Bioequivalence and Bioavailability Forum

Hi Loky do,

❝ Regarding BLQ samples in the middle of plasma concentration curves, which remain BLQ after reanalysis, what is the best practice for those samples in calculations?

My personal points of view…

❝ To be considered as (missed or Non-reported samples …

Why? You measured them (confirmed even in a second analysis).

❝ … or blank)?

Do you mean setting them to zero? If yes, that would not make sense.
BTW, replacing BQLs by LLOQ/2 does not make sense either.¹

❝ Or to keep it as it is <LLOQ?

Any nonnumeric code would be fine.² However, I recommend to give them with the more common code ‘BQL’ and use the linear-up / logarithmic-down trapezoidal rule. If you use the linear trapezoidal rule, the AUC will be positively biased (i.e., the value ignored and a trapezoid calculated connecting the time point before and after with a straight line). See this article (esp. the example about missing values).

Of note, the lin-up/log-down trapezoidal is recommended in the trainings for Phoenix/WinNonlin (Certara) and PKanalix (LIXOFT). It is the default in the -packages bear, PKNCA, and ncappc.

---

1. I use it only for plots since I always give the geometric mean.
   
2. If you want to make assessors, inspectors, and consultants happy: Use more informative codes in the bioanalytical report, e.g.,
   • ‘BQL’ for values which where <LLOQ,
     
   • ‘NR’ for not reportable ones (say, problems in sample processing or chromatography),
     
   • ‘missing’ for no sample (vial broken, etc.).
   All nonnumeric codes are ignored in software. However, data pretreatment is a must. A common procedure is to set all nonnumerics before t_max to zero and keep the code for later time points.