Download presentation
Presentation is loading. Please wait.
Published byAlisha Martha Blake Modified over 9 years ago
1
Tips for a Sensible Animal Justification Reid D. Landes Disclaimer: the opinions and thoughts contained herein are not necessarily those of UAMS or any of its affiliates (including the IACUC).
2
Outline Reviewers’ questions Answers for –Non-statistical justifications –Statistical justifications Elements of a (normal) power analysis Pilot studies
3
Reviewers ask What are you measuring? What are the groups? What do you want to learn? Do the animal numbers match your need?
4
Non-statistical Justifications Reviewer: Measuring what? Learning what? –PI: “From the animals? Nothing.” Typical protocols with these answers –Breeding protocols –Materials protocols
5
Comment #1 Breeding protocols: How many animals… –are needed for future research objectives and maintenance of colony? –will not be used for any purpose? Materials protocols: How much material… –is needed & why that amount? –does one animal provide?
6
Tips for Statistically Justifying Your Animal Use Answers to Reviewers’ Questions for Statistical Justifications
7
Comment #2 Many protocols have more than one experiment Some differ in their design or size Some differ in their outcomes Usually power analyses are needed for each unique design-size-outcome combination
8
What are you measuring? Define your outcome measures Give their units Several outcomes in one experiment? –Which is most important? –Which is most variable? –On what outcome is the power analysis based?
9
Outcome Measures Examples Cancer study: “Tumor burden” –Number of cancer cells per l. Or tumors. –Volume of largest tumor. Or all tumors. ALS Study: Several measures –Time on a rotating rod (seconds) –Survival time (days; power based on this) –Oxidative damage markers (more definition needed here)
10
What are the groups? Describe the “experiment design” List the groups (and how they are formed) –The Methods section contains what happens to animals in the various groups
11
Experiment Design Example “Mice within a strain equally randomized among the 5 drug-dose combinations” 6 mice / strain-drug-dose combo X 10 strain-drug-dose combos = 60 mice StrainWTKO Drug--AB AB Dose0LoHiLoHi0LoHiLoHi
12
Experiment Design Example (cont) PO Define strains, drugs, and drugs’ low & high doses outside of the table If sample size differs among Strain-Dose-Drug combos, then add a new row StrainWTKO Drug--AB AB Dose0LoHiLoHi0LoHiLoHi
13
What do you want to learn? “What are your research hypotheses?” Identify comparisons are of interest Sometimes, comparisons are ordered –Primary, secondary, tertiary –Base power analyses on primary ones Often possible comparisons exceed comparisons of interest
14
Comparisons of Interest Example Possible comparisons: 10 C 2 = 45 pairs Primary: Within Drug, compare strains (2) Secondary: Within Strain and Drug, compare each dose to control (2x2x2=8) StrainWTKO Drug--AB AB Dose0LoHiLoHi0LoHiLoHi
15
Do the animal numbers match your need? PO Provide a well-described power analysis Attributes of “well-described” 1.Describes statistical analyses that are consistent with experiment design 2.Describes a power analysis that is consistent with the statistical analysis can be reproduced 3.Presents and justifies the parameters and assumptions required in the power analysis
16
Comment on Attribute #1 Must you describe the statistical analysis? –For me? Not required, but definitely appreciated Often, a well-described power analysis sufficiently describes the statistical analysis –For simpler experiments If a statistical / power analysis is beyond investigators’ capabilities, consult / collaborate with a statistician
17
Comment on Attribute #2 Must have reproducible power analysis? –I’m looking for this Sample sizes based on heuristics (trial & error, “standard” for a particular field, etc.) are generally turned back
18
Comment on Attribute #3 Parameters of a power analysis –Significance level, Power, Sample size, Effect size, …and others depending on statistical analysis –All are required to reproduce the power analysis Particular parameter = F (other parameters) Often have to assume / estimate one or more of these –Explanation or references for estimates should be given
19
Elements of a Power Analysis Normal-distribution based Need to identify (or calculate) the previous 4 parameters, plus –parameters specific to the experiment (e.g., number of groups, factors and levels) –statistical analysis (e.g., 1-way, 2-way ANOVA, two-sample t -test, paired t -test, etc.) –Possibly comparisons of interest
20
Effect Size Comment What about a SD? Or difference to detect? –Encompassed in an effect size When comparing two groups’ means an effect size = difference in means within-group SD difference = 5, SD = 10 effect size = ½ difference = ¾, SD = ¼ effect size = 3
21
Example of a Power Analysis Strains-Drugs-Doses Assuming –Y, outcome of interest, is normal –All Y s are mutually independent –k =10 groups are mutually independent –“one-way” ANOVA Significance level=.05, power=.80, and…
22
and n =6 … forces effect size= –1.17 (not controlling for multiple tests) –1.30 (controlling for 2 tests with Bonferroni’s method) and effect size=1.0 … forces n = –8 (not controlling for multiple tests) –10 (controlling for 2 tests with Bonferroni’s method)
23
Example Write Up “With 6 mice per strain-drug-dose combination, we can detect a difference of 1.3 SDs for the primary comparisons of interest with at least.80 power using a.05 level two-sided t -test, adjusted for 2 tests with Bonferroni’s method and conducted within an one-way ANOVA context. SDs of survival times have been estimated to be as high as 10 days [refs]. A difference of 2 weeks was found between Drug A and a control in [ref]; hence, providing evidence that a difference of 13 days is not unreasonable to expect.”
24
Pilot Studies PO Reviewers still ask the same questions Animal measurements informing scientific objectives call for justification Often, statistical justifications are possible and sensible for pilot studies When not possible, decision rules should be in place
25
Objectives of Pilot studies Estimating a statistical parameter; e.g., SD, mean, proportion, etc. –Report widths of 95% Confidence Intervals for requested n Determining feasibility –Feasibility defined in terms of measurement –i.e., define a rule by which feasibility is determined
26
Pilot Study Example #1 Estimating a statistical parameter Estimate SD of Y with n =10 animals “Assuming Y is normal in distribution, we will have 95% confidence that the true SD is no more (less) than 1.83 (0.68) times the estimated SD.”
27
Pilot Study Example #2 Determining Feasibility Want to know if a new procedure is reliable (i.e., feasible) in producing a desired effect. Requesting 6 animals Investigator defined rule: “The procedure will be deemed ‘feasible’ if the effect is produced in all 6 animals. We will examine each animal in sequence stopping if the effect is not produced.”
28
Contact Information Reid Landes rdlandes@uams.edu 501-526-6714 COPH Room 3224 www.uams.edu/biostat/landes
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.