HLTH 207 Week 6 - Case Control Designs

Does the entire population need to be studied in a case control study?

Not the entire population of interest needs to be studied in a case design; a sample can simply be taken

3 parts of a case control design

1. Identify a group of individuals with the outcome of interest (cases) along with a suitable comparison group of individuals without the outcome (controls)

2. Identify history of exposures of all study participants

3. Determine which exposures are more (or less) common among cases than controls

Outcome frequency measure for case control designs

Cannot calculate

Measure of association calculated for case control designs

Exposure odds ratio (EOR)

4 basic steps of designing a case control study

1. State the research question

2. Design the case-control study

3. Conduct the case-control study

4. Analyze and report data

3 aspects of stating the research question

1. State the study hypothesis(es) based on the research question

2. Define outcome(s) with case definitions

3. Define the exposure and reference group

4 aspects of designing the case control study

1. Determine whether you will conduct a primary or secondary-base case-control study

2. Identify the sample population from the source population

3. Decide on a strategy to identify cases and controls

4. Decide whether you want to conduct matching

Aspect of conducting the case-control study

Select cases and controls, obtain informed consent, and measure the past exposure status of the participants

3 aspects of analyzing and reporting data

1. Calculate measures of association (and conduct a matched analysis, if appropriate)

2. Report findings using the STROBE reporting guidance

3. Consider the strengths and limitations of the study

When are case-control studies used?

when investigators are interested in studying a particular health outcome and aim to identify exposures associated with that outcome

5 questions that case control studies should address?

1. What is the population of interest?

2. What is the health outcome of interest and how is it defined?

3. What is the exposure of interest and how is it defined?

4. What measures of frequency are being compared?

5. What is the direction of the hypothesized association

Functions of cases in a case-control study

provide an estimate of the frequency of the exposure in people with the outcome in the source population

Controls

selected from the same source population as the cases, and would have been identified as cases had they acquired the outcome

Function of controls in a case-control study

provide an estimate of the frequency of the exposure in the source population

What is the most important aspect to define in a case-control study?

Source population

Source population

refers to the underlying population that gave rise to the cases

Case control designs require a ____________ definition to determine who is eligible to be considered a case

rigorous case definition

2 qualities of exposures used in a case-control study

1. Typical for several different exposures to be investigated in case-control studies

2. Exposures should be factors that are plausibly related to the outcome of interest

“Gold standard” meaning

The term “gold standard” is used to describe a measurement method that is widely accepted as the best or most reliable way to measure an exposure or outcome

2 classifications of case-control studies and how they are categorized

Case-control studies are classified as primary- or secondary-base depending on how the source population is identified and used for participant recruitment

Primary base case-control studies

 with a well-defined source population, and then cases and controls are selected from that population

Function of primary base case control study

This strategy ensures cases and controls are sampled from the same source population and is therefore the preferred approach

Secondary base case control study

investigators begin by identifying a feasible mechanism for selecting cases from the source population

When should a secondary base case-control study be used?

Sometimes it may not be feasible to clearly define and sample from the source population that gave rise to the cases

Since there is not a well-defined source population in secondary base case control studies, what must investigators do?

Attempt to identify an appropriate comparison group of individuals who would have been identified as cases had they developed the outcome of interest

Label the following table for case-control studies

1. Cases

2. Controls

3. Exposed (E+)

4. A

5. C

6. Unexposed (E-)

7. B

8. D

9. Total

10. M1 (A+B)

11. M2 (C+D)

Measure of outcome frequency used in case-control designs

None

Why is there not a measure of outcome frequency used for case-control designs?

• Investigators decide how many cases and controls to enroll in a case-control study

• Thus, there is no meaningful measure that describes the frequency of an outcome for this design

Measure of exposure frequency used in case control design

Odds

Function of odds (measure of exposure frequency)

We typically do not report the frequency of the exposure on its own - however, we use this information to help calculate the odds ratio, which is the appropriate measure of association for case-control designs

Odds of exposure among cases calculation

probability of exposure among cases/probability of non-exposure among cases = (A/M1)/(B/M1) = A/B

Odds of exposure among controls calculation

probability of exposure among controls/probability of non-exposure among controls = (C/M0)/(D/M0) = C/D

Measure of association used in case control designs

Exposure odds ratio (EOR)

What may EOR also be referred to as?

