction is an independent output ,that means many dependent/variables inputs -if included-will make it closer to reality and away from false estimation so more inputs means more precise prediction .
See below simplified example :
1- imagine that one employee stays in his office from 8 am to 5 pm from Monday up to Thursday ,according to this data assuming it for year ,we say that he may be present there on 1st of July if its working day- in the next year will be 90 % lets say..
2- however if we know that guy takes vacation on June /or July or august ,we will be less certain of his presence in that date..
3-however imagine we got more data of this employee that in the previous 6 yrs he took vacation on July,:4 times -one on june and one on august now, we become quasi certain that we may not find him on that day (our prediction becomes more precise as we got more data !
4- now,imagine his dad is sick and about to a undergo surgery on 30 of June ; so that employee is most unlikely to be found in 1st of july (prediction becomes much more certain) !
5-more data means more certainty !!!!
To sum up,
more than 40 factors affect blood glucose -in addition to various types of lag time -discussed later- so its so important to figure them out once applicable -otherwise our prediction will be non logic and misleading in many cases...