Replace an animal in 4×4 Latin square design?
November 25, 2008 1:37 pm ScienceHow do others handle the issue of loosing a telemetry animal in a 4×4 Latin square design. For example if you loose one animal after two doses and still have two additional dosing periods.
What is the current practice in the industry when this occurs? One suggestion was to substitute in an animal and simply finish out the study in the normal 4 doses. Another was to substitute in an animal and give 4 more doses. Yet another is to continue on with missing cells. What is your opinion or experience?
I would also like to get the perspective on how you feel these options might impact the statistics.
November 25th, 2008 at 11:23 pm
Of the options presented, the second option, replacing the animal and administering all 4 doses is the most applicable from a scientific as well as a statistical analysis perspective. The goal is to have 4 animals receive all doses in a random order with no carryover effect between doses. The second option does introduce the variability of time as one of the animals will continue an additional week. But the power obtained by having data for 4 animals at each analysis data point positively outweighs the negative effect of time. Missing data points decrease the power and thus, the can reduce the sensitivity of the study. The power of the 4X4 comes in the fact that each animal is it’s own control, thus reducing inter-animal variability. The first option is not acceptable as it violates the Latin square study design. Option 3 significantly reduces the statistical power, and thus the sensitivity of the study is reduced. Over the past 8 years my experience and consult with my favorite statastician, is that replacement of the animal provided data that had little alteration to the standard error and maintained the statistical sensitivity to detect small yet biologically relevant changes in electrocardiographic (eg. QTc changes of 5-8 msec p < 0.05) and hemodyanamic parameters (eg. MBP changes of 3-5 mmHg p < 0.05). A fourth option is to stop the study, replace the animal, provide a sufficient washout period for all animals previously dosed, and then perform the study. In essence, the study is restarted to ensure that 4 animals are present for all dosing events.
If the study design were a double 4 X 4 (also referred to as a 4X8), then the latter option would have less of an impact on the statistical power of the analysis as there would still be 7 animals to provide confidence and statistical power in the interpretation and data analysis.