Ch 3 (assoc & effect) & 4 (interpreting assoc)

Question 1

Identify different measures of association –

Answer 1

prevalence ratio,
risk ratio,
odds ratio,
and incidence rate ratio

• four relative frequency comparisons to help us quantify the association between an exposure and an outcome to detect causal relationships that may help identify effective interventions
• measures strength of association between exposure and outcome
• high risk does not prove causation

Question 2

What are different measures of impact ?

Answer 2

If there is evidence of a causal association, we can assess the impact of an exposure as an absolute difference in frequency between the exposure groups using:

attributable risk,
attributable fraction,
preventable fraction,
and population attributable fraction (To estimate the public health impact of removing an exposure)

Question 3

Define, calculate, and interpret each measure of association and impact

Answer 3

Measures of association:
prevalence ratio, risk ratio, odds ratio, and incidence rate ratio

Measures of impact:
attributable risk, attributable fraction, preventable fraction, and population attributable fraction

Question 4

What do the different RR (relative risks) below tell us?
RR > 1
RR < 1
RR = 1
RR very far from one

Answer 4

RR > 1 – exposed group more likely to have outcome

RR < 1 – exposed group less likely to have outcome

RR = 1 – no difference between exposed and unexposed in getting the outcome

RR very far from one– stronger the association

Question 5

Draw out a standard cross-tabulation 2x2 table.

Answer 5

Reference blue notebook.

Question 6

Formula for– prevalance ratio

Answer 6

prevalance of outcome in exposed divided by prevalance of outcome in unexposed. Or (A/ (A+B))/ (C/ (C+D))

Used for cohort studies

Question 7

How does the prevalance ratio differ from the risk ratio?

Answer 7

Same formula, but the data is different. Prevalance is related to existing cases. Risk is related to new cases.

Question 8

Formula for– odds ratio (using cross-tabulation)

Answer 8

(A/B)/ (C/D) = AD/ BC

Used for case-control series.

Question 9

What is the formula for risk ratio and odds ratio per the 2x2 table? In what scenario will the answers to both equations be very similar.

Answer 9

risk ratio= (A/ (A+B))/ (C/ (C+D))
odds ratio = (A/B)/ (C/D) = AD/ BC

The rarer the outcome, the more likely the two will be the same because the denominator is so large that it’s similar.

Question 10

Define– attributable risk

Define– attributable risk fraction.

Answer 10

Helps determine how much of an outcome is attributed to (explained by) an exposure.

Expresses what fraction of the outcome was due to the exposure in the exposed group, to explain the increased risk of the exposure to the pt.

Question 11

Define– background risk

Answer 11

Frequency of an outcome in the unexposed group.

Question 12

Define– preventable fraction

Formula?

Answer 12

Helps measure the effect of a protective factor, where incidence is greater in unexposed to exposed group.

= (freq in unexposed- “ “ exposed)/ freq in unexposed

OR

= 1- relative risk

Question 13

Define– population attributable fraction (PAF)

Answer 13

proportion (fraction) of the outcome that could be prevented if the exposure could be eliminated from the population

PAF is rarely 100%, because an outcome is usually the result of more than one factor, and there is usually some outcome (background risk) in the unexposed group

Question 14

What is the difference between– ratio, proportion, rate

Answer 14

Ratio– comparison of 2 variables
Ex– 5 women: 7 men

Proportion– the numerator is included in the denominator
Ex– 5/12 are women, or 0.42, or 42%

Rate – is proportion related to time, often expressed in person years in the denominator (expressed in multiples of 10)
Ex– 0.000081 is 8.1 deaths per 100,000 person years

Question 15

What is the difference in the formula for PAF (population attributable fraction) compared to attributable risk fraction?

Answer 15

PAF= (incidence in gen pop - incidence in unexposed)/ incidence in gen pop

ARF= (incidence in exposed- incidence in unexposed)/ incidence in exposed

Question 16

What are the 3 main alternative explanations for an association that isn’t really associated, but just looks like it is.

Answer 16

Bias
Confounding
Chance

Question 17

How do we increase probability of measuring the true value while minimizing the variability?

Answer 17

Increase your sample size or increase number of observation. This will reduce chance or random error.

Question 18

How do we know if the observed relative risk from study sample actually represents the study population? What 2 measures help us determine this?

Answer 18

P- value

confidence interval

Question 19

Define– p-value

A smaller p-value means what?

