negative y-intercept value

[EVEL]

Currently performing a linearity study on a related substances HPLC-UV method for a non-UV absorbing molecule. The method uses Nitric acid as mobile phase and a PRP-X100 column; the analyte peaks elute as negative peaks.

A linearity study was performed on concentrations ranging from LOQ (0.1% of spl conc) to 120% of limit concentration (0.28% of sample concentration); this gave a good correlation co-efficient value (> 0.99) however the y-intercept wrt 100% spiked solution obtained was -17%.

A seperate linearity study performed at higher concentrations gave a y-intercept of 0.5%.

It is evident that in this case, the lower the concentrations, the more negative the y-intercept being obtained.


Would appreciate if anyone can give me any pointers with regards to:

- what is the significance of such a low -ve % y-intercept value (-17%)?

- how does this result impact the suitability of the method?

- are there any suggested modifications that can be done to the method to improve the y-intercept? The sample concentration cannot be altered since it is the highest possible concentration to give an adequate S/N ratio at LOQ concentration, while maintaining an adequate specificity between the main peak and other known impurities in the test solution.

[Peter Apps]

The problem is not the y-intercept, but the x- intercept - the non-zero quantity of analyte that gives a zero peak area,and correspondingly increased LOD and LOQ.

Just a guess - are you sure that there is nothing (besides the mobile phase) that absorbs UV eluting at the same time as your analyte ?

Peter

[lmh]

also, how linear is your calibration curve? R-squared values are often a dreadful measure of linearity - they can look really high, while the residuals (deviations from the straight line) are obviously very non-random (high at the ends and low in the middle, or vice versa). If present, deviation from linearity can be either a genuine effect of the detection system (in which case I personally think it better to find a fit that matches the detection system, rather than pretend it's linear, but I don't have to deal with regulatory bodies who can't understand anything that doesn't go in straight lines) - or it can be a problem such as Peter described, or a limited number of binding-sites for an analyte in effect titrating out the analyte at low concentration.

[mckrause]

You fail to mention your wavelength that you are using. Nitrate absorbs intensely in the low UV; you may be seeing the effects of trying to measure small changes in a large absorbance.

As lmh points out, definitely run residuals on your curve. We normally toss relatively linear curves into log space and then do a least squares fit; it does a better job of fitting all of the points, not just the high end of the curve.