Relative response factors - analytes at the same conc.?

[Rob Burgess]

Hi All,

I am debating internally at my company that when determining relative response factors (RRF) in HPLC, do the analytes necessarily need to be at the same concentration levels on the linearity plots (i.e. RRF's are being calculated from the slopes of the linearity plots).

For instance is it acceptable to compare slopes from the main analyte peak at say 10 to 150% of a nominal loading with impurity linearity slope from 0.1 to 1.0% (assuming 4-5 data points across these levels)? Or is it preferable to perform the linerity loading plots at equal concentrations i.e. both main and impurity at 0.1 to 1.0%?

Statistically/mathematically which is the preferred approach - equal concetrations for main and impurity peak - or it does not matter?

[tom jupille]

Ideally it shouldn't matter -- so long as you stay within the linear range in both cases.

As a matter of practice, I think you would be better served working at equal concentrations. The argument for that is the same as the argument for using weighted least squares instead of ordinary least squares for the plot: chromatographic data are inherently "heteroscedastic" (that's my six-bit vocabulary word of the day!), which means that the absolute variability is not constant across the entire range (percentage errors tend to be constant, so absolute errors increase at the high end). The result is that variability near the top of the range tends to swamp out variability at the bottom, and the results at the low end can be wildly off.

If it were my problem, I'd do it both ways at least once and then decide based on the results.

[mattmullaney]

Hi Rob,

I agree with Tom and for exactly the same reason(s). The RRFs you determine will eventually be used so that standard curves will not be generated for subsequent analyses for the analytes that are not likely to be present at comparable amounts to the "main" analyte of interest. It seems to me that it makes some sense to compare like concentration levels in pre-validation/validation of the analytical method between the main analyte and the other analytes that will be present only at a fractional amount of the main analyte.

It's also a one- or perhaps several-times demonstration...and it removes any question(s) an auditor would possibly bring up at a later date. For me, this is (sadly?) another "reason" to prefer the approach of calculating RRFs in which the main analyte curve is created at similar concentration levels to the minor analytes.

All of this said...it shouldn't matter which way you do the work to determine the RRFs...as long as the curve fitting the main analyte may be modeled the same way as those curves fitting the minor analytes and all data for all analytes are homoscedastic.