case control
case-control
- observational, retrospective
- take condition (often rare) and matched control
- look back for exposure to certain thing
- give odds ratio
cohort
cohort
- observational, prospective
- follow cases (w/ certain exposure) vs controls and look for development of problems
- give relative risk
sensitivity and specificity
spec: true neg/ not got
- high specificity helps rule in
sens: true pos/ have got
- high sensitivity helps rule out
absolute risk reduction
(rate of event in controls) - (rate of event in group)
Number needed to treat =1 / ARR
funnel plots
show publication bias in meta analysis
cross sectional survey
snapshot of clinical scenario
low statistical value
odds ratio
if 80 people take drug 40 get better
odds of positive outcome is 1 (40/40)
if 100 were control placebo and 25 got better
odds is 25/75=0.33
odds ratio is 1/0.33=3
power and p value
power= 1 - prob of type 2 error
p value = chance of type 1 error
type two say null is true when not (false neg)
type one say null is false when its true (false pos)
comparison tests
parametric
- students t test
- pearsons - correlation
non-parametric
- chi squared - percentage
- man whitney - unpaired data
- wilcoxon - 2 obs on same set
error type 1 and 2
T1 : null rejected but true (p value is chance of this)
T2: null accepted but not true
null hypothesis is that there is no relationship between 2 measured phenomena
1-(T2) = power
standard deviation
root of variance
- 3% w/in 1sd
- 4 - 2sd
- 7 - 3sd
95% within 1.96sd
