Bioequivalence and Bioavailability Forum

Dear Thi Nguyen,

❝ BUT: core run 1 has one LLOQ 498% and one LLOQ 727%, the other four LLOQs have good accuracy 〜100%. Core run 2 has five LLOQs with accuracy 〜100% and one LLOQ 3000%. For core run 2, 3000% can be regarded as outlier (from Grubb's test - calculated by Prism) but for core run 1, both values 498% and 727% are not outliers.

Forget about outlier tests in such situation. The validation is there to demonstrate that your method works. If you exclude those values that may indicate it doesn't, you're losing the plot, whatever the result of the outlier test.

Now there are several possible explanations, including contamination, sample spiking error and carry over. It is rather difficult to comment further without more information. I would go back to the data and look at, for instance:

• how were the samples spiked practically ? Risk of contamination of LLOQ samples ?
  
• do the concentrations measured match with the nominal concentration of LQC, MQC or HQC samples, or of a calibration sample ? Could be just the use of the wrong spiking solution.
  
• were these samples injected right after a sample with a high concentration (carry-over) ?
  
• are there "strange" results at other QC levels too, even if "less" abnormal ?

Plenty of possible reasons and aspects to look at. But don't exclude any result unless you have identified what happened... And "we must have made a mistake somewhere" is not considered an "identifiable cause".