Feedback
on the VOHC system and suggestions for improvements in the VOHC
process are welcome. Contact VOHC.
Statistical
Analysis Information
Statistical
Presentations for Inclusion in VOHC Submissions
One of the
most common reasons for concerns being raised about a submission
during review by the VOHC Council and Statistical Consultant
is failure to use and adequately describe appropriate study
design and analysis procedures to demonstrate that the VOHC
statistical hurdles have been met.
Use of inappropriate study design and/or analysis criteria can
be an expensive mistake if the trial has to be repeated. VOHC
strongly suggests pre-trial consultation with a qualified statistician.
It is the responsibility of the sponsor to ensure that appropriate
methods are used.
Requirements for the statistical methods and results sections
in the submission:
1. Provide a clear description of the statistical analysis methodology
employed, with justification for choice of the specific tests
used. The tests used must be appropriate for the type(s) of
data (categorical, ordinal [ranks] or continuous [actual metric])
analyzed.
2. Note that there is a risk when using a small sample size
(fewer than 30 animals per group, as is typical in VOHC trials)
that the data may not be amenable to parametric analysis. Again,
for small samples (as just defined), if the group sample variances
are greatly disparate (ratio > 2:1), assume the data are
not treatable by a parametric approach and apply a non-parametric
test (such as Wilcoxon Rank-Sum test) instead of the t-test.
Note: there are rank tests for several groups, even when stratified
or blocked, to replace ANOVA when necessary.
3. Use a two-tail test when determining p-level.
4. Randomization of subjects to treatment groups is required.
The randomization method must be described, and can be used
to your advantage. For example: if subjects
are assigned to treatments within a blocking structure (to reduce
variability), the blocks can be included in the analyses to
reduce the size of the residual error term (i.e.: the denominator
of the F-Test).
5. Tabulation of the results of the statistical analyses is
required in order to assess the validity of the claim made in
the submission. For example: when analyses of variance are used,
ANOVA tables are to be presented, including mean squares, F-
tests and p-levels. When multi-way layouts are used (for example
when various centers are used) the interaction of treatments
with the other levels used must be evaluated to insure additivity
(that the treatments compared are consistent across the other
levels employed in the design).
6. Where several groups are compared, adjustment for type-error
must be applied, i.e.; for multiple comparisons.
7. In trials using a cross-over design, effects for carry-over
(sequence effect) and period must be included and evaluated
in the applicable ANOVA tables. In the event of a statistically
significant sequence effect in a 2 period cross-over, only the
first period data can then be used in a parallel analysis –
the second period data must be ignored. In cases of cross-over
designs with 3 or more periods, statistical protocols are available
to adjust for carry-over effects detected in the analysis.
8. A descriptive summary table with sample sizes, means with
standard deviation and standard error, percentages differences
in means of the analyzed variable (e.g. plaque or calculus [tartar])
and where applicable, p-levels for the statistically significant
findings, must be included.
9. An electronic copy of the raw data in spreadsheet format
(including a clear explanation of the data formatting) is to
be included. Including the individual tooth data in the main
submission document is not required – in the main submission
document, include only the data and analysis information required
in items 1-8 above.