Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
33 Cards in this Set
- Front
- Back
Goals of epidemiology; what do they all require?
|
describing disease in population, understanding underlying causes of disease, preventing disease
all require reliable & valid measurement |
|
Central tenet of measurement theory
|
all scores equal the 'true' score plus measurement error
|
|
About reliability
|
higher measurement error => lower reliability
consequences of low reliability: 1) tendency to mask significance (i.e. biases studies towards finding no differences) 2) as a result, need larger sample size for sufficient statistical power |
|
two measures of reliability
|
inter-observer agreement and test-retest reliability
|
|
What's Cohen's Kappa? What is it for?
|
Cohen's kappa is a measure of inter-observer agreement; equivalent to:
extent to which two measures agree beyond chance / maximum possible improvement in agreement (% agreement observed - % chance agreement expected) / (100%-chance agreement expected) % agreement observed = (agreed '+' + agreed '-') / (total responses) i.e. ( a + d / (a+b+c+d) ) % agreement by chance expected = '+' agreement expected + '-' agreement expected '+' chance agreement expected = ('+' from person A / total responses from person A) * ('+' from person B / total responses from person B) |
|
sensitivity
|
the proportion of 'true' cases which the measure identifies as 'positive'
|
|
specificity
|
the proportion of 'true' non-cases which the measure identifies as 'negative'
|
|
positive predictive value
|
the proportion of positive-identified cases that are, in fact, 'true' cases
|
|
negative predictive value
|
the proportion of negative-identified cases that are, in fact, 'true' non-cases
|
|
What do sensitivity, specificity, PPV and NPV have in common?
|
they all are measures of 'criterion validity' i.e. of validity against a gold standard
|
|
Outbreak, epidemic, pandemic
|
All 3 describe 'elevated' levels of disease; outbreak is most localized geographically (i.e. county/city level), epidemic is more widespread and pandemic is global
|
|
Epidemiologic Triad
|
Vector is in the center; Environment, Agent, and Host are at vertices
|
|
Attack rate & what influences it
|
(# of exposed and sick)/(total # of exposed)
Some influencing factors: herd immunity, incubation period, virulence |
|
Sample, Population, Bias
|
Population: everyone for whom you want to draw a conclusion
sample: representative sub-group (i.e. as congruently comprised as possible) of population bias: when sample isn't representative of population |
|
prevalence, incidence, odds
|
prevalence: # of cases / sample size
incidence: # of NEW cases / sample size odds: # cases / # non-cases |
|
When comparing two adult groups, what is most important variable to adjust to reach sound conclusions? Why?
|
Age is most important variable to adjust, because of physiological differences between people over chronological time
|
|
Case-fatality rate, years of potential life lost, disability-adjusted life-years
|
case-fatality rate = (# of dead from disease) / (# of people with disease in sample)
YPLL = Σ(65 - age of death); speaks more to early deaths Disability-adjusted life-year: 1 year of healthy life lost to either death or disability |
|
Advantages of survival analysis
|
1) accounts for censoring
2) addresses *time* until an event occurs |
|
Most Impt. thing to remember about survival analysis
|
'cumulative survival' is multiplicative but the denominator of each proportion changes as people are censored/die
|
|
Hazard Ratio
|
relative risk integrated over time; used in survival analysis
|
|
With respect to timing of measures, what are the three epidemiologic research designs?
|
prospective (i.e. cohort), retrospective (i.e. case control) and concurrent
|
|
Observational Study Vs. Experiment
|
Experiment involves researcher intervention (i.e. is treatment-based); observational study only collects data
|
|
Ecologic study
|
exposures unlinked to outcome case-by-case, usually uses incidentally collected data (i.e. city-wide, country-wide, company-wide etc)
|
|
Cohort study
|
samples on *risk factor* and follows forward for outcome; no need to oversample for rare risk factors because oversampling is inherent
well-suited for relative risk because relative risk describes chance of *developing* disorder |
|
Case-control study
|
sampled on *outcome*, tracks back through survey or records to risk factor; can oversample for rare outcomes (but is hard to deal with mathematically)
well-suited for odds ratio because OR describes odds of *having been exposed* given that you are disordered |
|
Relative risk
|
describes relative chance of developing disorder;
( incidence in exposed) / ( incidence in unexposed) (a / (a+b) / c / (c+d) ) |
|
Odds ratio
|
describes odds of having been exposed to risk factor
(odds of past exposure among diseased) / (odds of past exposure among healthy) (a/c) / (b/d) |
|
Attributable risk; Population attributable risk
|
excess cases among exposed (i.e. risk BEYOND background risk);
AR = incidence in exposed - incidence in unexposed PAR = AR * prevalence of risk factor in a population answers the question "how much could the incidence of this disease in pop. be reduced if risk factor eliminated?" |
|
Null hypothesis
|
in population, there is no association or group difference
|
|
Type I; Type II error
|
Type I error: rejecting true H0 (i.e. false conclusion of difference)
Type II error: accepting false H0 (i.e. false conclusion of no difference) |
|
p-value
|
the probability of obtaining the test statistic in question *if H0 is true*
|
|
95% confidence interval
|
interpreted as '95% probability that true value lies within bounds' -- really means that in repeated sampling value would appear within this interval 95% of the time
|
|
determinants of statistical power
|
# Effect size
# Prevalence of outcome and exposure # Significance level (p) # Reliability of measures # Number of covariates (measured confounds) |