Answer 19

Measures probability that a measure (RR, frequency, etc.) occurred by chance (as in doesn’t actually exist).

The smaller the P-value, the stronger the evidence that the observed value is real.

Question 20

Define– p-value

A smaller p-value means what?

What does a p-value of 0.05 mean?

Answer 20

Measures probability that a measure (RR, frequency, etc.) occurred by chance (as in doesn’t actually exist).

The smaller the P-value, the stronger the evidence that the observed value is real.

A P-value of 0.05 indicates a 5% probability the observation occurred by chance, and anything below this is generally considered low enough to imply that the result is not due to chance.

Question 21

Define– confidence interval

Answer 21

The confidence interval is a range of values, estimated from a sample, which includes the ‘true’ population value based on a predefined probability (usually 95%, but occasionally 90% or 99%).

Question 22

For relative risk, if the 95% confidence interval does not include 1, what does that mean.

Answer 22

In RR, 1 = no association. So if not included in the 95% CI, then we can say there’s a <5% chance there’s no association.

Long version:
For a relative risk (prevalence ratio, odds ratio, risk ratio, incidence rate ratio) where the value 1.00 represents no association, if the 95% CI does not include 1.00, we can say that the association is significant at the 5% level. This is because there is less than 5% chance that there is no association, so P < 0.05.

Question 23

Define– bias

How does it differ from chance?

Answer 23

Bias is systematic (non-random) errors that lead to deviation from the truth.

Different from chance because chance is related to random errors related to sampling variability.

Question 24

Does sample size affect bias?

Can statistical methods adjust for bias?

How is bias avoided?

Answer 24

No. Sample size does not affect bias.

No. Statistical methods cannot adjust for bias.

Bias is avoided via study design.

Question 25

Bias can be divided into what two types?

Answer 25

Selection bias– systemic difference between characteristics of individuals or from the pop. being sampled

Misclassification bias–

Question 26

What can cause selection bias?

Answer 26

How study pop. or comparison groups are defined. Or related to missing or incomplete information about the groups.

Question 27

What can cause selection bias?

How can it be avoided?

Answer 27

How study pop. or comparison groups are defined. Or related to missing or incomplete information about the groups.

Can be avoided via random selection of study participants or random allocation into study groups.

Question 28

Bias can be divided into what two types?

Answer 28

Selection bias– systemic difference between characteristics of individuals or from the pop. being sampled.

Ex–a study of access to routine health care using household lists drawn up by village elders, recent migrants or marginalized groups may be excluded from the study, making it less representative of the population

information bias– classification of exposure or outcome is inaccurate

Question 29

What can cause selection bias?

How can it be avoided?

Answer 29

• How study pop. or comparison groups are defined.
• Or related to missing or incomplete information about the groups.
• Inclusion/ exclusion criteria.

Can be avoided via random selection of study participants or random allocation into study groups.

Question 30

What can cause information bias?

Answer 30

Can be introduced by the investigators (observer bias), the study participants (responder bias) or measurement tools, such as weighing scales or questionnaires (measurement bias).

Question 31

What is observer bias?

How to avoid it?

Answer 31

Misclassification caused by the observer’s knowledge of the comparison group.

Blinding– which hides the exposure or outcome to the person measuring

To reduce the effect of observer bias, the exposure status (or outcome status in a case-control study) should be concealed from the person measuring the outcome or exposure – this is known as blinding

Question 32

What is responder bias?

How to minimize it?

Answer 32

Responder bias are systemic differences in the information provided. May appear in the form of recall, non-response, or reporting.

ex–study of sexual behaviour and prevalence of sexually transmitted infections (STIs), individuals with a higher number of sexual partners who are at higher risk of STIs may also be more likely to under-report their number of partners due to social stigma (reporting bias), resulting in an underestimate of the association. People who are less sexually active may choose not to answer questions they find embarrassing or intrusive (non-response bias).

Can be minimized by blinding the study participants with a placebo.

• asking questions in different ways
• asking additional questions, not just on the areas of interest
• correlate responses with records
• trialing the questions out before study starts

Question 33

What is non-differential misclassification?

What is the consequence of this?

Answer 33

When two groups are equally likely to be misclassified. May cause the comparison groups to be more similar than they really are.

