HPLC Related substance analysis by using Slope & intercept

[ananthjkumar]

How to use the negative intercept for the calculation as such or change the sign.

Actual calcualtion is (Peak area-Intercept) ,If peak area is coming less than intercept, the result value showing negative,then what is the result?

[pharmaguy]

Our standard response to this situation has been to list value as <LOQ as stated in the method, this is assuming your (-) intercept is relatively close to zero.

[lmh]

I'd extend Pharmaguy's answer.

The fact your result cannot be correct (because it's negative) has waved a red flag that you've correctly seen. However, if you'd looked at samples containing just a bit more analyte, the answer could have been positive, but still very, very wrong.

If your calibration curve begins to deviate from the measured standard points, then it is no longer reliable in this region of concentrations. It'd be better to assess limits of detection and limits of quantification routinely, and flag all values below LOQ, negative or positive.

If your calibration curve deviates from linear, but always does so in a repeatable way, it may be acceptable to fit a curve. I must admit I have occasionally forced a curve to the origin, but this is also frowned on; there was a thread back here and an interesting LC-GC article a while back which concluded it's only appropriate to do so when the curve already passed insignificantly close to the origin (in which case it makes no difference anyway). This does eliminate the possibility of negative results, but the important question isn't whether the results look possible: it's whether they are correct.

[ananthjkumar]

Thanks for both

The intercept sometime getting positive and negative also.

Calculation, the same sample with positive intercept it give low value at the same time with negative intercept ,it give high value of impurity

[tom jupille]

What are you typically getting for the standard error of the y-intercept?

[pharmaguy]

Very good clarification Imh.
This is an inevitable (good) problem when you get very good linear response for a compound even at very low levels, unfortunately, it is near impossible to get a zero-intercept repeatably so you have to deal with it.
There are two 'solutions' I've used, with a curve we design the lowest standard to be at or below the regulatory limit and report as <LOQ for peak areas below it, or we prove linearity during validation and use a single-point, which is, in essence, a line through the origin, and guarantees a result with any detectable peak.
Each has it's drawbacks, and I'm surprised no regulatory agency has established a guideline on how to handle this better.