Greetings,

I am currently writing a validation protocol for an Assay and Related Substances HPLC method. After performing linearity from LOQ to about 5xLOQ and 60 to 150% of active, I am trying to evaluate my results. The two curves are very linear but the y-intercept is significantly different than zero at the RS level. I am using a percent bias calculation {% bias = [b/(mX+b)]x100% where b is the y-intercept, m is the slope, and X is the concentration at level 3 of my 5 data points}. Is anyone familar with this calculation or acceptable limits to the bias?

I am trying to use a single point standard calibration for the method and believe I must show that the y-intercept isn't significantly biased. Is this true?

thanks in advance.

You are right when you say that you have to show that the y-intercept isn't significantly biased.

How did you prove that your intercept is significantly different from zero?
Because when you use some statistics like ANOVA, you can show an interval, so where you can say that you are for 95% sure that the intercept is between these values. If 0 is in between this values, there is no problem at all.

I did not use a calculation to prove the intercept was significantly different than zero. The intercept value is about one-third the y-value for my 3rd of 5 data points. Common sense suggests the intercept value is very different than zero over this range.

I guess I am trying to use the percent bias equation to determine an acceptable deviation from zero for the intercept. I am trying to avoid using confidance intervals, even if I am 95% sure of them.

thanks again.

Hi Kris,

Just a few quick thoughts.

I see lots of protocols that use a criteria for the y-intercept of NMT 1-2% of the 100% area. If you are trying to quantitate related substances and assay active with the same injection, these criteria are probably useless.

When you are doing linear regressions with HPLC data over large ranges you have to be careful because random errors at the high levels will dominate the statistic, and can generate nonsensical intercepts. HPLC data is not homoscedastic and is not really suitable for unweighted linear regressions. This is also why an ANOV analysis of the intercept is also useless.

I think the best approach is to look at the specific respose of the entire range LOQ - 150%. Where specific response is area/amt and plot this against % label claim or whatever. If your specific response is constant +/- whatever % accuracy you need then a single point calibration will be sufficient.

This thread covers the same ground. Sorry, its long. Good luck.
http://www.lcresources.com/discus/messages/2401/2367.html?MondayApril2220020506am