HTLH 207 Week 3 - Analytic Epidemiologic Study Designs and Measures of Association

Incidence vs. prevalence

• Incidence refers to new health outcomes (requires some degree of follow up)

• Prevalence refers to preexisting health outcomes (mostly cross-sectional design)

Measures of association

Measures of association: quantitative comparisons of measures of frequency (from descriptive EPI) across 2 or more groups

Exposures

May be any factor potentially associated with an outcome, generally be categorized as person, place, or time characteristics

3 examples of exposures

1. Smoking (in relation to lung cancer)

2. Low birth weight (in relation to adulthood diabetes)

3. COVID infection (in relation to brain fog)

Study hypothesis

statement that posits how an exposure is associated with an outcome

Importance of study hypothesis

Foundational to conducting an analytic epidemiological study

General structure for a study hypothesis

In a [population], those who are [exposed] have a <higher/lower/same/different> frequency of an outcome compared to those who are unexposed

Population of interest (in the context of a study hypothesis)

must consider the population that you hope to apply your study results to (i.e. make inference about) and select study participants from this group

Health outcome (in the context of a study hypothesis)

clearly defines how those with and without the outcome will be classified

Exposure of interest (in the context of a study hypothesis)

clearly defined those with and without the exposure will be classified - may need to account for the dose and duration

What is the unexposed group called in public health studies?

Referent/reference/comparison group

2 ways of comparing measures of frequency

• Relative/ratio: compared by division

• Absolute: comparison by subtracting

What determines what type of comparison between measures of frequency can be used?

Study design

What question should be asked in relation to the direction of a hypothesized association? Then, what should the next step be?

• Is the frequency of the outcome different or not different across exposure groups?

• If different, can decide if you wish to specify the direction or frequency of the outcome

Exploratory analytic epidemiological studies

may be used to evaluate association between several different exposures and outcomes

2 uses of exploratory analytic epidemiological studies

1. Useful if little is known about the etiology of an outcome

2. May be used to generate hypotheses for future studies

What caution must be taken with exploratory analytic epidemiology studies?

it is important to carefully select exposures and outcomes that could plausibly be related

Structure of exploratory analytic epidemiology study question

In a [population], do those who are [exposed] have a different frequency of an outcome compared to those who are [not exposed]?

2 classifications of analytic epidemiology studies

1. Experimental

2. Observational

Experimental study

A study in which the investigator aligns the exposure

How are experimental studies often achieved?

Randomization (i.e. random assignment)

2 examples of exposures that cannot be assigned

1. Smoking

2. Neighborhood

Observational studies

Studies in which the investigator observes the exposure status that has already occurred or will occur anyway

Unit of analysis

the level at which study data are analyzed

2 divisions of unit of analysis

1. Individual

2. Group

Individual unit of analysis

investigators have information about the exposure and outcome status for each individual in the study

Group unit of analysis

investigators only have access to aggregate data, such as the average frequency of exposure and/or outcome across different groups

Function of group unit of analysis

May be useful to compare the effect of an exposure across defined groups (i.e. states or countries)

Main type of experimental design used in public health research

Controlled trials

4 types of observational designs

1. Cohort studies

2. Case control

3. Cross-sectional

4. Ecologic

Controlled trial

study sample is enrolled from a population that is at-risk of developing the outcome of interest

Steps of a controlled trial

• Study participants are assigned to an exposure group by the study investigators, often via randomization

• An intervention of interest may be compared to a placebo, which is an agent that appears identical to the intervention

• Participants are followed for a specific time period to measure new (incident) cases of the outcome

How is the association between exposure and outcome quantified in a controlled trial?

Using risk-and rate-based measures of association

Cohort design

the study sample is enrolled from a population that is at-risk of developing the outcome of interest

Steps of a cohort design

• Study participants are then classified by their exposure status, which is observed/recorded (not assigned) by the study investigators

• Participants are followed for a specific time period to measure new (incident) cases of the outcome

How is the association between exposure quantified in a cohort design?

Using risk- and rate-based measures of association

Case control design

Investigators enroll a group of individuals with the outcome of interest (cases), along with a suitable comparison group that does not have the outcome (controls)

Steps of a case control design

• Participants’ historical exposure status is ascertained using tools such as questionnaires or medical records

• Investigators explore whether the cases were more (or less) likely to have certain exposures than the controls

How is the association between exposure quantified in a case control design?

Using the exposure odds ratio (EOR)

Cross-sectional design

investigators enroll a population of interest and measure their exposure and outcome statuses simultaneously

Does follow-up occur in cross sectional studies?

No

How is outcome frequency reported in cross-sectional studies?

Prevalence

How is the association between exposure quantified in a cross-sectional design?

Using prevalence-based measuers of association

Ecologic study design

• exposure and outcome information are reported in aggregate (for distinct groups, not individuals)

• Evaluation of exposures that can only be measured at the population (i.e. policy changes)

Benefit of ecologic study design

Exploration of existing aggregate data to generate findings more quickly and less expensively than individual-level and data collection

Ecologic fallacy

findings at the group level are presumed to apply to individuals

Traditional hierarchy of analytic epidemiology study designs (worst to best)

1. Ecologic

2. Cross-sectional

3. Case-control

4. Cohort

5. Randomized control trials

Modern way to think about which epidemiology study design is best

well-conducted study using the design that appropriately aligns with the hypothesis of interest, as well as ethical and practical considerations

How are dichotomous exposure and outcomes organized?

