March 19 - Biostats/epidemiology Flashcards
Clinical meaning of high sensitivity and high specificity
High sensitivity means a negative test is good for ruling out diagnosis
High specificity means a positive test is good for ruling in a diagnosis
Attack rate formula
number who become ill / number at risk
Negative and positive skey
Neg skew: large values predominante in data set (tail at neg end of curve). Mean less than median less than mode
Pos skew: small numbers predominate in data set (tail at pos end of curve). Mode less than median less than mean)
Power and beta
Power = 1-beta = ability to detect a difference when it exists
beta=probabiliy of comitting a tyep II error (failure to reject null)
Relative risk ratio
=risk in exposed/risk in unexposed = a/(a+b) / c/(c+d)
Can’t be used in case control, except for rare diseases where OR equals RR. Prevalence less than 10%.
Odds ratio
Used in case control studies. OR=disease in exp/disease in unexp // no disease in exp/no disease in unexp
=a/c / b/d
Attributable risk and attributable risk percent
attributable risk =risk in exposed - risk in unexposed
attributable risk %=RR-1 / RR = (risk in exposed - risk in unexposed)/risk in exposed = percent of disease explained by the risk factor
Number needed to harm
1/attributable risk
Secondary vs tertiary prevention
Secondary=interrupting disease process before symptoms develop
Tertiary prevention=treating established condition to minimize progression/complications
Ecological study
Looks at populations rather than individuals. Compares frequence of characteristic and outcome in different populations. Can’t be used to draw conclusions, only to make hypotheses
Relationship between confidence interval and p value
When 95% confidence interval does not contain null value (e.g. RR of 1), then p is less than 0.5 and result is statistically significant
Standard error
Std error = SD/sq rt (n)
As sample size increases, std error decreases
Calculation of confidence interval
95% confidence interval = mean +/- 1.96 x SE
where SE=SD/sq rt(n)
Case control vs cross sectional study
Case control
- takes those with and without disease and looks back at their exposures
- caclulae exposure odds ratio
Cross sectional
- takes those with and without exposure and looks at diseaese
- calculate prevalence odds ratio
Hawthorne effect
Tendency to change behavior when know being studied
Berkson’s bias
Selection bias of using hospitalized patients as control grup
Effect modificiation vs confounding
Effect modification: effect of exposure on outcome modified by another variable; shows difference between strata
Confounding: no difference between strata. Effect seen on whole due to, for example, those who drink more also smoking more contributing to false association between drinking and lung cancer
Probability that at least one test result will be positive: formul
=1 - probability that all will be positive
true whenever have independent events
Attrition bias
results from loss of subjects if they differ in their risk of developing the outcome than those that are retained. Form of selection bias
