This site is 100% ad supported. Please add an exception to adblock for this site.

Medical Epidemiology

Terms

undefined, object
copy deck
Attributable Risk( AR)
Incidence attributable to a specific exposure. AR = I(exposed) - I(unexposed)
Proportion of Attributable risk
(PAR)
PAR = (I (disease Exposed)- I(disease unexposed)) / I(disease exposed)
Bradford Hill's Criteria
Blow jobs Cause STDS:
Biological Plausability (do they give an explanation)
Consistency (Have other papers been published to suport this?)
Strength of Statistics (is the RR or OR statistically significant)
Temporality (Can causality be determined)
Dose-response(Does disease exacerbate as the dose is increased)
Specifity( Does the exposure have a single outcome)
Case Control Study
(retrospective)
Define “cases” who have disease and “controls” who are similar to cases but do not have disease.
Look for exposures within each sub group. Sampling is based on outcome
· OR (Odds Ratio) only estimates RR (Risk Ratio) if frequency of disease is low
· Adv: Relatively inexpensive, req. small study population
· Disadv: Need exposure to be frequent among cases. Does not include newly diagnosed cases.
***Cannot determine causality***
· Possible Biases: Assessment of exposures (incomplete info), recall bias, survival bias (you’re only
interviewing people who are living with disease)
Cohort (prospective) study
· Gold standard of epidemiology. Determines temporality.
*The only better study is randomized*
trials since this method randomly associates exposure with people. This eliminates confounders
· Interview and divide study nondiseased population into exposed and unexposed. Study to
determine where disease occurs
· Can be done concurrently (following a population) or retrospectively (using published data)
· Adv: Can calculate absolute risk (AR) & relative risk (RR)
· Disadv: long follow-up, expensive, requires lots of $$
· Possible sources of bias:
o In assessment of outcome
o Info bias: quality of info for exposed vs. unexposed
o Non-responsive vs. losses to follow-up
o Analytic bias
· Need to know risks with some certainty, need to have methods for good follow up, easier to do
when pathogenesis & latency are short. May be difficult to find unexposed
Cohort effect
Artifact created when comparing cross-sectional data because an age group (“cohort”) moves
through time as a discrete population. You’re studying the same population more than once
Common null hypothesis
RR = 1.0 Thus, the exposure has no effect on the outcome. You want to be able to reject this and
accept your HA (alternative hypothesis) that the RR¹1
Confidence Interval
Range around a calculated result in which if the study was performed 100 times, the result would
fall within the CI 95 times.
· CI = 1 – _
Confounder
· Factor that is related to exposure and outcome but is not in the casual pathway
· E.g. race, socioeconomic status, level of education, immigrant status.
Cross sectional study
· Hand out surveys and have people fill ‘em out Cheap, easy, fast.
· Determine exposure and outcome simultaneously
· Get odds ratio
· Can suggest risk factors but cannot determine temporality or etiology
Incidence
= (# new cases) / [ population*time ]
Nested case control
· Mix of cohort & case-control study
· Begin with defined population. Perform interviews and collect samples. Perform follow up study
and divide population into “case” and “controls” and used to initial data to determine exposures
· Removes recall bias, determines temporality, cheaper than cohort.
Odds ratio (OR or RO)
· Ratio of odds that disease will develop in exposed person to odds that it will develop in
nonexposed person. RO approximates RR when disease frequency is low, controls & cases are
representative of the population
· = ad/bc
Overmatching
· Process of making case and controls too homogenous; not sampling of general population
· Can occur when neighborhood or friend matching is performed
P value
· If p < 0.002. Assuming the null to be true, the probably of obtaining study results even more
inconsistent with the null, if the study were repeated many more times, is less than 0.002.
· If p < _, reject null hypothesis
· Depends on sample size, standard error, and magnitude of difference
Prevalence
· = (# total cases at present time) / (population # at present time)
· Period Prevalence = (# cases per given time) / (population during that time)
· = incidence * duration (only for diseases with short duration)
Power
Power · = (1 – _)
· The ability of your test to find difference when it actually exists
· If power = 0.8, then 80% of the time, you will detect difference
· The sample size and magnitude of difference determines _
Relative Risk (RR)
· Ratio of the diseased exposed to the diseased unexposed
· = (Incidence in exposed) / (Incidence in unexposed)
· = ad/bc
· calculated directly ONLY in cohort study
o if RR > 1, there is a positive association between exposure & outcome
o if RR = 1, there is no association (cannot reject null hypothesis)
o if RR < 1, there is a negative association (exposure minimizes outcome)
· association does not mean causality
Stratification
Stratification · Process of removing confounders by performing statistics for each group separately (i.e. separate
out race or age groups)
Direct Standardization
· Applying a set of weights to population
· Indirect standardization involves more than one set of rates
Standard Error
s / (sqrt n)
Type I Error (_)
· You reject the null hypothesis when it is true. False positive
Type II Error (_)
· You accept null hypothesis when it is false. False negative

Deck Info

23

permalink