Clinical Epidemiology

Question 1

Internal validity:

Answer 1

The degree to which the results obtained from a study sample reflect the truth.

Question 2

Truth:

Answer 2

Defined as the results that would be obtained if everyone in the source population participated in the study, and all of the measures were completely accurate.

Question 3

Source Population:

Answer 3

The population from which the sample was drawn

Question 4

External validity:

Answer 4

(sometimes called generalizability) is the degree to which the results of the study are applicable to other (external) populations that are not even being sampled. This is an issue of judgment, and it depends on knowledge of the subject matter.

Question 5

If the characteristics that predict the outcome (disease) or response to treatment are dramatically different between the sample and some external population, then the study has _____ external validity for that specific external population.

Answer 5

If the characteristics that predict the outcome (disease) or response to treatment are dramatically different between the sample and some external population, then the study has LOW external validity for that specific external population.

Question 6

External validity in a clinical trial (an intervention study) is also based on the degree to which the intervention itself can be ________

Answer 6

External validity in a clinical trial (an intervention study) is also based on the degree to which the intervention itself can be REPLICATED
e.g. PTSD treatment with CBT by trained psychiatrists vs. untrained

Question 7

Bias is:

Answer 7

Systematic error; the systematic difference between the observed findings and the truth

Question 8

The greater the bias, the _______ the internal validity

Answer 8

The greater the bias, the LOWER the internal validity

Question 9

Magnitude of Bias:

Answer 9

The magnitude of bias is the size of difference between the observed effect and the truth.

Question 10

Direction of Bias:

Answer 10

Based on the relationship among: a) the truth; b) the observed findings; c) the null value. If the observed findings are farther from the null value than the truth is from the null value, then the bias is away from the null.

Question 11

Confounding Bias:

Answer 11

A distortion in the results due to a mixing of the effects of exposure with the effects of some third, extraneous, factor. Confounding occurs when some third factor is associated with both the exposure and the disease.

Question 12

Confounding is _____ a threat to the validity of observational studies

Answer 12

ALWAYS; because exposures which cause disease do not occur randomly

Question 13

What are examples of observational studies?

Answer 13

• Cohort studies,
• case control studies,
• cross-sectional studies

Question 14

3 ways that confounding can be prevented in the design of a study:

Answer 14

1. Randomization
2. Matching
3. Restriction

Question 15

Randomization:

Answer 15

If an exposure is randomly allocated to study participants, then any potential confounding factor, such as age, ethnicity, etc, should, on average, be distributed equally across the intervention groups. If a factor is equally distributed across intervention groups, it cannot be a confounder.

Question 16

What is the best way of preventing confounding in human research?

Answer 16

Randomization

Question 17

Matching:

Answer 17

If participants are matched on a potential confounding factor, then the distribution will be equal across groups and that factor cannot confound the association.

Question 18

Restriction:

Answer 18

If sex is a potential confounder, and a study is restricted is women, then sex cannot be a confounder.

Question 19

Stratified analysis:

Answer 19

In a stratified analysis, the data are analyzed separately in each level of the confounding variable, and then combined. For instance, in the study of gray hair and heart disease, the data would be analyzed separately in different age groups, and then mathematically combined to produce an overall result which is not confounded.

Question 20

Regression analysis:

Answer 20

A statistical model; produces results in which the potential confounding factors are statistically controlled. One of the major advantages of modeling techniques is that MANY confounders can be controlled at the same time.

Question 21

Selection bias:

Answer 21

A distortion in the results due to a difference between subjects who participated in a study and those who are theoretically eligible but did not participate (i.e., the source population).

Question 22

Selection bias can develop in 3 ways:

Answer 22

1. The way that subjects are selected for inclusion in a study (e.g. coffee related to pancreatic cancer, but all controls were also GI patients who reduced coffee),
2. Refusal of eligibles to participate (e.g. people completing surveys are either thrilled or angry),
3. Withdrawal of loss to follow-up of research subjects in a cohort study or randomized trial

Question 23

What is the difference between confounding and selection bias?

Answer 23

-In confounding, the association between the extraneous factor and the exposure and between the extraneous factor and the outcome exist in the source population. Selection bias is the distortion in the association between an exposure and a disease when the procedures used to select or maintain the sample create differences between the exposed and unexposed groups on factors that are also associated with the disease.

Question 24

Problem with selection bias:

Answer 24

it is frequently difficult to identify the imbalances in risk factors that are created by the processes used to create and maintain the sample, so it is difficult to control for.

Question 25

How is selection bias different from external validity?

Answer 25

In selection bias, the issue is whether the results are distorted due to differences between the groups being compared that arose from the processes used to create the sample. For external validity, the issue is whether there are differences between the results in the sample and the results that we would expect in some external population.

Question 26

Measurement bias:

Answer 26

A distortion in the results (e.g., the relative risk for the association between an exposure and a disease) due to error in measurement of the exposure, or measurement of the disease, or both.

Question 27

Nondifferential measurement bias:

Answer 27

Occurs when errors in the measurement of exposure (second hand smoke) are independent of the disease (sinusitis), or when the errors in the measurement of disease (sinusitis) are independent of exposure. e.g. if people tend to under-report their exposure to second hand smoke, but that people who don’t have sinusitis under-report second hand smoke to the same extent as people who do have sinusitis.

Question 28

In general, nondifferential measurement bias produces bias _____ the null (towards/away from)

Answer 28

In general, nondifferential measurement bias produces bias TOWARDS the null

Question 29

Differential measurement bias:

Answer 29

Occurs when errors in measurement of exposure depend on the disease, or when errors in measurement of disease depend on the exposure. e.g. If persons who had sinusitis were less (or more) likely to under-report exposure to second hand smoke than persons without sinusitis, then there would be differential measurement bias of exposure.

Question 30

Differential bias in measurement of exposure is more likely to occur when:

Answer 30

the disease status of the subject is known prior to measurement of exposure, as in a case control study.

Question 31

Differential bias tends to produce bias _____ the null (towards/away from)

Answer 31

Both towards or away from - the direction of the bias depends on the particulars of the situation

Question 32

Randomized trials are less likely to be plagued by ________ than ______ or _____ because the randomization process creates groups which should be similar for all factors which can affect the outcome.

Answer 32

randomized trials are less likely to be plagued by CONFOUNDING than COHORT STUDIES or CASE CONTROL STUDIES because the randomization process creates groups which should be similar for all factors which can affect the outcome.
(***less likely, not impossible)

Question 33

Cohort Studies:

Answer 33

track people forward in time from exposure to outcome.

Question 34

Case-control studies:

Answer 34

Work in reverse, tracing back from outcome to exposure.

Question 35

Cross-sectional studies:

Answer 35

Like a snapshot, which measures both exposure and outcome at one time point.

Question 36

Ratio:

Answer 36

a value obtained by dividing one number by another. (can either contain the numerator in the denominator or not, depending on study)

Question 37

Rate:

Answer 37

Measures the frequency of an event in a population. Contains numerator in the denominator, and uses time and a multiplier. E.g. 11 cases of TB per 100,000 persons per year

Question 38

Proportion:

Answer 38

Like a rate, must have the numerator contained in the denominator. But is unit-less, and does not have a time component.
E.g. prevalence - 27 of 100 at risk have hay fever.