Lecture 11- Analytic Epidemiology Flashcards
Descriptive epidemiology
Used when little is known about the disease, rely on preexisting data, who, where when, illustrates potential associations
Analytic epidemiology
Used when insight about various aspects of disease is available, rely on development of new data, why, evaluates causality of associations
Study designs of analytic epidemiology
Case control, cohort, prospective, retrospective
Case control study design
Study population divided into cases vs. controls
Cases and controls divided into exposed and unexposed
Odds ratio
Measure of association between exposure and outcome
If Odd ratio =1
Exposure does not affect odds of outcome
Odds ratio >1
Exposure associated with higher odds of outcome
Odds ratio <1
Exposure associated with lower odds of outcome
Where does a prospective cohort study start
With the unexposed and exposed group
Where does a retrospective cohort study start
With the disease and no disease of exposed and unexposed
Relative risk
Measure of the strength of association based on cohort studies and randomized clinical trials
=incidence in exposed/incidence in unexposed
Relative risk =1
Indicators no difference between groups
Relative risk>1
Indicates risk among exposed group is higher
Relative risk <1
Indicates risk among exposed group is lower
When is relative risk used
When comparing outcomes of those who were exposed to something to those who are not exposed, calculated in cohort studies and randomized clinical trials, can’t be calculated in case control studies because the entire population at risk is not included or represented in the study
When to use odds ratio
Used most commonly in case-control studies, odds of exposure among cases is divided by odds of exposure among controls, provides a rough estimate of relative risk
What do epidemiology study designs include
Experimental and clinical trials
Experimental study design
Conditions are highly controlled- variables manipulated by researcher
Causality can be inferred by effects of influence of
What is the measure of association in randomized clinical trials
Relative risk
Cross over clinical trials
After the study progresses, the groups are switched, valuable when number of subjects is limited, potential confounders if the intervention has effects that carry over to the next portion
Blind clinical trials
Blinding can reduce various types of bias
Single blinded clinical trials
Subjects do not know which group they are in
Double blinded clinical trials
Neither subjects or investigators know which groups the subjects are in
Randomized clinical trials
Can maximize validity and minimize bias, subjects are recruited from a population and randomly allocated into groups, usually called study and control groups
Community trial
Intervention designed for the usual purpose of educational and behavioral changes at the population level
Ex: water fluoridation in two communities in NY
What does the 2x2 table compare
Your test to the best test that we have
Sensitivity
% of truly positive identified by your test
= A(true positive)/ A +c(false negative)
Specificity
% truly negative
=d(true negative)/d+b(false positive)
Example: a specificity of 95.6% indicates that only 4.4% of truly negative people would receive a false positive test
Why is sensitivity vs specificity important
Determine whether its more important to rule a disease in or out, costs of being wrong
What does high sensitivity mean
We are confident that the animal with a negative test does not have the condition
What does high specificity mean
That we are confident that an animal with a positive test has the condition
SnNOUT
Sn= sensitive
N= negative
OUT= rule out
SpPIN
Sp= specific
P= positive
IN= rule in
How to improve sensitivity
Parallel testing- running different tests simultaneously, disease positive if either or both test is positive, disease negative if both tests are negative, proves you not to have the disease
How to improve specificity
Serial Testing- run screening test- then a confirmatory test if screening test is positive
Disease positive if confirmatory test is positive
Disease negative- if first test is negative
Proves you do have the disease
Positive predictive value
Helps determine of the positive tests how many are truly positive
As prevalence decreases, PPV decreases too
True positive/ true positive +false negative
Negative predictive value
Of my negative tests how many are truly negative
True negative/true negative +false positive
Likelihood ratios
Includes the risk of the patient in front of us (pre test probability) and tells us how much that risk changes when he test is positive or negative (post test probability). Incorporates the probability that the test is positive in patients with and without the disease
2x2 contingency table example: prevalence=10%, sensitivity=96% and specificity=98%. What is the positive predictive value
Prevalence=10% that means that (0.1x1000) 100 animals have the disease and (1000-100)=900 do not
Sensitivity=96% then (0.96x100)=96 (a- true positive) and c=(100-96)=4 (false negative)
Specificity=98%= (0.98x900)= 882 (d=true negative) and (900-882)=18 (b=false positive)
So PPV= a(true positive)/a+b(false positive)= 96/(96+18)=84%
Equation for sensitivity
A(true positive)/a+c(false negative)
Equation for specificity
D(true negative)/d+b (false positive)
Equation for positive predictive value
A(true positive)/a+b(false positive)
Equation for negative predictive value
D(true negative)/d+c(false negative)
Calculating odds ration example: determine whether cats with hyperthyroidism and without hyperthyroidism differ in exposure to flame retardants
N=100,000
50 have hyperthyroidism
50 with hyperthyroidism, 38 of their owners indicated they have a couch treated with flame retardant
50 cats without hyperthyroidism 20,000 of the owners indicated they had owned a couch treated with flame retardant
38=A
20,000=B
C= 50-38= 12
D= (100,000-50)-20,000
OR=(A x D)/(B x C)= (38 x79,950)/ (20,000 x 12)= 12.7
Cats with hyperthyroidism were 12.7 times more likely to have exposure to flame retardants than cats without hyperthyroidism
Calculating relative risk example: curious to see if the risk for developing FIP is different for cats belonging to smokers or non-smokers among a group of 500 kittens
A= 80
B= 120
C=15
D=285
RR- (A/(A+B))/ (C/C+D))
RR= (80/80+120)/ (15/15+285)
RR= 8.0
The risk of FIP is 8 times greater among cats belonging to smokers
Example: you suspect hyperadrenocorticism in a 9yr, FS dog with 2 month history of increased appetite, increased thirst and urinary accidents. Which diagnostic test do you trust the most if he has a positive result
Urine cortisol creatinine ratio: sensitivity=90%, specificity=25%
ACTH stimulation: sensitivity=80%m specificity=85%
Low dose dexamethasone suppression: sensitivity=95%, specificity=50%
ACTH stimulation- most specific- fewer false positives
Example: you serological test 140 wallabies for a disease, 35 wallabies test seropositive and 105 test seronegative. However, postmortem data reveals 5/35 of the seropositive wallabies are disease free and 4/105 of the seronegative wallabies are disease
What is the sensitivity of this serologic test:
A. 95%
B: 96%
C. 88%
D: 86%
E: 77%
Sensitivity= A/a+c= 30/(30+4)=88%
C. 88%
example: you serologically test 140 wallabies for disease, 35 wallabies=seropositive, 105=seronegative, however postmortem data reveals 5/35 seropositive wallabies were disease free and 4/105 seronegative wallabies were diseased
What is the predictive negative value of this serologic test
A. 95%
B. 96%
C. 88%
D. 86%
E. 77%
NPV= d/(c+d)= 101/(4+101)=96%
B. 96%
Example: you are using an FeLV test with sensitivity of 90% and specificity of 95%. Assuming the prevalence of FeLV in you area is 5% what is the positive predictive value of your test
A. 45%
B. 48%
C. 55%
D. 88%
E. 90%
PPV= a/(a+b)
Prevalence =5% so (0.05x1000)= 50 of 1000 animals have disease and 950 don’t
Sensitivity=90%= (0.90x50)= 45 (A)
C= 50-45= 5
Specificity=95% (0.95 x 950)= 902.5 (d)
B= (950-902.5)= 47.5
PPV= a/(a+b)= 45/(45+47.5)= 48%