Cross-product

EOR calculation

EOR = odds of exposure among cases/odds or exposure among controls = (A/B)/(C/D) = AD/BC → may be referred to as cross-product

Structure for interpreting an exposure odds ratio

The odds of exposure among cases is [EOR] times the odds of exposure among controls

Purpose of a case control study

ascertain whether an exposure is more or less common in those with the outcome compared to the source population, therefore the interpretation should only reference the exposed condition among cases and controls

Null value for EOR

1

The case-control design is useful because it allows investigators to evaluate the association between one or more exposure and a health outcome without requiring what 2 pieces of information?

1. Exposure data on everyone in the source population

2. A follow-up period

Odds ratio in other study designs

• The odds ratio is the only measure of association that can be calculated in a case-control study

• However, odds ratios are sometimes calculated for in other study designs

The cross-product/EOR can be calculated in any study with __________________

traditional 2×2 data

What is the cross-product called in a cohort study?

risk odds ratio

What is the cross-product called in a cross-sectional study?

prevalence odds ratio

Appropriate use of odds ratio vs. risk/rate-based measures in other types of studies

It is best to report risk-, rate- or prevalence-based measures of association for these other study designs, and seeing that investigators have reported an odds ratio isn’t enough to determine that they have conducted a case-control study

Method 1: Sampling after enumeration of all cases

Begins by identifying all cases occurring in the source population and including either all of these individuals as cases or a random sample of these individuals as cases in the study

When is method 1 (enumeration of all cases) common?

when the health outcome of interest is systematically collected or reportable (i.e. outcome data are available in health registries, vital records, or previous cohort studies)

What may occur in method 1 (enumeration of all cases) when it is not possible to enroll all of the cases that were identified?

• In this situation, a random sample of cases may be enrolled

• In a random sample, each case has the same probability of being selected and enrolled

What should the frequency of exposure be if a random sample of cases is enrolled in a case control design?

In this situation, the frequency of the exposure in the random sample of cases should be the same as the frequency of exposure among all those with the outcome in the source population

If it is not possible to identify all individuals with the health outcome of interest, what is a common approach?

One common approach is to identify patients receiving care within a particular healthcare facility as cases

2 disadvantages of not identifying all individuals with the health outcome of interest in the source population

• sampled cases may be different than those who would have been selected if there was a complete listing of cases available

• May miss some cases in the source population

Selection bias in case-control designs

if investigators only enumerated and sampled cases from one of these healthcare facilities, the estimate of the frequency of exposure among cases would not be the same as the source population

When selecting cases for case-control study, investigators must also consider whether they are selecting _________ or __________ cases of the outcome

incident, prevalent

Is it generally preferable to select and enroll incident or prevalent cases in case control studies? Why?

• It is generally preferable to select and enroll incident cases because the exposure could be associated with survival among those with the outcome

• When the exposure is associated with survival, unexposed cases live longer and would be overrepresented in a sample of prevalent cases

Why might prevalent cases not provide an accurate measurement compared to incident cases in a case-control design?

If exposure is associated with survival, prevalent cases may not provide an accurate estimate of the frequency of exposure among those with the outcome in the source population

Controls

group of individuals who should be selected from the same source population as the cases, and who should provide an estimate of the frequency of exposure(s) in the source population

3 main sources from which to select controls

1. Population-based controls

2. Hospital and clinic based controls

3. Relatives, neighbors, or friends of cases

Population-based controls

 occurs in primary base studies and is the most robust approach to identify controls

Population-based controls method

Investigators have access to complete list of the source population, which is often defined by a geographic area

Advantage of population-based controls

frequency of the exposure(s) among controls randomly selected from a population-based roster should be similar to that of the entire source population

2 disadvantages of population-based controls

1. May not always be feasible

2. Individuals selected from a population-based roster may be less likely to agree to participate in a research study than those identified via other approaches

Hospital- and clinic-based controls

When cases are identified at a healthcare facility, investigators may enroll hospital- or clinic-based controls

Source population in hospital- and clinic-based controls

The source population is conceptualized as a group of individuals who would be treated at a particular facility if they were to develop the outcome of interest

