STAT 231 PROJECT

SPRING 2012


THE EFFECT OF SUBTHERAPUTIC ANTIBIOTIC USE ON MILK YIELD AND THE
AMOUNT OF CARRY OVER IN COWS MILK
































1. Define the objectives of the experiment
Ranchers and farmers have been feeding antibiotics to the animals we eat since they discovered
decades ago that small doses of antibiotics administered daily would increase the yield of milk
and meat. Stuart B. Levy, M.D., who studied the subject for years, estimates that there are 15-17
million pounds of antibiotics used sub-therapeutically in the United States each year.

There is evidence to suggest that the use of sub-therapeutic antibiotics in animals can pose a health
risk to humans. If a group of animals are treated with a certain antibiotic over time, the bacteria
living in those animals will become resistant to that drug.

Concern about the increasing level of drug-resistant bacteria has led to the banning of sub-
therapeutic use of antibiotics in animals products in countries such as Canada and the European
Union. In the United States, however, the use is still legal. The World Health Organization is
concerned about the use of antibiotics in the animals and its associated health risk to humans. It
has therefore advocated for a decline use of antibiotics in animals.

The aim of this project is to study the effect of sub-therapeutic oxytetracycline supplementation on
milk yield in dairy cows in order to test the farmers claim that administering small doses of
antibiotics daily would increase milk yield.
The quantity of milk produced will be measured at specific times. This will help inform the
farmers about the maximum milk production after administering of antibiotics. The quantity of
antibiotics in milk produced will be measured in order to test the hypothesis that antibiotics
administered to cows will be carried over into milk consumed by humans.

2. Identify the sources of variation that will be accounted for in the experiment
(a) Experimental units (EUs) Cows
(b) Observational units (OUs) Milk samples

(c) Treatment factors and their levels
Antibiotic 0 (Control group); 250 mg/100lbs; 500mg/100 lbs (Maximum legal)
http://pubs.ext.vt.edu/400/400-008/Table_1.html

Cows: 1 to 15. Highly inbred lines (Same father, similar mothers), in similar conditions (Age,
weight, number of lactations)

Time after starting the treatment: 0, 5, 10, 15, 20, 25 weeks

Indicate which factors you are considering as fixed or random effects
Cow - Random effect
Antibody - Fixed effect
Time - Fixed effect

3. Specify the measurements to be made, the experimental procedure, and any anticipated
difficulties.
Fifteen cows with same age, weight and breed will be used for the experiment. For all the cows we
will use the same diet, varying only the subcutaneous injection (Excipient for the control group,
excipient + antibiotic 50/50 or antibiotic). All fifteen cows will have their ears tagged, and then
using random number generator, a set of 5 cows will be selected and put into assigned pen. To
avoid the effects of confounding variables like amount of food intake by cows, which will
increase milk production, cows will be given equal amount of food for 2 weeks before starting
the experiment. Cows will be milked twice a day, 8 am and 5 pm on selected days. Milk yield
will be recorded at each milking. The difficulty of using cows by the same weight and/or breed is
anticipated.

Antibiotic analyses
The concentration of the antibiotic in milk will be measured with a ELISA assay (Detection limits
of ppts)

Carryover (CO) calculation
The antibiotic CO in milk will be calculated at plateau condition.
Statistical analyses
We will measure over time: 1. Food intake (kg/day) 2. Antibiotic excretion in milk (mg) 3.
Antibiotic concentration in milk (ug/L) 4. CO in milk (% intake) 4. Milk production (l/day)


Control(ug/L) Concentration 1(ug/L) Concentration 2 (ug/L)
14 11 15
10 6 8
3 4 5
2 7 13
12 9 1


4. Specify the statistical model for the experiment
a completely randomized design with the following model:
Milk yield_ijk = mean + alpha_i + b(alpha)_j + time_k + (alpha*time)_ik + residual_ijk

alpha_i = Fixed effect of treatment
time_k = Fixed effect of time after treatment
(alpha*time)_ik = Sampling interaction treatment*time
b(alpha)_j = Random effect of individual cow within treatment
residual_ijk = Experimental error




5. Outline the analyses that you would perform

We expect all the variables to stabilize over time so we will take the means when they are constant
and use them for our analysis (The same number of time points for each treatment and variable).
If the Time variable is significant the milk yield/AB concentration wont be in the plateau phase
yet.

The Trt x Time variable will tell us if during what we consider plateau phase are interactions
between treatment and time.

With contrasts and Dunnett method (Control = Treatment 1) we can see which treatment is better
(Better milk yield).

With orthogonal contrasts (-1 0 1 ; 1 -2 1) for the treatments we will study the response, if it is
linear or quadratic.

Plots:
- Time vs. AB in milk for the treatments --> Calculate the carry over in the plateau, see
interactions.
- Time vs. milk yield for the treatments --> Calculate the milk yield in the plateau, see
interactions.

