Principles Of Epidemiology Final

Front 1

Goals of epidemiology

Back 1

Describe distribution, understand causes, prevent spread

Front 2

Epidemiologic triad

Back 2

The agent (disease-causer) is sometimes directly transmitted to the host and otherwise is transmitted through a vector; agent, vector and host are all affected by the environment

Front 3

Population, sample, bias (in this context)

Back 3

population: everyone for whom you want to draw a conclusion

sample: representative subgroup of population

bias: when sample isnt representative of population

Front 4

Prevalence, incidence, odds

Back 4

prevalence: # cases/total # in sample

incidence: # NEW cases/total # in sample

odds: # cases/# non-cases

Front 5

When comparing mortality rates in two groups, what is most important variable to adjust for?

Back 5

AGE; because of physiological changes over chronologic time

Front 6

Years of potential life lost; disability-adjusted life years; why devised?

Back 6

YPLL = sigma(65-age)
DAYL = 1 healthy life lost either to death or disability

Both seek to describe amont of potential life lost to disease quantitatively

Front 7

Two advantages of looking at death/discrete outcomes using survival analysis

Back 7

1) Describes time until an event occurs
2) takes censoring into account

Front 8

Relative risk

Back 8

describes relative chance of developing disorder; (incidence in exposed) / (incidence in unexposed)

Front 9

Odds ratio

Back 9

describes odds of having been exposed; (odds of past exposure among diseased) / (odds of past exposure among healthy)

Front 10

Attributable risk

Back 10

(Incidence in exposed) - (incidence in unexposed)

Front 11

Population attributable risk

Back 11

attributable risk * Prevalence of risk factor in population

Front 12

Hazard ratio

Back 12

Relative risk integrated over time; used in survival analysis

Front 13

Central tenet of measurement theory

Back 13

Score = true score + measurement error

Front 14

Relationship between measurement error and reliability

Back 14

high measurement error => low reliability

high measurement error biases towards the null hypothesis (i.e. biases towards finding no difference)

Larger sample sizes are needed when measure is less reliable to achieve adequate statistical power

Front 15

Three kinds of epidemiologic research designs (with respect to time)

Back 15

prospective, retrospective, concurrent

Front 16

What is an observational study design? Three most common obs. study designs

Back 16

As opposed to an experiment, an observational study doesn't involve research intervention

Three kinds: cohort, case-control, ecologic

Front 17

What is the null hypothesis?

Back 17

H0: in population, there is no difference/correlation between groups

Front 18

95% Confidence Interval

Back 18

on repeated sampling, 95% of generated parameters

Front 19

What does this result mean (no exposure is the reference group)?

Relative risk = 1.60 (95% CI: 1.1 – 3.3)

Back 19

??? Those exposed in the sample were 1.6 times as likely to develop the condition of interest than their unexposed counterparts. ???

Front 20

Determinants of statistical power

Back 20

Effect size, prevalence of outcome/exposure (rare exposures/outcomes require larger sample sizes), significance level (p), reliability of measures, number of covariates (measured confounds)

Front 21

Genotype, phenotype, heritability

Back 21

Genotype: DNA sequence of the individual
Phenotype: the outcome under study
Heritability: proportion of variation in a phenotype in the population that is due to variations in genotypes

Front 22

How can sequences of nucleotides in the nucleus of our cells influence our health and behavior?

Back 22

By coding for the synthesis of proteins when the DNA is transcribed

Front 23

What is epistasis?

Back 23

Epistasis = gene x gene interaction

Example: "Polymorphism X increases risk for cancer, but only in the presence of polymorphism Y"

Front 24

What does it mean to say that the great majority of health phenotypes are "complex" or "polygenic"?

Back 24

Complex (polygenic) phenotypes are influenced by many polymorphic genes, each of which has a small effect on the phenotype

Front 25

What is the premise of gene-evironment interaction?

Back 25

The combined influence of genetics and environment is more than additive; that is, the influence of the environment (on phenotype) depends on our genes, and conversely, the influence of our genes depends on the environment

Front 26

What is gene-environment correlation and how does it arise?

Back 26

Our genes and environments are generally not independent of one another, but are correlated

G-E Correlation arises in THREE ways:

1. PASSIVELY: we receive both our genes_and_ our early environments from our parents

2. ACTIVELY: because of our genetically-influenced temperament we 'select' some environments over others

3. EVOCATIVELY: because our genetically-influenced temperament changes our environment

Front 27

Why is gene-environment correlation a major problem for studies of causal environmental risk factors?

Back 27

What appears to be a causal *environmental* effect may actually be a causal effect of correlated genes

Front 28

What are the two biological mechanisms for gene-environment interaction? That is, what are the two biological mechanisms for the reciprocal influence of environment and genetics on phenotype?

Back 28

1. Environmental influences on transcription factors

2. Environmental causes of epigenetic modifications of the genome that increase or decrease gene transcription

Front 29

What does causal inference mean in epidemiology? Why is it so challenging? Why are observational studies not very useful in this regard?

