---start----

epidemiology
1/21/98

first we did problem set #3. 

now, 
he's telling some story about sheep and mathemeticians...

Prognosis:

a 
prediction of the expected outcome, with or without treatment...
it is 
expressed as the probability that something will occur in the future.


practicioners should avoid statements that can be misconstrued as 
statements of a contract, eg if this...then that. this is when you start 
making implied contracts and becoming vulnerable to lawsuits.

a client 
should be made aware of the probabilities of favorable and unfavorable 
outcomes. this is not to say you shouldn't maintain a positive attitude. 
you can't be sued for being cheerful. but don't say "if you give it this 
drug, he'll get better."

as we shall see, these probabilities usually 
take the form of cumulative incidence of some event (death, cure, 
disability, complication, etc)
we're not using anything different. it's 
the proportion of instances where an outcome occurs with regard to the 
total number of animals at risk. cumulative incidence.

there have been a 
number of studies on what constitutes a good expression of prognosis. an 
ideal case is:

things that should be included in a prognosis: 

1. 
variability in course relative to treatment options (eg, if you do this, 
you're likely to see this occur, but maybe 5% of cases will have 
complication x)
2. a time reference - if death from disease will occur in 
5 yrs, that is a different prognosis in a 15 yr old animal than a disease 
that will kill it in 5 days. or, "most dogs recover, but sometimes it 
takes 4 years to recover..."
3. risk of treatment related death or other 
complication
4. cost - while we cannot put a  cost on human lives, people 
can and do put costs on the lives of their pets. ditto racehorses - 
insurance companies will make those decisions.
5. nature of the benefit 
attainable - can you cure the animal? arrest the disease? what is cure? 
when something else kills you first? when the symptoms go away? something 
else?

generally, casual expressions of prognosis won't include all the 
above...but you should try

expressing prognosis - some expressions used:


survival: proportion of patients that survive a defined interval from 
some defined point in the course of disease. it's meaningless to discuss 
survival without time. you can't say "there is a 50% survival rate with 
this disease." you have to say "50% rate of survival over one year from 
diagnosis" or something. often, survival will differ depending on when 
you start - from beginning of tx, or from time of dx, or time of first 
occurrence of dz, or what? often, when you quote survival rates, you 
aren't told when time zero is. if you see a survival rate, always check 
this.

case fatality: proportion of patients who die of the disease. 
really should have a time reference. how do you know you got to the end? 
maybe an animal died of the disease when it was 21...

response: 
proportion of patients improving after treatment. time- how soon? 
absolutely when you follow all cases forever?no, you can't do that.


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

recurrence: proportion of patients who experience a 
return of disease after remission. again, timeline is important. 

all 
these expressions are used. there are a whole number of problems when you 
try to work this stuff out. 

problems in trying to measure /describe 
outcomes useful in prognosis:

1. natural history of dz is rarely 
observed, because sick animals are treated - no baseline for comparison. 
what do you compare your prognosis with? if you say "death rate of cats 
over 5 years with this dz is x" what do you compare it to?

2. reports of 
prognosis from vet teaching hospitals may not be representative of cases 
seen in typical practice - referred cases have often been referred 
because they are more sick than other cases. 

3. qualitative expressions 
of prognosis mean different things to different people. you know...they 
talk about someone in the hospital doing "very poorly." what does that 
mean? at death's door? having a bad hair day? when someone desperately 
hopes for a favorable outcome they may always look for the silver 
lining...

qualitative terms for clinical outcomes:

prognosis			
probability of recovery (%)

excellent			90-100
good 				70-89
fair				
40-69
poor				10-39
grave				0-9

(Crowe, 1985)

essentially, we're 
talking cohort study. you take a group of animals with the same disease 
and figure what's going to happen. problems:

1. zero time - do you start 
from the time you start tx? the time animal presented with dz? the time 
owner first noticed signs of dz? different measures of prognosis use 
different starting times. 

2. interval of followup - comparing tx x with 
tx y or no tx, you may be comparing your work to someone else's work, and 
often, for many reasons, the interval over which the event of 
significance is recorded, is different. now, cumulative incidence 
increases with time. if you measure incidence of death in animals given 
tx x, over 2 yrs, and compare it with animals given tx y, you'd also want 
to do it over 2 yrs. but sometimes you do it over a different time 
period.  if you are designing the study, you have to watch the animals 
long enough to figure out the prognosis - if you want to see an event, 
your observation period must exceed the induction period. so interval of 
followup really has to be the same for all cohorts bcause cumulative 
incidence is meaningless unless time period is defined.

example of this 
problem:

mortality of FeLV infected and noninfected cats:

cause of 
death		FeLV+	FeLV-

LSA					15.2%	0.6%
other				13.0%	0.2%

FeLV+ figures 
- two year followup
FeLV- figures - three year followup. these negative 
cats have had a whole extra year of being watched in which they had a 
chance to die. their incidences would really be less if only watched for 
2 yrs. so FeLV+ really confers a worse prognosis than this indicates. 


another problem:

survival analysis: summary survival rates only tell us 
what patients with a given condition are likely to experience at some 
given point in time. 

japanese people have low infant mortality and high 
longevity. US has dreadful infant mortality rate - very high. kenya has 
huge infant mortality and mortality during first two years - then go on 
to live as long as japanese people. 

survival analysis provides 
information about the average time to event for any point in the course 
of disease. 

eg: survivorship curve for postop survival of 15 cats 
treated for hemangiosarcoma. plot survival vs time. these curves are 
often difficult to contract because some patients drop out of the study. 
consider the average time to death - after 20 weeks, maybe 80% of the 
cats are still alive....after 120 weeks, only 25% of them are still 
alive.

using the same zero time and the same period of observation, you 
can compare survival rates. the chart he shows now indicates that cats 
with + LNs after mammary CA resection have lower survival rates than cats 
with - LNs.  the problem with this is that to construct these curves 
requires that none of the patients drop out of the study. if someone 
moves away with their cat, and the cat is lost to followup, you're 
screwed, analytically speaking. then you have to do "life table analysis" 
and the difference between that and survival analysis is that in life 
table analysis you can account for the fact that some animals have 
dropped out of the study. 

life table analysis: can be used when some 
patients drop out of the study. the essence of life table analysis is 
that the number of individuals at risk (of dying or whatever) over each 
interval can be adjusted for those that drop out.

interval	censored	dead	
at risk	survival%
0			0			0		15		100
6			0			1		15		93 (15-1/15)
13			0			
2		14		87 (15-2/15)
15			0			3		13		80	(15-3/15)
20			*2			4		10		72 
(13-4/13)

* censored = withdrew from study - moved to connecticut
			

the only thing to remember is that life tables will account for those who 
drop out of the study. that's it. that's all. 

one last thing. if you 
get a survivorship curve that starts high and ends low, don't think that 
because slope decreases, animals at the right of the curve have an 
increased probability of surviving any given interval. although the slope 
of the curve is decreasing, the probability of surviving any given 
interval is the same irrespective of how far away you are from zero time. 
the probability of dying is the same everywhere - but you keep losing 
survivors. so an exponentially decaying curve represents a constant 
probability of an event occurring. it's just curved because as time 
increases, the number of animals at risk decreases. 

the exam will not 
have anything on it about prognosis.


---end----