ANOVA analyses:

Treatment
Means
SE Significance Orthogonal
contrast
1 2 3 F.E.
Trt
F.E.
Time
F.E. Trt
x Time
R.E.
Cow
Linear Quadratic
Food intake
(kg/day)

AB concentration
in milk (ug/l)

AB excretion
Carry over in milk
(%)

Milk yield (kg/day)


For the experiment we also could have used different farms, adding them as a blocking variable
and we could extrapolate the results to all the cattle population.

We expected to get a small AB excretion and a yield increment with the antibiotic use; if the AB
excretion is significant greater than 0 the antibiotic use wouldnt be allowed; if the AB excretion
is not significant greater than 0 more studies would be necessary to prove there is or not
relationship between the AB use and the increase of AB-resistant bacteria.

6. Provide a power analysis for your study by determining the number of observations that would
need to be taken in order to detect a pre-specified effect. You can frame your power analysis in
terms of standard deviation units, but you must give some estimate of the magnitude of the
expected variance components in your study. Explain how you arrived at the variance estimates
used in the power analysis and the target for the alternative hypothesis.

Mean milk (Daily)
Mean = 31.56 kg SD = 6.45 kg (Multiple-Study Analysis of the Effect of Live
Yeast on Milk Yield, Milk Component Content and Yield, and Feed Efficiency; The Professional Animal Scientist 2
6 ( 2010 ):661666 ; M. B. de Ondarza , C. J. Sniffen L. Dussert , E. Chevaux J. Sullivan PAS, and N. Walker.)

Mean = 56.7 kg SD = 7.4 kg (Association of milk yield and body condition score indices with the
commencement of luteal activity after parturition in high producing dairy cows. Iranian Journal of Veterinary
Research, Shiraz University, Vol. 12, No. 3, Ser. No. 36, 2011 . A. Tamadon, M. Kafi, M. Saeb and M. Ghavami.)

Mean = 32.16 kg SD = 4.9 kg (Iodine and selenium carry over in milk and cheese in dairy
cows: effect of diet supplementation and milk yield; Animal (2010), 4:1, pp 147155; M. Moschini, M. Battaglia, G.
M. Beone, G. Piva and F. Masoero)

Mean = 7073 SD = 1425 (Not units specified) (Dairy Production 342-450 Milk Yield & Composition)

Well use mean = 50 L SD = 3,25 L for our data (As we are doing the experiment in a single farm
the SD is expected to be less than in the studies above, the results only could be applied to the
farm itself).


In order to find the power of the test, we need to find the non-centrality parameter

, which is the
parameter that the F distribution has when the null hypothesis is true. The function of the non-
centrality parameter is

u=

t
=
D
o
r
2t
(the reference is our textbook), where t is the number
of treatments, D is the biologically difference that we think is significant,

o is the standard
deviation, r is the number of replicates. So, to apply this formula, we need to have an estimate of
the standard deviation for the variables that we are interested in, and since we are interested in
the effect of antibiotics dosage on milk production, and on the effect of antibiotic dosage on
antibiotic concentration in milk. After calculating the ANOVA table, MSE was found to be
3.25L/day, and we would consider a difference of 10L/day as a biological significant difference.
So applying the formula, we would have

v
1
= (t 1) = (31) = 2,

v
2
= r(t 1) = 2r, and trying
r=5,

u =
10
3.25
5
(2)(3)
=2.8, and looking at the power table and using

o = 0.05, we find the
power to be

~
96.5% which is good.

Now to find the power analysis for the concentration of antibiotic in milk, we need to compare the
two treatment groups in the amount of antibiotic in the milk, so 5 one liter samples of milk are
taken randomly from each group (1 L from each cow in the three treatment groups) and the
amount of antibiotic is measured using the technique of ELISA, and a difference of 5mg/L is
considered biologically significant, the standard deviation for the first AB group is 1.7mg/L, and
the standard deviation for the second AB group is 1.9mg/L, and the formula for calculating the
sample size needed for a 90% power with

o = 0.05 is

r =
2(10.5)S
pooled
2
d
2
, where 10.5 is a
multiplier which depends on the significance lever and power and comes from the normal
distribution (the reference is Bland M. An introduction to medical statistics. 3
rd
ed. Oxford
University Press, 2000), and since we are assuming that the variance is equal in the two groups,
the pooled variance is

s
1
2
(n
1
1) + s
2
2
(n
2
1)
n
1
+ n
2
2
=

(1.7)
2
(4) + (1.9)
2
(4)
8
= 3.25, and

r =
2(10.5)(3.25)
25

=2.73 or 3 cows, so in our experiment we should have more than 90% power since we have
selected 5 cows in each group


CODEBOOK

Variable name Levels Data type
AB 1 Excipient
2 Excipient + 250mg/100lb
3 500mg/100lb

COW 1 to 15 Number of the cow (Randomized)

TIME 1 0 weeks
2 2 weeks
3 4 weeks
4 6 weeks
5 8 weeks
6 10 weeks