Control group in hospital- and clinic-based controls

It is assumed that the control group would receive treatment at the same facility as the cases if they actually had the outcome of interest

2 advantages of hospital- and clinic-based controls

1. May be easier to find, more motivated to participate than controls recruited using other methods

2. May be more similar to cases than population-based controls on confounders

Limitation of hospital- and clinic-based controls

These controls may not accurately represent the frequency of exposure in the source population

Relative, neighbors, or friends of cases method

In instances where population-, hospital-, or clinic-based controls cannot be identified, relative, neighbors, or friends of the cases may be selected as a control group

3 advantages of relative, neighbors, or friends of cases method

1. Do not have to enumerate the source population

2. May be more motivated to participate given the social connection to cases

3. May be more similar to the cases on confounders

2 limitations of relative, neighbors, or friends of cases method

1. Reliance on cases to provide contact information for potential controls

2. Frequency of exposure(s) among these controls group may be more similar to the cases than to the source population due to proximity

Defining feature of a cohort study

a group of individuals who are at-risk for the outcome of interest are followed over time to see who develops the outcome of interest

What changes if historical data is used to conduct a cohort study?

the underlying structure is the same

Outcome frequency in cohort studies

Outcome frequency can be measured in all types of cohort studies

3 characteristics of case-control study that is different from cohort study

1. Does not follow a cohort of at-risk individuals over time

2. First classified by outcome status

3. Outcome frequency cannot be estimated using case-control data

Matching

occurs when investigators select controls who have specific attributes that correspond to those of the cases

Requirements of matching attrivutes

must be factors that are likely to be confounder of the relationship between exposure and outcome

Confounders

factors that are associated with both the exposure and the outcome and may partially or completely explain an observed exposure-outcome association

What happens (in terms of data analysis) if confounders are ignored?

EOR will be biased or incorrect

What happens in terms of matching once a case is identified?

Once a case is identified, investigators may seek to identify a control or set of controls that has (have) the same matching factors characteristics (i.e. age) as the case

4 types of matching

1. Pair-matching

2. Exact matching

3. Category matching

4. Caliper matching

Pair-matching

occurs when one control is matched to each case while n-to-one matching means more than one control is matched to each case, where n is the number of controls

Exact matching

matching on the exact same value of the matching factors; however, this may be challenging for certain factors

Example of exact matching

exact matching on age and sex means a 36-year-old female case is matched to a 36-year-old female control

Category matching

mutually exclusive categories for the matching factor are identified and a control is selected that falls in the same category as the case

Caliper matching

controls with a value within a specified range of the case are identified

Advantage of matching in case-control study

Control for the confounding by the matched factors

2 limitations of matching in case-control study

1. More difficult to identify controls for inclusion in the study

2. Can be time-consuming and costly

What happens to a factor once it is matched on?

Once a particular factor is matched on, it can no longer be considered as a potential exposure because matching has forced the cases and controls to have the same frequency of the matched factor

How is the data layout for a pair-matched case control study different from an unmatched case control study?

each cell represents the number of pairs with a given exposure pattern

Label the following case control 2×2 table

1. Exposure status of the case

2. Exposure status of the control

3. Exposed (E+)

4. Unexposed (E-)

5. Exposed (E+)

6. Unexposed (E-)

7. W

8. X

9. Y

10. Z

W (on case-control 2×2 table)

both the case and control are exposed

X (on case-control 2×2 table)

the case is exposed and the control is unexposed

Y (on case-control 2×2 table)

the case is unexposed and the control is exposed

Z (on case-control 2×2 table)

both the case and control are unexposed

2 exposure patterns/statuses

1. Concordant

2. Discordant

Concordant exposure pattern

the exposure status is the same for the case and control (W and Z)

Discordant exposure pattern

the exposure status is different for the case and control (X and Y)

What can we evaluate when the exposure status of the case and control are concordant?

we don’t gain any information to evaluate the relationship between the exposure and outcome

What can we evaluate when the exposure status of the case and control are discordant?

e can evaluate whether there is an association between exposure and outcome

Pair-matched odds ratio (mOR)

ratio of discordant pairs: X/Y

Pair-matched odds ratio (mOR) null value

1

What can we conclude if the pair-matched odds ratio is greater than 1?

there are more pairs where the case is exposed and the control is unexposed