Bioequivalence and Bioavailability Forum

Dear D. Labes & Yung-jin!

❝ In the manual selection of time points I have observed:

❝ Sometimes the very last point could not be selected.

Confirmed. First I thought this might happen when the last datapoint is very low (close to the x-axis), but this is not the case. Sometimes I couldn't select a datapoint at ≈50% of the y-axis; I don't see a clear pattern. Dataset upon request.

BTW: Why 2-4 points - not more?

---

Edit: Well, the datapoints were selected, but the label was printed outside the viewport. Better if I moved the graphics window to my second screen in full resolution (1280×1024), but even on my first screen, full res (1600×1200) sometimes the labels were only partly visible. After getting the list of timepoints (hurrah - all are there), and saving the results, I got:
Error in pooledSD(object) :
  No degrees of freedom for estimating std. dev.
... and end of story.

Dear D. Labes,

Fantastic suggestions. I did have the same situation (it just crashed there!) when using automatic selection algorithms (TTT/AIC/ARS with combinations). The only way to get the job done is the manual selection. Your suggestion sounds feasible. It only needs to write several screening functions for each automatic selection algorithm to scan/check the all data points by subject first, and report if the selected method is applicable for the data set. Even we can show list of subjects who cannot use the selected method. Great idea.

❝ I have some another ones, but more bullet proofs rather than new features:

❝ [cut here]

❝

❝ Moreover: In the manual selection of time points I have observed:

❝ Sometimes the very last point could not be selected. But

❝ [cut here]

Yes. I understand. What I am doing, when selecting data points, is to try not to click the exact location of the data point (sounds weird?), especially when there are crowded data points around or when the data point is very close to zero (as Helmut said above). You can do data point selection by just clicking around the data point (it's a trick!). The display screen of the graph seems cheating on us (?). In bear, the selection of data point does not necessarily to click the exact location of data point. I don't why, but it is just like that. You can try this next time. I know it's really frustrated when you try to select data point showing in the graph and get no luck.

❝ BTW: If you have very much spare time I would highly recommend to refactor the bear code. It's a nightmare to make a

Very true! I have been working on it whenever I revise it. Very sorry about this. As the matter of the fact, I always wish in my mind to invite some colleagues (you, Helmut and many others) in this Forum to together maintain bear if they are interested in coding R. However the source code of bear is really a mess right now. I wish I have much time to do this in the future. Thank you for your messages.

Dear Helmut,

Thanks for your messages. Yes, I do plot the conc. vs. time on a semilog scale with mean + sd. Hopefully, I wish that I have fixed bear to take care of this, such as using 'foo <- na.omit(foo)' for conc. < LLOQ. And I find the package I used (sciplot) can ignore the data point with value of zero (the first data point). For some reasons (which I still cannot figure out), semilog plots do not show at all, specially for multiple-dose, parallel study. I don't know if it is data dependent. However, the linear plots always works very well. Probably, I should consider to include both linear and semilog plots.

❝ Problems arise if at least one concentration is <LLOQ. Some people (like me) don’t report the geometric mean for this time point, others calculate the mean if ⅔ or ½ of concentrations are >LLOQ (excluding the BQLs). For the plots I would suggest to use backtransformed values – either in linear scale or with a log-axis.