Comparison of methods for analysis of functional observation battery (FOB) data

Different methods for the analysis of behavioral observation data were compared to evaluate how the interpretation of a data set may depend on the analysis method employed.

Three methods of analysis were used to evaluate the same four sets of rodent behavioral FOB data: (1) evaluation by a trained behavioral toxicologist (ToxRev); (2) Kruskal-Wallis statistical analysis of variance for nonparametric data followed by Wilcoxon pairwise tests (KW) and (3) Cochran-Mantel-Haenszel statistical analysis of variance for nonparametric data followed by Wilcoxon pairwise tests (CMH). The FOB consisted of 9 behavioral, 10 neurologic and 7 autonomic parameters that were evaluated following administration of either vehicle or diazepam (1 or 4 mg/kg; 1, 2 or 5 mg/kg) to male and female rats. The chosen data sets were labeled A, B, C and D. The outcomes of the three analysis methods were compared to identify similarities and differences.

ToxRev, KW and CMH analyses were in agreement in determining the no-effect level (NEL) for each data set. All methods were also in agreement calling the fewest FOB parameters in data set C, and correctly identifying the most effects in the behavior functional domain in each data set. The 3 methods were also in agreement in correctly not calling parameters such as convulsion. No single analysis method stood out as remarkably more permissive in identifying effects of diazepam on FOB parameters, although CMH appeared to be the most conservative method, identifying the fewest effects across all data sets.

Factors contributing to these patterns of outcome are discussed, including variability within and between dose groups. ToxRev, KW and CMH are all viable methods for evaluating FOB-type data, however minor differences in a study's outcome using these analyses may be dependent upon the method selected.