Animal Models and Experimental Design Flashcards
Study design: Latin sqaure
complex study design
Testing more than two conditions - multiple groups (similar in size, age, number ect) is treated with different treatments (A/B/C - one being no treatment = control) over multiple time periods:
Treatment periods seperated by a wash-out period
Treatment: A/B/C
Test: 1 2 3 Group 1: A B C Group 2: C A B Group 3: B C A
Time consuming. the treatments must not make irreversible changes
Control: no treatment can be one of the treatments
Study design: Completely randomized
How to make two groups, what is the risk and advantage?
flip a coin - Dividing the animals (an individual = unit of study) in the experimental group and the control group by randomization.
Risk: all the small/big animals are together in one group = might have influence on the results!
But we need to ensure that the groups are at same size ! so it is not possible to just flip a coin…
Study design: Randomized block - heavy pigs and randomized dividing
Dividing the heaviest animal in one group, which is again divided into an experimental group and a control group by randomization.
The rest of the animals are divided in to the two group by randomization.
Other factors could be the main focus, this example is focused on size (weight) being an important factor = not a bias anymore
Fidelity
A model that is very similar to the thing being modelled = it looks a lot like the thing the model it mimics
Discrimination
The does not look like what the model mimics but only specific aspects of the model being modelled. animal models often has high discrimination = specific aspects of the model mimics human
External validity - how well does the model translate to the species it is modelling?
Construct validity
The extent to which both the animal model and the human phenomenon can be explained – in terms of origin, underlaying mechanism and function – by the same theory
The cause of the disease is the same both in animal and human.
External validity - how well does the model translate to the species it is modelling?
Face validity
Similarity in appearance – what is observed in the model resembles that observed in the modelled phenomenon. The phenotype of the model is the same as what is observed in what is being modelled.
FIV in cats has a high face validity to HIV in humans
External validity - how well does the model translate to the species it is modelling?
Predictive validity
performance in the test predicts performance in the condition being modelled = A measure of how much a drug has the same effect in human and models being tested
Relationship between construct and face validity
Often goes hand in hand –> both high or both low
Reproducibility crisis:
- external validity
- internal validity
Many studies cannot be replicated!
External validity: how well does the animal model resemble the disease we are trying to model in various aspects.
Internal validity: does you trust the study? Experiment design randomization, blinding, statistics. GLP, PREPARE guidelines, sufficient reporting.
To what extent does the study ensembles as trustworthy- is the study well argued in the descussion. Is it possible to replicate the study?
External validity: is the results also seen in reality = in human?
Effect on sample size: Variation
Variation: the greater the expected variation, the larger sample size
Small variation: inbred mice of same age, weight and sex.
Effect on sample size - Effect size
Effect on sample size: the smaller effect size of interest, the larger sample size
use literature or a small pilot study to determine the effect size
Effect on sample size: level of significance
Level of significance (0,05 = 95%): the lower the risk of false positive (type I error) (fx. 99%), the larger the sample size
Effect on sample size: statistical power
Statistical power (0,80 = 80%): the lower the risk of false negative (type II error), the larger the sample size (fx 0,9 instead of 0,8)
If a false negative result isn’t beneficial, you must increase the statistical power = higher sample size
What is bias in an animal experiment?
Bias is every possible thing with the opportunity of affecting the study – subconsciously or done on purpose. Resulting in unreliable research = we have a risk of finding a different between the group that has nothing to do with the thing being tested!
Bias: subconsciously factors
Knowing which group is the treated and which is the control – altering the method used for sampling data (measuring tumor size and squeezing harder on the treated tumors)
prevention: blinded, random data collection
Bias: know factors
Only publishing the positive results and not publishing the results that weren’t expected.
Bias prevention:
- Randomization group dividing
- Blinding datacollection
Randomization and blinding are the main actions done to avoid bias.
- Randomization: the animals should be assessed in a random order. The animals (if not same size/age/sex) should be randomly divided into the groups.
o OBS if weight is important the animals should be either divided randomly in blocks avoiding all the heavy animals in one group OR by matching them in pairs by weight and randomly place the animals in control/experimental group.
- Blinding: the person collecting the data does not know which animal is treated and which is not, ideal nor does the caretakers or anyone else handling the animals.
What is an experimental unit - example of different sizes
The smallest element in the study for which a conclusion can be drawn –> often it is an individual animal but, in some cases, it is a small group of animals.
Individual: If a mouse is treated with an anti-cancer drug it is possible to conclude on the difference in the tumor size on the treated mice compared with the mice in the control group (non-treated)
Group: if we want to measure how much water the animals drink or food they eat – they are caged in groups of three and it is not possible to make the individual measurements. = the experimental unit is the cage = a group of three animals
Calculation of sample size prior to conducting an animal study - why ?
