EBM Day 4 Flashcards
Cohort Study vs Case Control
Cohort-know exposure, look how exposure level effects disease-start with cohort who are exposed (at risk for disease)
Case control-know disease, what factors give rise to disease-start with people with/without disease
Case control OR vs Cohort RR
Exposed CasesxNot exposed Noncases/not exposed cases x not exposed non cases
Exposed Cases x Exposed noncases/nonexposed cases x non exposed noncases
When OR is what, good idea of RR
Low-looking at disease that is not found much in population
Case control studies facts (that cohort can do)
- can’t yield incidence rates
2. can not give risk ratios
Case control strengths
- rare diseases
- diseases with long induction time
- explore wide range of exposures
- quick/cheap/easy/yields potential hypot
Bonferroni Correction
Correcting p values over multiple studies-leads to error
Random error, systematic error, bias
what do they look like
random error has points more spread out
bias moves points toward or further away from normal
systematic looks like a pattern
Oversampling
distorts odds ratio
Recall bias
Take unexposed or exposed and place in opposite group (because for recall)
Case definition traits
clear, specific, but not overly restrictive
misclassification if too broad
limited sample size if too strict
Should cases be incident?
Yes stop recall bias/less effect due to prolonged exposure
How to select controls
Should have same oppurtunity to have been exposed
-should be population risk at becoming a case
Should be sampled independent of exposure
- want people who are very similar except in exposure
-if not have selection bias
2x2 table
draw
Diagnostic/Workup Bias
Case selection influenced by physicians knowledge of exposure
Nested Case Control Studies
Select cases and controls from cohort study
More eficient- already have most info
healthy worker effect
People who work are more healthy then those who do not
Matching
control selection coupled with experimental selection to reduce confounding variables
More controls
Increase power until 4, then not worth
Non differential misclassification vs differential misclassification and what error they lead to
Exposure unrelated to disease (chance)
All different groups (variables) have equal rate of being misclassified
Leads to type II error
Different groups have unequal rate of being misclassified
Leads to type 1 or t2 error
Information bias
bias due to measurement error
How to minimize recall bias
Using records of exposure to disease, use incident cases, appropriate control, blind study, etc.
Investigator bias and how to not have
Investigator does something like leading questions because knows exposure status he is looking for
Standerdized protocols, objective measurements, blinding
Adv of case control studies
rare disesase, new disease, outbreaks, induction period is long, inexpensive, multiple expusres
Disadv of case control studies
Bias (recall and misclassifcation), singe out come, inefficient if low freq, does not calculate incidence rate directly
Cross sectional vs Cohort vs case control
CS-sample population , no follow up, compare disease experience among groups in present
Ch-identify exposed, follow exposed through disease course
CC-identify disease cases and noncases, compare histories of past exposure
When to use regression
Looking for trend in data between two variables
Microarrays
Adjusting for confounding variables
When to use correlatoin
When don’t know IV or DV
Examine relationship between two variables
Variance and which kind of tests assume equal
How far numbers are spread out
Parametric
R^2
Correlation coeffecient
Amount of variability in Y contributed by x
Meaningfulness of the correlation coefficient
Effect of outlier
Destroys parametric correlation because variances are unequal
Nonparametric tests have no problem
How to deal with outlier
Drop it
Log transform
Leave it
Nonparametric test
Different types of regressions
Linear-DV=continuous, IV=single and continuous,
Nonlinear-DV=continous, IV=1 or more and continous
MV-DV=continous, IV is continous or categorical
Logistic=DV=categorical, IV is continous or categorical
MV analysis
Looking at multople variables at a time
Allows simulataneous assessment of different variables and adjust for confounders
Possibly look at interaction between terms
Stepwise regression
chance of getting something by doing this, then this and that, then this that and other thing
OR
odds ratio of dead person with condition/
odds ratio of alive person with condition
Multiple logistic regression
Gives odds ratio for each independent variable
Adjusts for confouding
Logtistic Regressions
when outcome or dv is binary
adjusts for confounding
good for odds ratio in case controls
Principal Component Analysis
Takes many variables and reduces by regression
ex. 100s of diet items (put into 3 categories)
Couple with logistic regression to get odds ratio
ex. 3x chance of getting cancer if only eat meat and fat
Zero time point (and examples)
start of study
now
date of randomization
first MI
Median survival time
Half of sample reaches the event (death or discharge usually)
MI risks for first and second
They are same, and prevented same way
Can you use experience to say how long someone has to have an MI?
Not really, each patient has different propensity to mi
Equipose
genuine lack of consensus in the medical community about a treatment or prognosis
-only way to have RCT on patient
Case fatality
percent of of patients with disease who die due to it
response
percent of patients showing some improvement following an intervention
Kaplan Meier Survival curve and CI and how to match?
x is usually month/year
y starts at 100% alive, and decreases (cum probability to survive)
Can draw CI and if one curve within CI of another curve, no difference
PROPENSITY
Truncation
Entering study
Event occurred before start of study
Event occurred after start of study
Censoring
Leaving study
Incomplete followup
Event occurred and left study early
Kaplan Meier limitations
Does not handle co-variates (use proportional hazards)
Cox Proportional Hazards (Regression)
Hazad Ratio=risk ratio, can control or adjust for other factors
(ex BMI and sex)
HR1
HR1 and CI no include 1=worse off
