---start---
epidemiology
1/26/98

Dr Smith remarks that he enjoys 
etymoloy, evolution of words...
"I carried on until the bitter end..." 
doesn't make sense until you realize it means "I carried on until the 
better end" and it referred to the rope used on boats that would hang 
down and rot in the water, so you'd extend the rope past the bad part to 
get to the better end...

american english has retained a lot of archaic 
usages. "gotten" instead of "got"
"momentarily" - means "in a short 
while" to americans (and to shakespeare) but to Brits it means "for a 
short time." 

sodium - Na+ from latin natrium, from arabic word meaning 
headache. 
1805, an english PI was testing efficacy of "magnetotherapy" 
on syphillis patients or something. he tested wooden rods painted silver 
as a control - 4 out of 5 of those also claimed the treatment worked. he 
called these "placebo" which means "i will be giving pleasure"

leading 
in to talk about Clinical Trials:

in terms of the order in which one 
does things - first you must decide what kind of clinical trial are you 
going to do? there are many kinds. we'll only go over two. one, just to 
dismiss it - the uncontrolled clinical trial. this is a very common 
trial, actually. you take a sick patient, treat it, and see what happens. 
compare status before and after tx. that's fine if you have a disease 
whose untreated course is obvious eg bovine hypocalcemia leading to 
death. but many diseases have uncertain natural courses, or highly 
variable natural courses. you have no idea if what you see after tx is 
due to tx, or would have happened anyway. it's thought that about 80% of 
tx used on people have been tested only in this way. a huge amt of 
veterinary therapies have only been tested this way. so it is the bedrock 
of clinical practice, but we're not going to go over it b/c you never can 
be sure, even a little, that your changes are attributable to treatment. 
eg, canine ehrlichiosis - a tick borne rickettsial disease with a very 
severe acute phase from which most dogs recover irrespective of 
treatment. uncontrolled trial also called "before and after" trial. 

a 
randomized, controlled clinical trial:

summary:

you start with a group 
of patients - it's essentially a cohort study - all patients are defined 
by having a disease.

1. patients are allocated to either treatment or 
control group. please be aware that control doesn't mean "untreated." It 
could mean "on standard, accepted therapy" and then treatment group is 
getting "new, improved" therapy or something.
2. both groups are treated 
identically with the exception that the treatment group(s) receive(s) an 
intervention believed to be beneficial. all other parameters should be 
identical.
3. control group gets a placebo (an intervention designed to 
simulate the act of treatment, lacking beneficial components) if it isn't 
actually getting a standard treatement or whatever. 
4. differences (end 
points) which emerge between the two groups are attributed to the 
treatment. 

you follow the cohort, measuring differences that occur b/w 
treated and control groups...any differences in endpoints are generally 
regarded as being attributable to the treatment. 

after deciding what 
kind of trial to perform, you have to formulate a hypothesis. the 
hypothesis has some characteristics - namely, it is almost always 
expressed in the null form. you hypothesize that there will be no 
difference b/w the treatment and control groups wrt the defined 
endpoints. so hypothesis must include some statement about what the 
endpoint actually is. a good rule is to be as specific as possible. try, 
like in any good title, to encapsulate the relevant points. what's the 
disease? what are the endpoints? what are you interested in? and couch it 
in the null form. the real trick is selecting appropriate endpoints. you 
could measure a lot of things, but which are important clinically? well, 
sometimes what you think is significant isn't the same as waht the 
patient/owner thinks is significant. you have to make judgements about 
this. you have to also decide how to measure a canonical endpoint - the 
manner in which you measure this may be to kill all the animals - but 
owned animals can't be used in that study, then. so your endpoint has to 
be able to be reached in reasonable manner - you need ease, compliance, 
money, etc. some endpoints might take 10 years to reach, that takes a lot 
of time and money. some diseases result in higher death rate but the 
deaths occur decades after the onset of disease. you can't get enough 
money to track the animals long enough to use death as an endpoint. you 
use intermediate endpoints - physiological measures that appear soon, but 
predict later endpoints, that you're more interested in. 

