I have clients conducting LD50 studies. I sometimes find their data
is not sufficient to estimate the LD50 and 95% confidence intervals.
However they have sacrificed laboratory mice to no avail. That upsets
me. I want to be sure they bring me quality data that I can give my
clients good estimates, so they don't sacrifice the laboratory mice
Therefore I am looking for information on software to design LD50
studies that ensures that sufficient data is collected within 5 doses to
estimate the LD50 and its 95% confidence interval.
My concern is that when an initial dose leads to 0% mortality one has
little to guide you to the next dose. However once that next dose
is selected the dose interval is determined. Eg if X is the initial dose
with 0% mortality and you chose a higher dose of Xd, then you must
continue with (Xd)d, (Xdd)d, and (Xddd)d. What do you do if the mortality
for the last dose: (Xddd)d is only 40%. Clearly the dose multiplier d was
too small to get sufficient data within 5 doses. Does there exist
software that will guarentee that the dose multiplier chosen ensures
the 5th dose results deliver a mortality greater than 50%, preferably
closer to 100% ?
My problem with the scenario that leads to 40% mortality at the final
dose, is that the animals are sacrificed for no purpose as the data is
insufficient to get an estimate of LD50. All we know is that the LD50 is
greater than Xdddd. However if Xdddd were greater than 50% we would have
our estimate and confidence interval. Better still if Xddd (the 4th dose)
had been greater than 50% we could dispense with the 5th dose and get an
estimate, with confidence intervals of the LD50. If the mortalities for
the 3rd and 4th doses exceeded 50% the LD50 will be even better
estimated. The animals would not have been sacrificed unneccesarily. Also
less animals would have been used to get better data.
I am more interested in the study design, than the LD50 software, though
I would also welcome information on the statistical software.
Please understand that I am working under the assumption that the LD50
studies have to be done (and the animals have to be sacrificed). I want
to be able to advise on improving the study design to get better data,
and if we can sacrifice less of the animals all the better.
If such software does not exist, and if there is an express need for it,
I will finalise and market the software I am developing for my client,
I therefore would appreciate whatever information I can get.