I think the opposite is true. To achieve medical usage the correlation should be linear. If this is not the case, a small variation of glucose could lead to big

I think the opposite is true. To achieve medical usage the correlation should be linear. If this is not the case, a small variation of glucose could lead to big jump in the reading, which would decrease the accuracy. In real life, most of the biological system or test exist in a buffered environment, which often show a range of linear relationship for substrates, and tests are often performed by adjusting into such linear range.