types of end 
points - 
survival
quality of life
complications
productivity/economic 
benefit - farm animal medicine - productivity may be more important than 
disease status.

step three - sample size determination. you've decided 
what trial to do, you've written your hypothesis and chosen reasonable, 
significant endpoints. now you have to decide how many cases you need to 
start with in order to test your hypothesis. we almost never,in a 
clinical trial, have access to the whole population you want to study. 
so, you sample the population - therefore you run the risk of choosing a 
nonrepresentative sample or introducing sampling error that materially 
affects the results. 

									"the truth"
your conclusion		treatments 
differ	treatments don't differ

treatments differ	correct				type I error

treatments don't	type II error		correct

type I error = you conclude 
there is a difference when there is not - you falsely reject the null 
hypothesis.
type II error = you conclude there is no difference when 
there is one - you falsely accept the null hypothesis.

every year, 
students ask - why do you have to mention type I and type II errors? 
well, when you do a clinical trial, there is always the risk that your 
conclusion will be wrong, due to sampling error. we want to minimize that 
risk. we can do this by calculating a sample size such that we define the 
risk of type I or type II error. by defining in advance an acceptable 
level of risk for type I or type II error, you can calculate sample size.


probability of making Type I error is the alpha level
probability of 
making type II error is the beta level

it's conventional to decide in 
advance what alpha and beta levels you will accept before starting. 
usually you accept a 5% risk of making type I and 20% risk of making type 
II error, so alpha = 5%, beta = 20%. now you use a formula to figure out 
how many cases you need. you set these levels for that purpose. then at 
the end, you can state your known risk of type I or type II error. 


note: power = 1 - beta
you have to use beta as a proportion, not a 
percent, though. this is the probability of detecting a difference when a 
difference actually exists - so with beta = 20 you have power = 80% = the 
probability of finding a real difference when it actually exists.

also - 
decide how big a difference b/w treated and control groups you want to 
detect. if you're interested in survival of animals, and you get a 1% 
difference b/w treated and nontreated groups, you probably won't think it 
is significant. when does it become significant? you have to decide. it's 
a judgement call, really. 

thing to remember - you have to know three 
things in advance: alpha, beta, and magnitude of detectable difference. 
once you know all three of those, and you use your own judgement because 
there are no rules for them, then you can determine the sample size. the 
higher the number of cases, the better off you are. so as alpha, beta, 
and detectable difference get smaller, your number of cases gets larger. 


case definition:
what patients will you include in the clinical trial? 

1. it may be difficult to define a set of signs that will include all 
true cases and exclude similar but unrelated conditions
2. few cases show 
the complete range of signs. therefore, as you increase the number of 
required signs, to exclude noncases, you also exclude a larger and larger 
numbers of true cases.

example - we're familiar with the idea that lymes 
dz in people is signalled by bullseye rash. but less than 50% of people 
in PA with lymes dz remember such a thing. 

so case definition is a 
problem. you have to be very specific about how to define your case, who 
to accept into the trial. 

allocation of patients to the treatment and 
control groups: this is the really important thing. you have to do this 
without bias, do it by randomization.

the purpose of randomization is to 
achieve an equal distribution of all factors related to prognosis among 
the groups. if you really really think a therapy is going to work, it's 
very simple to assign all the animals you think will recover anyway into 
the treatment group. people do this by accident all the time. so you have 
to randomize.

1. allocation not influenced by researcher's preference
2. 
allocation not influenced by allocation of any other patient
3. it is 
expected that if patients have been randomly allocated, demographic (age, 
breed, gender, etc) and known prognostic factors will be the same in all 
groups. this can and should be checked. make a list of prognostic 
factors, known and uncertain, and see if they are equally represented in 
both groups. it's possible if for example you are convinced that gender 
is very important, you can fix that factor so there are equal numbers of 
either gender in both groups - block randomization, or stratification.


difficulties with clinical trials:

1. migration bias: occurs with the 
patients that leave a study are systematically different from those that 
remain. sick animals may be withdrawn - particularly unpleasant 
treatments may cause owners to withdraw their animals.