Back 29

Causal inference: identifying a risk factor that /causes/ a health problem

Difficult b/c many criteria for comfortably declaring correlation 'causative'

Observational studies aren't so useful for establishing causation because they aren't highly controlled (and so are more likely to have confounds which are not adjusted for)

Front 30

Minimum requirements for inference of causation from a correlation (ESSENTIAL)

First two: Measurement criteria

Back 30

1. Standardized, reliable, and valid measurement of hypothesized causal risk factor and outcome

2. Measurement of candidate risk factor is not biased by the outcome (e.g. retrospective recall bias in depressive vs non-depressive subjects)

Front 31

Minimum requirements for inference of causation from a correlation

Third requirement: Sampling

Back 31

3. UNBIASED sampling from population of reference!! i.e. Sample must represent group of interest

Front 32

Minimum requirements for inference of causation from a correlation

Fourth and fifth: Timely and 'Forward'

Back 32

4. Temporal precedence of candidate causal risk factor before outcome (i.e. exposure BEFORE onset of outcome)

5. Reverse causation ruled out, controlled, or incorporated in causative model (e.g. reciprocal causation)

Front 33

Minimum requirements for inference of causation from a correlation

Sixth and seventh: Comparable and Consistent

Back 33

6. Full comparability across levels of risk factor (i.e., lack of confounding)

7. Consistency of finding (replication)

Front 34

What is a confound?

Back 34

A variable that is correlated with both the risk factor and the outcome, which may be the true cause of the outcome.

Front 35

What kind of confound is the most serious enemy of causal inference?

Back 35

The UNMEASURED CONFOUND

Front 36

Which of the seven requirements for causal inference is most directly relevant to the statement "correlation doesn't equal causation"

Back 36

6. Full comparability must be achieved on all levels of risk factor (i.e. no confounds)

Front 37

What is a moderator (also called an effect modifier)?

Back 37

A variable that statistically interacts with another predictor variable as it influences the outcome. That is, they combine more than additively.

Front 38

What are two kinds of validity in research designs?

Back 38

Internal validity: the logic of the design is sound and allows causal inference

external validity: the results of the study generalize to the population of reference

Front 39

What two general kinds of designs allow us to move beyond correlation and make causal inferences?

Back 39

Experimental (i.e. Randomized control trial) and quasi-experimental

Front 40

What are the two research designs that allow for causal inference and what are their relative strengths?

Back 40

RCTs have strong internal validity but the results are difficult to extrapolate to the general population.

Quasi-experiments are always flawed internally (i.e. _full_ comparability not achieved) but have good external validity.

Front 41

What is confounded with exposure in a typical cohort or case-control design?

Back 41

EVERYTHING@!!!!

Front 42

How do randomized control trials (RCT) move us beyond detecting correlations to strong causal inference?

Back 42

They achieve _full comparability_ by breaking all genetic and environmental confounds through random assignment.

Front 43

What kind of risk factors do adoption and twin studies disentangle? What questions can they answer?

Back 43

Twin studies separate genetic and environmental risk factors, and they are useful for addressing two main questions:

1. How much genetic and/or environmental causal influences are there on the health phenotype?

2. Are the environmental exposure and outcome related because they share the same genetic influences?

Front 44

How can an environmental exposure and an outcome share the same genetic influences? How is it tested?

Back 44

GENE-ENVIRONMENT correlation!! Passive, active or evocative

The genes that influence an outcome often influence risk environments through passive, active, or evocative gene-environment correlation.

Whether or not an environmental exposure and an outcome share the same genetic influences is tested using multivariate twin modeling.

Front 45

Twin studies divide the variance in a phenotype into what three causal components?

Back 45

Genetic: variation in the phenotype due to genetic polymorphisms

Shared environment: variation in the phenotype due to experiences that are similar by siblings and make them more similar

Nonshared environment: Variation in the phenotype among individuals due to unique experiences that make siblings _less_similar_ ** PLUS MEASUREMENT ERROR**

Front 46

What are three kinds of health services outcome studies?

Back 46

from least to most pragmatic:

Efficacy: Does the program work in an ideal setting?

Effectiveness: Does the program work in the field?
a. Can it be implemented with fidelity?
b. Will the field adopt it? Do patients tolerate it? Does it work with real-world patients?
c. Does it work with real-world patients?

Efficiency: How effective is the program relative to its costs?

Front 47

What is the gold standard for designs for health services outcomes research?

Back 47

Randomized control trials!!!

RCTs are the gold standard for health services outcome studies. ~ But quasi-experiments have a role ~

Front 48

Evaluating health screening programs: do they promote health and well-being?

Back 48

Effective screening requires screens with high: sensitivity, specificity, PPV, and NPV

Front 49

What are sensitivity, specificity, positive predictive value, and negative predictive value?

Back 49

Sensitivity: proportion of true cases the measure identified as positive

Specificity: proportion of true non-cases that the measure identified 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

Front 50

What are the four kinds of bias that complicate the evaluation of screening programs?

Back 50

Any evaluation of a screening program must address and eliminate four potential sources of bias:

1. Selection bias
2. Overdiagnosis bias
3. Prognostic bias
4. "Lead time" bias

Front 51

Selection bias

Back 51

In observational studies, when people can select to be screened or not, healthier, more educated (etc) people voluntarily get screened. Because healthier people live longer, this inflates the apparent benefits of screening.

RCTs effectively control selection biases.

Front 52

Overdiagnosis bias

Back 52

If a screening program leads to many false positive screens, then this overdiagnosis of healthy persons will falsely indicate better outcomes, even if the screen does not lead to effective treatment

Front 53

Prognostic bias

Back 53

People with better long-term prognoses may have longer preclinical phases, and therefore may be more likely to be detected by screens during their _longer_ preclinical phase

Front 54

Lead time bias

Back 54

The lead time created by early diagnosis can give a false impression of longer survival even when death is not actually delayed. Screening is only helpful when it actually results in a gain of life (delay of death)