I really enjoyed this paper, thank you for this contribution as the back transformation of limits of agreement would be very useful for people quickly looking at data. However, I found a couple of drawbacks. Firstly, there is no accounting for systematic bias in the formulae you propose to calculate the mean measurement dependent limits of agreement. I propose that this can be countered by using any systematic mean bias calculated from the absolute value differences, and adding this to your formulae. However, there remains the second problem that your limits of agreement will decrease to zero as the mean measurement decreases to zero. For any measurements that include numbers close to zero as common and meaningful (for instance, a large proportion of laboratory serum sample tests, such as CRP), this is inappropriate, as a lot of method comparison studies will show a variation near these values. If there is any way of accommodating this, or anyone has any suggestions, I would really like to hear about them! Many thanks.
Article