2. 
non-compliance: there are many reasons for this, including forgetfulness, 
disappointment with results, realization that animal is in placebo group 
- might cause owner to go out and buy alternative treatment. then what? 
you keep noncompliers in groups to which they were assigned - even if 
they effectively switch groups. clinical trials are based on "intent to 
treat". so, usually the data for noncompliant animals/owners is left in. 
trials measure "effectiveness" - ie, trial meausres how well treatment 
works among those OFFERED the treatment. that's different from seeing if 
it works among those actually treated.

note: people on aspirin trials 
who were on placebo ditched the placebo and went out and bought aspirin. 
this was a form of noncompliance.

3. reporting bias on the part of the 
researcher or owner - both are likely to be influenced by their own 
desires or prejudices when evaluating the outcome. BLINDING - with 
blinding, neither owner nor  person measuring outcome knows which group 
the animal belongs to.

one author said he did a trial of a particular 
anthelmintic. animals were divided into two groups - one got a new drug, 
one got an old one. endpoint was a fecal egg count. author said "this is 
an objective measure, so there is no possible bias." but, fecal egg 
counts are not objective. you do 30 in a row, and the last 29 were 
negative, you're really less inclined to look carefully. if you know the 
slide is from the effective treatment group, you won't look as carefully. 
if you know it is from the less effective group, you look more carefully. 
so there is possible bias. you have to blind the person taking the 
measurments - render him/her unaware of origin of measurement.

questions 
to ask yourself when reading a clinical trial:

1. is the case definition 
sufficiently explicit to exclude similar conditions?
2. are patients 
allocated to treated and control groups without bias?
3. is treatment 
(and only that treatment) experienced by all in treated group and none in 
control group?
4. is the outcome assessed without regard to treatment 
status (blinded)?
5. how certain can we be that the outcome could not 
have resulted from chance alone (what test of significance was used)?


read the literature critically. there are tons of errors in the journals.

---break----


problems:

Q1.						disease
				+					-
t	+			A					B
e	-			
C					D
s
t


1. false positives? B
2. false negatives? C
3. sensitivity? 
TP/ TP + FN
4. specificity? TN/ TN + FP
5. PPV? A/A+B
6. apparent 
prevalence? A+B /N
7. true prevalence? A+C /N    (note: N=A+B+C+D)

Q2. 

1. what formula links prevanlence P and true incidence i? 
P = iD/(1+iD) 
where D = duration of disease.

2. what approximate version of this 
formula is often seen, and when can this approximation be used?

P 
approximates iD when incidence is very small.

Q3.

Let ci = cumulative 
incidence in exposed group
let co = cumulative incidence in unexposed 
group.

1. relative risk? RR = ci/co
2. absolute effect? AE = ci - co
3. 
attributable risk? same as AE
4. relative effect? RE = ci - co / co

Q4. 
write down the order of events in the following:

1. cohort study to 
measure relative risk

first, divide a disease free population into 
exposed and unexposed groups
second, monitor animals for years
third, 
note which animals develop disease and which do not
fourth: calculate ci 
and co, calculate relative risk ci/co

2. case control study to measure 
odds ratio

first, find all the cases you can in a given population

second, select non-cases (controls) from the same population using 
predetermined criteria.
third, separate animals into exposed group and 
unexposed group. 
fourth: calculate odds ratio (OR=AD/BC)

3. randomized, 
controlled clinical trial

1. patients are allocated to either treatment 
or control group. please be aware that control doesn't mean "untreated." 
It could mean "on standard, accepted therapy" and then treatment group is 
getting "new, improved" therapy or something.
2. both groups are treated 
identically with the exception that the treatment group(s) receive(s) an 
intervention believed to be beneficial. all other parameters should be 
identical.
3. control group gets a placebo (an intervention designed to 
simulate the act of treatment, lacking beneficial components) if it isn't 
actually getting a standard treatement or whatever. 
4. differences (end 
points) which emerge between the two groups are attributed to the 
treatment. 

