chapter 3 biostats Flashcards
1, 2, 3 SD around the mean = percents?
68
95
99.7
sensitivity test?
true positive / people with disease
for screening
low false negative rate
specificity test?
true negatives / people w/o disease
for confirmation of disease
low false positive rate
tradeoff between sensitivity and specificity
changing the cutoff for screening
if higher: more FNs, fewer FPs (if higher, you miss more people)
if lower: fewer FPs, more FPs (if lower, you catch more people that are not sick)
PPV
probability of having disease given positive test
true positive / all people with positive test
higher prevelance = higher PPV
high sensitivity = low PPV, bc more FPs
NPV
probability of not having disease given negative test
true negatives / all negatives
lower prevalence = higher PPV
high sensitivity = high NPV, bc fewer FNs
attributable risk
cases attributable to one risk factor = the amount by which the incidence of a test will decrease if the RF is removed
incidence rate in population - incident rate in smokers = attributable risk due to smoking
relative risk
compares disease risk in those exposed to a certain factor with the disease risk in people who have not been exposed to the factor
in prospective or experimental studies (not retrospective) cannot be calculated otherwise
significant if NOT 1
- 5 x more likely to develop disease if exposed to factor in question
- 5x less likely to develop disease if exposed == protective factor
OR
estimates RR in retrospective studies (case control)
skewed distribution
meaning not normal distribution
positive skew: lots of high values, tail of the curve is on the right. mean>median>mode
neg skew: lots of low values, tail on left. mean
test reliability
=precision
=reproducibility and consistency of test
= random error reducers reliabilitly and precision
test validity
=accuracy
=trueness of measurement, does it measure what it claims to measure
=systematic error reduces validity and accuracy
define correlation coefficicent
what degree are two variables related
-1 to +1
more strong = absolute value away from 0
if positive = positive correlation
if negative = negative correlation
confidence interval
95% certainty that the mean is within this range
1sd - 68
2sd - 95
3sd - 99.7
experimental studies prospective retrospective case series prevalence surveys
order of relevance from best to lowest
experimental studies
well matched controls
sometimes ethical c/f is why we can’t do it
prospective studies
cohort or observation
dividing groups based on presence of RF and following to determine the outcome
can determine incidnece + RR
retrospective studies
choose population samples after the RF
only OR, no incdence + RR
less expensive, better for rare disease
case series study
describes presentation of disease, good for rare disease
prevalence survey
cross-sectional
prevalence of disease and risk factor
could suggest a cause of disease
incidence versus prevalence
incidence= #new cases of disease in a unit of time incidence = absolute risk of developing a condition
prevalence = total number of all cases at a point in time
what happens to incidence and prevalence if disease can be treated to the point where only people can be kept along longer w/o being cured, what happens?
incidence = nothing prevalence = goes up
short term diseases = incidence>prevalence
long term = prevalence>incidence
nominal = no numeric value (day of week)
ordinal = ranking
continuous data
nominal/ordinal data = use chi square
t-test/anova = continueous data
pvalue
if p<0.05, then less than 5% chance that the data were obtained by random error or chance
p value error?
type 1 error = claiming an effect when none exists = pvalue percentage
if<0.05, 5% chance of type 1 error
type 2 error
claiming no effect when it does exist
power
probability of rejecting the null when its false
increase by increasing sample size
confounding variables
affect independent and dependent variable
nonresponse bias
ppl don’t return surveys
list nonresponders as unknown
lead time bias
cancer screening prolongs survival only bc of early detection
admission rate bias
comparing MI
hospital A has higher mortality rate bc of higher admission criteria meaning sicker pts
recall bias
retrospective studies
overestimate risk factors
interviewer bias
absence of blinding
investigator makes an efect when doesnt exist
unacceptabilitly bias
pts don’t accept they do bad things so this changes the accuracy of data
