Biostats Flashcards
Incidence vs prevalence
Incidence is number of new cases per unit time
Prevalence is total number of cases in population
SD vs SEM vs CI
SEM = sd/ sqrt(n)
CI proportional to SEM so with given sd, must x4 n to get double precision and halve the CI
When to use chi square test?
To assess if groups are different along a discrete category (eg having disease vs not)
RR
Comparative risk of outcome with or without exposure
A/(a+b) all over c/(c+d) where a and b have exposure, c snd d don’t
Use in cohort studies
Odds ratio
Retrospective case control, of those who have disease or don’t, how many were previously exposed?
A/c divided by b/d where a and c have disease
OR vs RR
OR case control, retrospective
RR cohort, prospective
Sensitivity, specificity, ppv, npv
Sens - if ppl with disease, how many test pos? If high, neg rules out dz
Spec - of ppl wo disease, how many test neg? If high, pos rules in dz
Ppv - of people with pos test, how many truly have dz? Changes with prev
Npv - of ppl with neg test, how many don’t have dz? Increases with lower prev
ARR
Percent without tx who have outcome - percent with tx who have outcome
Nnt = 1/arr
Vs rrr is arr/risk without tx
Verification bias
When you do gold standard testing selectively to confirm (eg biopsy)
Susceptibility bias
Selection bias in which Control and test groups differ from prognostic standpoint
Denominator in maternal mortality
Live births
Length time bias
A screening test preferentially detects less aggressive forms of disease, and therefore increases the apparent survival time
Forms of selection bias
Whenever subjects aren’t representative of study pop
Eg high dropout raises risk, berkson (hospitalized pts), referral bias (pts from specialized center), nonresponse bias (eg sicker pts don’t respond), prevalence bias (neyman bias - selective survival), susceptibility bias - tx depends on illness severity
Types of measurement (information) bias
Recall and observer (aka ascertainment, detection, or assessment - investigator’s label is affected by knowledge of exposure status —> solution is blinding!)
When to use fisher exact test?
Like chi square (categorical proportions) but with small sample size