Q5. define:

1. survival -  proportion of patients that 
survive a defined interval from some defined point in the course of 
disease.

2. remission - proportion of patients entering a phase in which 
dz is no longer detectable.

3. case fatality -  proportion of patients 
who die of the disease. really should have a time reference.

Q6. what is 
the purpose of randomization in a controlled clinical trial?

to achieve 
an equal distribution of all factors related to prognosis among the 
groups.

Q7. the incidence rate ratio, relative risk, and the odds ratio 
are all measures of effect 

1. what is meant by "effect" in this 
context?

an "effect" of a factor is the difference in disease occurence 
between two groups which differ with respect to that factor.

2. under 
what circumstances is relative risk a good estimate of the incidence rate 
ratio?

RR approximates IR (ie/io) when true incidences in each group are 
very small.

3. when is the odds ratio a good estimate of the incidence 
rate ratio?

OR almost always a good estimate of IR

4. when is 
cumulative incidence a good estimate of true incidence?

when true 
incidence is very small

Q8. what is the primcipal difference between a 
survival analysis and a life-table analysis?

life tables will account 
for those who drop out of the study, survival analyses will not.

Q9. 
what are the two essential properties of a confounder?

1. it must affect 
incidence, and 2. it must be associated with some other factor  


----break---

Outbreak Investigation:

She says she did a MS in 
epidemiology at some point, and that's why she's lecturing in 
epidemiology. also, chicken people use epidemiology all the time in their 
work. two handouts today.

Principles of outbreak investigation:

Definitions:

epidemic: unexpected increase in number of cases over what 
you consider to be normal. normal may be different for each disease - eg, 
LT, laryngotracheitis - expect 5 cases of LT/yr in PA - anything above 
that is an outbreak or epidemic. With avian influenza, AI, 1 case in PA 
is an epidemic. so you need to know normal to define an epidemic

endemic 
- the normal frequency of disease in a population

sporadic: infrequent 
and without discernible pattern - TB in birds is sporadic. we saw 3 cases 
in past 13 yrs. 

epidemic curve: before you start asking questions about 
your population, you make a graph plotting number of cases over time. 
time can be any time frame - hours, days, weeks, month. but you plot the 
number of cases vs time to figure out if you have an epidemic or an 
endemic rate. 

p 2 handout has charts. the epidemic curve, if it were 
for LT vs weeks, would be very concerning.

the endemic curve is not 
concerning if it is LT over the years - sometimes there are 5 cases/yr, 
sometimes a bit fewer, or whatever. if the chart were for AI, would be an 
epidemic curve.

sporadic curve shows one tiny peak.

after you've made 
your curve, if you have an outbreak/epidemic, you then describe it. is it 
hours, weeks, days, months? describe based on time, animal, and place. 
For animal - is it young animals? old animals? what kind of animals? 
particular species, breeds? what's the sex of the affected animals? place 
- is it in one flock, two flocks, one herd, ten herds, two farms, one 
farm, industrywide? is it in one pen on one farm? two pens on one farm? 


objectives (4):

1. halting progress of disease
2. determine reasons for 
outbreak
3. institute corrective measures
4. recommend procedures to 
reduce the risk of future outbreaks.

you might do this over a long time. 
you can't do it within a week all the time. with AI they started control 
measures last january, and we still have AI in PA. at first, they were 
depopulating immediately and burying birds - this was spreading disease. 
so it was an awful control measure. they had to change it. ultimately you 
want to institute logical corrective measures. 

epidemic - does not have 
to be infectious. Could be genetic, nutritional, vaccine-related, toxin 
exposure. there is no limit on the geographical extent - could be 
pandemic or could be on one farm. also, time can vary.

two types of 
epidemics, based on time factors:

propagated (slow spread) - these are 
the ones which have infectious agents
point or common source (rapid 
spread) - these are ones where there is a toxin, a contaminant, a 
nutritional source that causes a problem, that everyone is exposed to at 
once - like a food-borne illness in people.