Ex.– Consider a case-control study of the association of oral contraceptive use with ovarian cancer, using 20 years’ clinic records to determine exposure. It is unlikely that all records of oral contraceptive use over the previous 20 years from a family planning clinic will be traceable. The loss of records is likely to be distributed equally among cases and controls, since record-keeping in family planning clinics is independent of the risk of developing cancer. If the investigators decided to classify all women without a record as unexposed to contraceptives, then the odds of exposure would be underestimated in both cases and controls. This would lead to underestimation of the effect of contraceptives on ovarian cancer.

Question 34

What is non-differential misclassification?

Name an example.

Answer 34

When two groups are equally likely to be misclassified. May cause the comparison groups to be more similar than they really are.

Ex.– Consider a case-control study of the association of oral contraceptive use with ovarian cancer, using 20 years’ clinic records to determine exposure. It is unlikely that all records of oral contraceptive use over the previous 20 years from a family planning clinic will be traceable. The loss of records is likely to be distributed equally among cases and controls, since record-keeping in family planning clinics is independent of the risk of developing cancer. If the investigators decided to classify all women without a record as unexposed to contraceptives, then the odds of exposure would be underestimated in both cases and controls. This would lead to underestimation of the effect of contraceptives on ovarian cancer.

Question 35

What is differential misclassification?

Name an example.

Generally due to what?

Answer 35

Classification of the exposure is dependent on the outcome or vice versa, and is generally due to observer or responder bias.

Ex– in a case-control study of lung cancer in smokers, if the exposure to tobacco smoke was determined by questioning the study participants, there may be a tendency for recall or reporting bias. Cases that are heavy smokers may underestimate their consumption because of the social stigma, leading to an underestimate of the association. Alternatively, given public knowledge of the association, cases may be more likely to blame their disease on tobacco exposure and report past consumption more accurately, while controls may have poorer recollection and underestimate past exposure, leading to an overestimate of the association.

Question 36

Define– confounder

A confounder can or cannot be on the causal pathway between the exposure and outcome?

Answer 36

an apparent association between an exposure and an outcome is distorted by another independent factor (confounder)

A confounding factor must be independently associated with both the exposure and the outcome, and must not be on the causal pathway between the two.

Question 37

Define– confounder

A confounder can or cannot be on the causal pathway between the exposure and outcome?

Answer 37

an apparent association between an exposure and an outcome is distorted by another independent factor (confounder)

A confounding factor must be independently associated with both the exposure and the outcome, and must not be on the causal pathway between the two.

Ex of not a confounder– coronary heart disease is associated with high blood cholesterol, which is a direct result of diet. Blood cholesterol does not provide an alternative explanation for a relationship between diet and coronary heart disease, but is an intermediary between the two, and is therefore a ‘mediator’ rather than a confounder.

Question 38

What are 3 ways to avoid confounding?

Answer 38

1. randomization
2. restriction
3. matching

Question 39

How does randomization avoid confounding? What studies can this be done in?

Answer 39

If sample size is large enough and participants have equal chance of being assigned to exposure and unexposed group, then it helps minimize the known and unknown confounders.

This can only be used in interventional studies.

Question 40

How does restriction avoid confounding?

Answer 40

Only allowing study participants who are similar in relation to the confounder.

ex– gender is a confounder, then study only recruits men, but the results can then not be extrapolated to women

Question 41

How does matching avoid confounding?

What kind of study can this be used in?

Answer 41

Selecting two comparison groups with equal distribution of confounders. On an individual level this is called “pair matching”, on a group level this is called “frequency matching”.

Used in case-control studies.

Question 42

What are 2 ways to control for confounding when analyzing data?

Answer 42

1. stratification– extension of frequency matching.
   ex– effects of marijuana on cognitive function. Age is a confounder. So analyze the data with participants separated by age. but the smaller the groups, increase risk of chance.
2. statistical modeling– allows for adjusting to several confounders simultaneously, using mathematical regressions

Question 43

What is effect modification or effect modifier?

Answer 43

Varying association between exposure and outcome for different subgroups.

Ex– smoking and lung cancer. Gender is an effect modifier because women have higher mortality than men, even when taken into account the amount of smoking.

Question 44

What are the 9 components of Bradford Hill criteria to prove causality?

Answer 44

1. temporality– exposure occurs before outcome
2. strength– stronger the association, less likely due to some other factor
3. consistency– repeatability
4. dose- response – higher dose of exposure, increase risk of outcome
5. plausibility– reasonable biological explanation
6. reversibility– intervention to reduce exposure would also decrease risk of outcome
7. coherence– logical consistency with other information
8. analogy– similarity to other established causal relationships
9. specificity– relationship is specific to the outcome of interest