HTLH 207 Week 10 - Effect Measure Modification

When may EMM occur?

when there are subgroups of the sample population for whom the exposure has a different effect on the outcome than in other subgroups

Example of a modifier

Smoking is an effect modifier if it modifies the effect of a medication on side effect

What study approach fails when there is EMM? Why?

“One size fits all” approach fails because we need more than one measure of association to describe how the exposure impacts the outcome in different subgroups

Strata

Subgroups of effect measure modification, which are mutually exclusive categories of individuals in the sample population

How must we report measures of association when EMM occurs?

we must report multiple measures of association: one for each relevant subset of the sample population

2 hypothesis structures for EMM

1. The association between an [exposure] and a [health outcome] in a [subgroup] of a [population] is different from the association between the [exposure] and [health outcome] in a [different subgroup] of the [population]

2. The association between an [exposure] and a [health outcome] in a [population] is modified by a [third variable/effect modifier]

Sometimes effect measure modification is referred to as ____________, but these two are distinct concepts

interaction

Effect measure modification

The phenomenon that arises when an exposure has a meaningfully different effect on the outcome in different subgroups of the sample population

Statistical interaction

A statistical test can be used to determine whether the stratum-specific measures of association are statistically different from one another

Biologic interaction

Two (or more) factors must be present for an outcome to occur (they work together to cause the outcome to occur)

Key question to determine if there is EMM in a study

Key question: are the stratified measures of association meaningfully different from one another?

What procedure determines if there is evidence of EMM?

A qualitative assessment of the difference between the stratified measures of association

3 steps of assessing whether there is evidence of EMM

1. Stratify crude data into mutually exclusive groups (strata)

2. Calculate and compare stratum-specific measures of association

3. If the stratum-specific measures of association are meaningfully different from one another, effect measure modification is present - report stratified results

Key question to determine if there is statistical interaction in a study

Are the stratified measures of association statistically significantly different from one another?

What procedure determines if there is evidence of EMM?

A quantitative assessment of the difference between the stratified measures of association

3 required components to assess for evidence of statistical interaction

1. Null hypothesis (H0)

2. Alternative hypothesis (Ha)

3. Statistical test and p-value

Null hypothesis (H0)

There is truly no difference between two measures that are being compared

Alternative hypothesis (Ha)

There is truly a difference between two measures that are being compared

What does the null hypothesis infer about statistical interaction?

There is no evidence of statistically significant interaction between the exposure and potential effect modified

What does the alternative hypothesis infer about statistical interaction?

There is evidence of statistically significant interaction between the exposure and potential effect modifier

What 2 values does each statistical test yield?

1. Test statistic

2. P-value

For analyses stratified by a potential effect modifier, the _____________ test for interaction is one statistical test that can be used to assess whether there is evidence of statistical interaction

Breslow-Day

Alpha level

A threshold that we used to determine whether a result has reached statistical significance - we compare the p-value from our statistical test to alpha

Typical alpha level for significance

0.05

If p-value is less than or equal to 0.05, we ____________________

do not reject the null hypothesis

2 conclusions that can be made from a p-value less than or equal to 0.05

1. No evidence of statistically significant interaction

2. The stratified measures of association are not statistically significantly different from one another

If p-value is greater than 0.05, we ____________________

Reject the null hypothesis

2 conclusions that can be made from a p-value greater than 0.05

1. Evidence of statistically significant interaction

2. The stratified measures of association are statistically significantly different from one another

What is OpenEpi?

Web-based tool that can be used for basic and some intermediate epidemiologic data analyses

What does OpenEpi return when data is entered into a 2×2 table?

Risk- and odd-based measures of association

The risk-based data can be used for a cross-sectional design, so long as _____________________

you apply the correct interpretation to the findings

What do both EMM and statistical interaction evaluate?

Whether there are differences in stratum-specific measures of association

What does EMM consider in its analysis (when compared to statistical interaction)?

EMM considers whether the stratum-specific measures of association are meaningfully different from one another

What does statistical interaction consider in its analysis (when compared to EMM)?

Statistical interaction considers whether the stratum-specific measures of association are statistically significantly different from one another affected by sample size

When should stratified results be reported>

If EMM was central to the research question and/or if stratified measures of association are meaningfully different from one another

What measure should not be used alone to determine when stratified results should be reported?

Statistical tests

What 4 scales can EMM or statistical interaction present on?

1. Multiplicative scale

2. Additive scale

3. Both scales

4. Neither scale

When does EMM (or statistical interaction) occur on the multiplicative scale?

When the stratum-specific ratio measures of association are meaningfully (or statistically significantly) different from one another

When does EMM (or statistical interaction) occur on the additive scale?

When the stratum-specific difference measures of association are meaningfully (or statistically significantly) different from one another

What typically determines the scale used to assess EMM and statistical interaction?

The measures of association that the investigators decide to use prior to the study

many analytic techniques used in epidemiology are inherently ____________, which yield ______ measures of association

multiplicative, ratio

What is the downside of the fact that most analyses are done on the multiplicative scale?

• EMM and statistical interaction on the additive scale are potentially important and often missed

Before you begin your data analysis for EMM and statistical interaction, what should you have an idea of?

Which variables you plan to consider as potential effect modifiers

How to get an idea of which variables you plan to consider as potential effect modifiers

Look to previous literature to generate a list of variables that could plausibly modify the association between your exposure and outcome

“Fishing expedition”

A practice where all possible variables in a dataset are tested to see if they are effect modifiers (should be avoided)

Is EMM relevant for all research questions?

No

What can present as a potential issue with the study when you want to consider a variable as an effect modifier?

The number of individuals in each of the strata of interest may be relatively small

Downside to calculating and comparing stratified measure of association when the sample is too small

You may not be able to detect whether the stratum-specific measures of association are statistically different from one another since statistical testing is heavily impacted by a study’s sample size

Why is EMM an important concept for public health

It helps investigators better understand the nuances of how various exposure impact outcomes differently for different subgroups of a population

3 ways in which EM is important from a public health perspective

1. Identifying high risk groups - who is more at-risk for an adverse outcome when they are exposed?

2. Targeting interventions to particular subgroups of a population

3. Obtaining a better understanding of causation