propagated epidemics: agent 
spreads from animal to animal by direct contact, or by vectors - over a 
protracted time period - over weeks, months, or years. over several 
incubation periods. when dealing with this kind of outbreak, you want to 
get answers right away, but you do have some time to do it. with point 
source outbreaks, it's more of a rush to figure out. so first, plot cases 
over time like we said. this will be over weeks/months for propagated 
epidemic. see figure 12.1 in second handout. remember, time, place, and 
animal. you know time, now you have to figure out stuff about animals. 
identify the first sick animals, were they the ones that just got 
purchased from auction market? did they have contact with other animals? 
which ones? could there be an environmental factor? look for it. where is 
the disease? is it just at one place? what's common between affected 
animals, and different b/w affected and normals? same for farms - what's 
common b/w affected farms, and different about unaffected farms? do they 
share equipment, etc?

point source epidemic- exposure quick, 
simultaneous. focus on feed, water, environment. classic is food borne 
outbreak. mostly this is used in human medicine, but also large animal 
practicioners use this. you may not, as small animal practicioner, use 
this, but you should understand how to figure out the problem. all cases 
occur within one incubation period - hours to days. you make you graph - 
plot your curve - it goes up, peaks rapidly, declines rapidly, within 
hours/weeks.
see p 2, fig 12.2 of second handout - how to investigate 
point epidemic.

take geneal history
see which are involved. ask about 
feed, water, environment. you need a fast dx so you can remove the 
problem. 

attack rate table: investigate association of factors and the 
disease. see table 12.1 p 3 of second handout. 

the attack rate is the 
total number of animals that develop disease during a specified time 
period following exposure divided by total number of animals exposed. - 
it's # sick / # exposed.

attributable rate: subtract the rate of the 
disease in the unexposed group from the rate in the exposed group. see, 
some animals may be diseased, but didn't ever get exposed to your source. 
the number you come out with is the % of disease you can attribute to 
that source. (food).

table 12.1:

food borne outbreak attack rate table. 

for pea soup - 59 people got sick of the 255 that ate the pea soup. 
attack rate is 59/255 = 23.13
for people who didn't eat the pea soup, 2 
were sick out of 24, so attack rate was 8.33. then, 23.13 - 8.33 is 
14.80, which is the attributable rate.

the highest attributable rate was 
39.8 which was for raw eggs in syrup. The outbreak was caused by staph 
aureus. now, other foods were also contaminated - because the worker who 
made the eggs in syrup could have also helped make the other stuff, and 
not properly washed up - or utensils, pots/pans, etc. 

a point source 
epidemic may resemble a propagated epidemic if the source is not 
discovered and removed. if you go through your investigation and don't 
figure it out, it will continue and start going over weeks/months - you 
have to remove the source right away - so if farmer has a full feed bin, 
you have to put a different bin out, or empty the bin and refill it, or 
something, even if you aren't sure of the exact problem. 

also, if you 
have a very virulent, highly pathogenic, infectious agent, it may look 
like a point source epidemic at first when you plot it. 

back page of 
second handout - two ways of looking for most probable time of exposure 
in point source epidemic. 

method one: do your plot. you need to know 
what disease you're dealing with. if you think it's a virus w/18 day 
incubation period, take your middle case, count backwards one incubation 
period, and that marks your probable date of exposure. then you can ask 
what happened that day.

method two: need to know range of incubation 
period - say 14-21 days is incubation period. take minimum incubation 
time and subtract it from date of first case. take maximum and subtract 
from last case. then you get a range of a few days to focus your 
investigation on. 

example:
clinical signs reported by farmer:

acute 
onset: 23% drop in egg production, increase in shelless eggs, loose 
droppings, increase in water consumption, increase in mortality

what do 
you do? well, if it's acute onset, it's probably common source/point 
source epidemic. so, ask the farmer some questions:

