Conc'n of standard in a calibration curve?

[zimanli]

Greetings.

Currently we are using a method. Concentrations are 20, 50, 100, 200, 500 ppb.

my question is:

1. shall we have more than 6 points for basic pharmaceutical requirement?
2. In the concentration interval reasonable? This curve is going to have more points at low concentration end, and error for high concentration end will lead to large error to the whole calibration.

thanks

[shaun78]

1. USP et al. only requires that you use a minimum of 5 points to generate your calibration curve. No one is going to bother you if you decide to use 6.

2. In most cases, you would want to cover nominal +/- a range. Generally something like 60-140% is used. It is more desirable to spread your points out evenly across the range, but there is nothing that tells you that you must do this.

If it were me and nominal expected concentration was 100 ppb, I would test 60, 80, 100, 120, 140 ppb. While I was at it, I would show that analyte responce factor across that range was within a maximum of +/-3%. Then I would switch to using a single point calculation.

Naturally, this also assumes that you have approximately the same amount of accuracy/precision across the range...

Shaun

[chromatographer1]

IN MY OPINION............. which has no weight to influence anyone of consequence..........

I would also do multiples (6?) of standards near the expected level and only one example of a standard at the lowest and highest levels across the measurement range.

Assuming the variability is a function of injecting or making the preparation (like a splitter in GC) the ±1% error at the very high (out of normal measurement) levels might inappropriately affect the calibration at the middle range.

Again, just my opinion.

Rod

[zimanli]

chromatographer1,
I am a little bit confused.
You have a group of midle points, one high point and one low point. And then you said you think high point error might affect middle points. Why do not you spread evenly? Do you use fit parameters or do you only consider mid group numbers?

thanks
I can understand shaun78's method.

[shaun78]

I believe that chromatographer1 is suggesting that you simply study the points closer to your target more closely than the ones at the extremes. This is commonly done, though it is not written anywhere that it must be done like that.

Generally, when I run a validation I would follow USP category 1 (most of the work I do is on actives):

1. Precision - six independent sample preps. at 100%. Each preparation injected twice. Average two injections of each sample and calculate %RSD of concentration from six samples

2. Accuracy - three placebo sample preparations each at 60, 80, 100, 120, 140% of nominal, where analyte is spiked into placebo solution. Each preparation is injected twice, averaged, and rsd pf n=3 is calculated. Recovery must be within 3% of spiked amount at each level. Calculate response factor at all 5 levels. Compare all to response factor at 100%; they should match within 3%.

3. Linearity - five independent sample preparations (one at each level) that spans the linear range (in my example this would be 60, 80, 100, 120, 140%). Inject each sample 3 times, average, and plot. R^2 must be greater then 0.98 with intercept near 0.

4. Specificity - placebo injections as well as acid, base, heat, light, oxidative degradation if applicable.

5. Stability - re-assay nominal standard and sample preparations for how ever long you want to show stability for.

6. Ruggedness - another analyst/instrument/column makes six independent sample preparations at nominal. Same requirements as Precision. Second analyst must match analyst 1 results within 3%.

7. Robustness - GC initial oven temperature variation of +/- 20%, flow variation of +/- 20%, oven ramp rate variation of +/- 50% (or however much you can change it and still have system suitability pass). Requirement is that system suitability must pass for all changes (peak area ratio NMT 2%, resolution NLT 2.0, symmetry between 0.8 and 1.5, plates greater than xxxxx <depends upon column length>.

8. LOD/LOQ - make solutions/dilutions to try to get close to S/N ratio of 3 and 10, respectively.

[zimanli]

shaun78, this is VERY useful. Thanks a lot!

ziman