Introduction to Epidemiology Flashcards
epidemiology
the study of the distribution and determinants of disease and health related events in populations
AND
the application of this study to controlling and preventing health problems
p value
probability of the data being compatible with the null hypothesis
null hypothesis
assumes there is no association between X and Y
high P value
high probability that there is NO association between X and Y
do not reject the null hypothesis
p > 0.05 means that you’d expect to obtain the data you did >5% of the time, therefore there is likely no association between X and Y
low P value
low probability that there is NO association between X and Y
reject the null hypothesis
p < 0.05 means that you’d expect to obtain the data you did <5% of the time, therefore it is likely that there is an association between X and Y
hypothesis testing
start by assuming the null hypothesis is true and determine the probability (p-value) of getting your results if the null hypothesis was true
alpha level
level of significance; cutoff value of P for rejecting/not rejecting the null
usually 0.05 or 5%
type I error
willing to err on the side of wrongly rejecting the null hypothesis 5% of the time
does statistical significance always correlate to clinical significance
NO
- low P value can reflect minimal clinical effects
- high P value can reflect significant clinical effects
what are P values a function of
sample size
large sample size = greater precision
statistically insignificant data from a study with a small sample size could be significant if the sample size was larger (type II error)
type II error
even if a difference between groups does exist, the study may not be statistically powerful enough to find those differences
ex. small sample size = low statistical power
confidence intervals
interval with an associated probability (confidence) of 1-alpha
“we are (1-a)% confident that the confidence interval contains the true population meausre
what does it mean if 1 is included in the confidence interval
no association - means that the risk could be equal to 1 (equal risk in both groups)
what are the source of error in a study
random: variance
systematic: bias
validity
unbiased/accurate data
what we are measuring is what we want to be measuring
precise
variances/standard deviations/standard errors are small (minimal variability in the data)
confounding bias
comparing study populations that would have different health or disease outcomes even if they had the same treatment/exposure
comparing things that shouldn’t be compared
how can confounders be controlled for
randomization
what is confounding by indication
treatment is not randomized but deliberately chosen for specific reasons (clinical indication, cost, availability)
selection bias
choice of individuals selected to be in studies leading to invalid results
follow up bias
not following up with all individuals over time (censoring) leading to bias
information bias
occurs when the treatment/exposure and/or the outcomes are measured with error
specification bias
doing the statistical analysis wrong