1. did you change 
your feed or feed supplier?
	last feed obtained was 3-4 days ago
2. 
what's the housing like?
	two flocks of hens on the farm, only one is 
affected
3. did affected hens come in recently?
	no, have been there 20 
weeks
4. when did signs begin? today.
5. in the affected flock - is it 
randomly distributed? well, it's the whole flock.
6. do they share water 
supply? yes, both flocks share water supply
7. are there management 
differences? same people look at both flocks
8. do you feed them both the 
same feed? no. two different feed companies.
9. how's the calcium level 
in the feed? fine. 

based on clinical signs, if you had increased water 
consumption, increased urinary output, what do you think of? diabetes. 
kidneys. sodium. hmm. too much salt in the feed? there was 10x normal 
sodium bicarbonate in the feed. they add this stuff to the feed to 
strengthen the eggshell. this is for acid/base balance reasons. they put 
very low amounts in during the summer because birds pant a lot. the 
supplier missed a decimal point. so extra sodium load, increased water 
intake, loose droppings, gout, renal failure. the increase in shelless 
eggs is also a sign of AI. with the imbalance of acid/base homeostasis, 
the egg is expelled prior to full shell formation, because of 
derangement of normal hormone levels. a shelless egg has only the shell 
membrane without the full mineral deposit. 

so, what do you tell the 
person? change the feed immediately. then, take feed samples for 
analysis. birds recovered somewhat, not fully, b/c kidneys were damaged. 
farmer ended up getting rid of the flock. the feed company ended up going 
out of business after making several other mistakes with other flocks. 


why was farmer using two different feeds anyway? well, different flocks 
on same farm were owned by different people. the grower owned the farm, 
the poultry producers own the birds, and they wanted different feeds. the 
feed company did reimburse them for loss of birds. 

Avian Influenza:

in 
12/85 in PA: there was a case of AI 12/24 brought to Penn State. H5N2. 
this was the same AI from '83, '84 where they had to destroy most of the 
chickens in PA. in 1/86 there were 9 cases, in 2/86 there were 3 or 4. 
this is a propagated epidemic. it was going farm to farm. they made a 
map, to see where the cases were geographically. so they can figure out 
who goes to all the affected farms, what kinds of birds are affected, if 
there's any interfarm bird movement, are the farms related, do people 
move between farms, etc. 

then, a curious situation arose. a case came 
in from NJ. layer birds. everything before then was in PA. now, it's in 
NJ. then, MA, NY, CT. how was this getting spread around the east coast? 
was it related to hatchery? nope, it wasn't, although that's a good idea. 
but AI occurs in birds over 4 wks of age. but, AI can be carried by shore 
birds and wild waterfowl - they thought it was coming from ducks, geese - 
but it wasn't. They don't vaccinate for AI so it can't be vaccinated for. 
(would interfere with the surveillance program). hmm. were all these 
places owned by the same company? no. what was going on was a new thing 
to the poultry vets at NBC. the live bird market system - involves flocks 
ranging from 10 birds to 40,000 birds, and dealers will go around and get 
some birds from each flock, and then go to a wholesaler or a retail 
market and there is no cleaning involved. the major vector was the crates 
- these were old wooden crates that never got cleaned. a crate that 
started out in PA ended up in MA the next week.  these live bird places 
also sell turkeys, geese, ducks, etc. these markets aren't cleaned very 
well.

control measures for live bird market:

plastic crates - major 
wholesaler in NY set up a system using new plastic crates, and dirty ones 
would come in to be washed, birds would be put into clean crates for 
shipping to market. good system. 

clean the trucks.
clean the people.

dealers would go from flock to flock - don't let them. bird owners need 
to bring birds to pickup spot, then go shower before returning to flock. 

We also now have an H7 in the live bird market system, which got into PA 
poultry facilities this year. The federal government does a live bird 
market surveillance program. The market people take birds out back, clean 
up, then bring birds back in. they are supposed to take birds out, kill 
them, clean, bring in new birds. 

---end---