2×2 tables

Which measure(s) of association can be calculated from the following table?

Rate-based measures of association

Which measure(s) of association can be calculated from the following table?

Risk/prevalence-based measures of association

Label the following table

1. Outcome (O+)

2. No outcome (O-)

3. Total

4. Exposed (E+)

5. A

6. C

7. N1 (A+C)

8. Unexposed (E-)

9. B

10. D

11. N0 (B+D)

12. Total

13. M1 (A+B)

14. M0 (C+D)

15. N

5 marginal totals of the 2×2 table

1. N

2. N1

3. N0

4. M1

5. M0

N (2×2 table)

total number of individuals in the study population

N1 (2×2 table)

total number of individuals exposed

N0 (2×2 table)

tota number of individuals who are unexposed

M1 (2×2 table)

total number of individuals with the outcome

M0 (2×2 table)

Total number of individuals without the outcome

A (2×2 table)

individuals who are exposed and have the outcome

B (2×2 table)

individuals who are unexposed and have the outcome

C (2×2 table)

individuals who are exposed and do not have the outcome

D (2×2 table)

individuals who are unexposed and do not have the outcome

Measures of association

derived by comparing measures of frequency for two groups

2 ways that measures of association are typically calculated

1. Taking the ratio of two measures of frequency

2. Taking the difference of two measures of frequency

Null value

the value that a measure of association takes on when there is no association between the exposure and outcome

What will the frequency of the outcome in the exposed and unexposed groups if there is no association between an exposure and outcome

The frequency of the outcome in the exposed group = frequency of the outcome in the unexposed group

Null value for all ratio measures of association

1

Null value for all difference measures of association

The null value for all difference measures of association is zero

Goal of calculating measures of association

Ultimately we will calculate a measure of association using our study data, and our goal is to determine whether it is different from its null value

What can we determine if our measures of association is significantly different from our null value?

This suggests that there is an association between exposure and outcome

Direction of a measure of association meaning

Refers to where it lies relate to its null value

Measure of association falls below the null meaning

Negative association - exposure is associated with a decreased outcome frequency

Measure of association falls above the null meaning

Positive association - exposure is associated with an increased outcome frequency

Magnitude of a measure of association meaning

Refers to the strength of association

What is teh global cutoff for classifying the strength of association?

There is no global cutoff

Whether an association is considered strong depends on the ___________________

research question

Risk ratio

a measure of the strength of the relative risk of an outcome between an exposed and unexposed group during a specific follow-up period

Risk ratio calculation

Risk in the exposed/risk in the unexposed = (A/N1)/(B/N0)

Risk difference

a measure of the absolute difference in the risk of an outcome between an exposed and unexposed group during a specified follow-up period

Risk difference calculation

Risk difference = risk in the exposed - risk in the unexposed = A/N1 - B/N0

Risk-based measures of association are appropriate for what 2 applications?

1. Some controlled trials

2. Some cohort studies

When might it not be appropriate to calculate risk-based measures

If follow-up is not uniform for all study participants

Interpretations of risk-based measures of association must include what information?

Information about hte length of the follow-up period

Prevalence ratio

a measure of the strength of the relative prevalence of an outcome between an exposed and unexposed group

Prevalence ratio calculation

Prevalence ratio = prevalence in the exposed/prevalence in the unexposed = (A/N1)/(B/N0)

Prevalence difference

a measure of the absolute difference in the prevalence of an outcome between an exposed and unexposed group

Prevalence difference calculation

Prevalence difference = prevalence in the exposed - prevalence in the unexposed = A/N1 - B/N0

Prevalence-based measures of association are appropriate for what application? Why?

Appropriate for the cross-sectional design since the observed outcomes are existing, rather than the newly-occurring

Exposure odds ratio (EOR)

the only measure of association that is appropriate to calculate for case-control studies

What measure does the EOR use?

The odds

EOR calculation

odds of exposure among the cases/odds of exposure among the controls = (A/B)/(C/D) = (AxD)/(BxC)

What must a table used to organize data to allow for calculating rates and rate-based measures of association incorporate?

Person-time

Label the following table

1. Outcome (+)

2. Person-time

3. Exposed (E+)

4. A

5. PT1

6. Unexposed (E-)

7. B

8. PT0

9. Total

10. M1 (A+B)

11. PT (PT1 +PT0)

PT1 (table)

total person-time contributed by all participants who were exposed

PT0 (table)

total person-time contributed by all participants who were unexposed

PT (table)

person-time contributed by the entire study population

Rate ratio

a measure of the strength of the relative risk of an outcome between an exposed and unexposed group

Rate ratio calculation

Rate ratio = rate in the exposed/rate in the unexposed = (A/PT1)/(B/PT0)

Rate difference

a measure of the absolute difference in the rate of an outcome between an exposed and unexposed group

Rate difference calculation

Rate difference = rate in the exposed - rate in the unexposed = A/PT1 - B/PT0