Just using a random number of animals without considering the sample size might result in a waste of animals and failure to interpret the results correctly.
A computer model can be used: www.powerandsamplesize.com
You must have determined the following:
Level of significance – accepted risk of false positive (type I error), often 0,05
Statistical power – accepted risk of false negative (type II error): often 0,8
Variation: literature or pilot study
The smallest effect size of interest: literature
Complex study-design:
case-control vs a latin square!
Case-control: simple set-up: a treated group is compared to a non-treated control group = does the tested drug work and how well does it work.
Latin-square (complex cross-over design): multiple treatments are tested on the same animal. And the animal is also used as control
A: treatment 1, B: treatment 2 and C: control group
Multiple groups tested:
A – B – C
C – A – B
B – C – A
The time frame is similar for all treatments. Separated by a wash-up period: the tested substance must be out of the body before the animals can be treated with a new drug.
The animals are used as their own control for two different types of treatments. = fewer animals and elimination bias as all animals are treated and used as control.
When do you need expert help? statistics!
Are you looking at multiple factors with possible impact on the result (fx. age of the animal, sex of the animal, weight of the animal) it calls for expert help as the statistics will be quite complicated (multivariate analysis).
The more complicated the study design, the more complicated the statistics – you might even need help to calculate the sample sizes needed.
You must know:
What the aim is with the study – is it to find out if a drugs works/not works? if a given factor have an effect on the drug effect?
Do you have one group or multiple different groups that you want to test the drug on? (age, sex, size)
Is there a possibility of multifactorial impact?
How can pilot studies help us when planning our experiments? Can we use a pilot study to estimate an effect size?
If no literature exists similar to your study, if you are using a newly developed model with no data registered, it is beneficial to do a pilot study, indicating the possible variation in data = an important factor in calculation your sample size (the bigger the variation, the larger sample size is needed).
If you have no experience with the methods you need to use – it could be a specific procedure/operation, it could be an advantage to conduct a pilot study, to eliminate all factors possible to prevent! Is the study design even possible to conduct at all or is it impossible to do the operations within the given time frame or with the given materials available0?
It is not possible to estimate an effect size based on a pilot study, if so, you have done the actual study! Use the literature.
Explain the principles of GLP in experimental animal research
GLP: ensures the consistency, reliability, uniformity and quality of chemical non-clinical safety tests. Aims to ensure consistency and reproducibility!
Following the principles of GLP you achieve research of quality with full transparency and traceability.
SOPs is essential as is documentation: examples:
- Animal room preparation
- Animal care
- Laboratory tests
- Histopatology
- Handling dead and moribund animals including the animals dying during the study
- Necropsy and postmortem examinations
Explain key issues to be reported when publishing experimental animal studies.
Use the specific guidelines from the journal you wish to publish in; abstract, background, objectives, ethical statement, housing and husbandry, animal care and monitoring ect. Should be written .
Use the ARRIVE guideline for key issues in reporting animal research:
1. Study design; groups (control group), experimental unit
2. Sample size including how the sample size was decided
3. Inclusion and exclusion criteria
4. Randomization – and how it was done
5. Blinding – who knew the allocation of the animals throughout the study
6. Outcome measures – what was assessed
7. Statistical methods, including software
8. Experimental animals: in depth description
9. Experimental procedure; what, when, where and why
10. Results – summary/descriptive statistics
Validition criteria (face-, construct- and predictive validity) in a spountaneous NOD mouse - Non-obese diabetic NOG mouse- spontaneous Diabetic 1
High face validity as the diabetic clinical sign mimics the symptoms in humans
High to low predictive validity when treated with insulin = alleviates diabetes as in humans (the animal model shows what we will se in human). but no cure has ever been found
Moderate construct validity: Autoimmune destruction of beta-cells in the pancreas as in humans. But the mouse needs to be inbred with a specific gene
Validition criteria (face-, construct- and predictive validity) in a induced animal model (diabetic 1 rat with surgical removed pancreas)
LOW Face validity: The rat will get the same symptoms (hyperglycemia, weight loss ect) BUT will also develop multiple other symptoms due to the lack of the hole pancreas and not just the beta-cells
LOW Construct validity: hyperglycemia is due to the lack of beta-cells (pancreas) in the rat, but in humans it is often due to the destruction of the betacells (autoimmne) which is not the same as seen in the rat with surgical removed pancreas
LOW to HIGH Predictive validity:
HIGH: if the rat is given insulin the clinical signs of diabetes will disappear as in humans. and LOW if you are to test a diet targeting the reason for developing diabetes.
ARRIVE Guidelines 1-10
Minimum included in the manuscript in animal research
- Study design
- Sample size
- Inclusion and exclusion criteria
- Randomisation
- Blinding/masking
- Outcome measures
- Statistical methods
- Experimental animals
- Exprimental procedures
- Results
