calibration curve

[syx]

In most LC softwares we used, in the calibration curve section we often find "force the origin" or "force through zero" option, but we are not sure of how or when to use this option. Some literatures mention not to force the calibration curve through zero. So the question is, are we really allowed to force the calibration curve through zero? in what case do we force the calibration curve through zero?

[aceto_81]

When you validate your method, you verify your linearity.
WHen doing this, you also check wether or not your intercept confidence interval includes zero. If this includes zero, you are allowed to use a force thru zero calibration, which only needs 1 datapoint to calibrate.

If you don't force thru zero, you have to calibrate on at least 2 different concentrations which increase the workload.


Ace

[mbicking]

Ace is correct.

You should always begin a new calibration with standards that bracket your expected sample range. The regression analysis should be standard slope-intercept methods (do not force to the origin) to evaluate if the actual intercept is different from zero. If not (that is, within the confidence interval for the intercept - calculations easily done in Excel), then you could switch to forcing the intercept to zero.

As Ace noted, a zero intercept means that you can proceed with a single point calibration system if desired, although there are situations where this approach is not always the best. But if you continue to use multiple standards, then I prefer to continue to use slope-intercept, since it allows me to monitor these values, to make sure that they don't change significantly.

Some data systems offer an "include zero" option for the intercept. This approach just adds the point (0,0) to your curve, and would tend to force the curve toward the origin. I do not like this approach. You want to determine if the curve really goes through the origin, not force it to go there.

[Slammy1]

Actually, intercept may be an indication of different issues (such as matrix) but if you're not running an impurities method, it's not that critical so long as you operate within the method's accuracy range and are specific. We'll typically run intercept as a percent of the 100% standard for validations, and perform bias testing for determining the number of calibration points needed (typically <2.0% for intercept and < 1.5% bias in the analysis range). It's quite rare to see a force through zero in pharma, though I've seen some people do it on some "questionable" methods...

[syx]

Thank you for your replies.