EPID 503 Effect Modification Previously Discussed Confounded Associations

EPID 503 Effect Modification

Previously Discussed Confounded Associations

Controlling for Confounding • Confounders can be accounted for in design or analysis phase ▫ Adjust for known confounding factors: Stratify results or adjust in regression model ▫ Prevent confounding by known factors: Match, restrict or randomize in design

Estimates Within Strata Will be the Same Non-Smokers D+ D- Total E+ 160 640 800 E+ 5 95 100 E- 40 160 200 E- 45 855 900 200 800 1000 50 950 1000 OR Smoker Strata= (160*160)/(40*640) = 1. 0 Crude : 2. 8 OR Non-Smoker Strata= (5*855)/(45*95) = 1. 0

Effect Modification • Biological synergism or antagonism results in different association between exposure and disease based on a third variable (effect modifier) Exposure Group 2 Exposure STRENGTH Group 1 Outcome Group 1 Group 2 Exposure DIRECTION

An Example: Finals Week & Ice Cream • Study Objective:

Would You Expect This Relationship to be the Same Among Everyone? • Study Objective: • What might modify this relationship?

Is Lactose Tolerance an Effect Modifier? Crude RR Lactose Tolerant RR 2. 5 3. 7 1. 7 to 3. 7 2. 5 to 5. 5 Lactose Intolerant RR 1. 2 1. 0 to 1. 4

Key Steps for Identifying Effect Modification • Stratify on levels of potential effect modifier • Estimate the measure of association AND confidence intervals • If confidence intervals do not overlap, then effect modification is present so stratified results must be reported

Are stratum-specific OR/RRs the same? Yes No EFFECT MODIFICATION Crude OR/RR = stratum-specific OR/RR? No Report stratum-specific OR/RR Yes CONFOUNDING NO CONFOUNDING OR Report stratum-specific OR/RR EFFECT MODIFICATION Report crude OR/RR

Are stratum-specific OR/RRs the same? Yes No EFFECT MODIFICATION Crude OR/RR = stratum-specific OR/RR? No Report stratum-specific OR/RR Yes Why do we not check for confounding here? CONFOUNDING NO CONFOUNDING OR Report summary measure (MH OR/RR) EFFECT MODIFICATION Report crude OR/RR

Confounding or Interaction? Data Set Stratum 1 RR Stratum 2 RR Crude RR 1 1. 00 4. 11 2 1. 02 3. 50 2. 02 3 0. 03 3. 50 1. 70 4 4. 00 4. 10 4. 20 Interaction? Confounding?

Confounding or Interaction? Data Set Stratum 1 RR Stratum 2 RR Crude RR Interaction? Confounding? 1 1. 00 4. 11 No Yes 2 1. 02 3. 50 2. 02 3 0. 03 3. 50 1. 70 4 4. 00 4. 10 4. 20 Stratum specific same as each other, but different from the crude remember 10% rule

Confounding or Interaction? Data Set Stratum 1 RR Stratum 2 RR Crude RR 1 1. 00 4. 11 2 1. 02 3. 50 2. 02 3 0. 03 3. 50 1. 70 4 4. 00 4. 10 4. 20 Interaction? Confounding? Yes Don’t evaluate Stratum specific different from each other if given confidence intervals, they need to not overlap.

Confounding or Interaction? Data Set Stratum 1 RR Stratum 2 RR Crude RR 1 1. 00 4. 11 2 1. 02 3. 50 2. 02 3 0. 03 3. 50 1. 70 4 4. 00 4. 10 4. 20 Interaction? Confounding? Yes Don’t evaluate Stratum specific different from each other if given confidence intervals, they need to not overlap.

Confounding or Interaction? Data Set Stratum 1 RR Stratum 2 RR Crude RR 1 1. 00 4. 11 2 1. 02 3. 50 2. 02 3 0. 03 3. 50 1. 70 4 4. 00 4. 10 4. 20 Interaction? Confounding? No Yes Stratum specific same as each other, but different from the crude remember 10% rule