CLASS SESSION 23 EFFECT MODIFICATION Epidemiology 503 Section

CLASS SESSION 23: EFFECT MODIFICATION Epidemiology 503, Section 2

The Big Picture

Last Class - Confounding: Situation in which a non-causal association between a given exposure and an outcome is observed as a result of the influence of Exposure Outcome a third variable Third Variable Effect within strata will be the same but will be different from the crude To control for confounding: � � To prevent confounding by known factors: Match, randomize or restrict in design To adjust for known confounding factors: Stratify results or adjust in regression model

Effect Modification STRENGTH or DIRECTION of the association between an exposure and outcome differs according to third variable (effect modifier) Exposure Outcome Third Variable Stratify by levels of potential effect modifier (third variable)

Exposure Outcome Group 1 Group 2 STRENGTH Crude Exposure Group 1 Outcome Exposure Group 2 Exposure DIRECTION

Example: Radon and Lung Cancer Radon is an odorless gas that is the 2 nd cause of lung cancer after smoking (5 yr survival ~10%). EPA states radon in homes causes more deaths than fires, drownings, and airplane crashes

Smoking is a Strong Modifier of Radon, Lung Cancer Relationship

Shared Biology Makes Smokers More Susceptible to Effects of Radon Level 20 p. Ci/L 10 p. Ci/L 8 p. Ci/L 4 p. Ci/L 2 p. Ci/L 1. 3 p. Ci/L 0. 4 p. Ci/L Smokers Non-Smokers Expected lung cancer (per 1, 000) Risk compares to 260 250 * risk of drowning 36 35 * risk of drowning 150 200 * risk of dying in a fire 18 20 * risk of dying in a fire 120 30 * risk of dying in a fall 15 4 * risk of dying from a fall 62 5 * risk of dying in a car crash 7 Risk of dying in a car crash 32 6 * risk of dying of poison 4 Risk of dying from poison 20 Average indoor radon 2 Average indoor radon 3 Average outdoor radon

Steps for Identifying Effect Modification 1. 2. 3. Stratify by levels of potential effect modifier Compute stratum-specific estimates AND confidence intervals If confidence intervals do not overlap, then effect modification is present so stratified results

Example – Finals Week & Ice Cream Would you expect this relationship to be the same among everyone? What might modify this relationship?

Example – Finals Week & Ice Cream Would you expect this relationship to be the same among everyone? What might modify this relationship?

Is Lactose Tolerance an Effect Modifier? Crude Lactose Tolerant RR 2. 5 3. 7 95% CI 1. 7 to 3. 7 2. 5 to 5. 5 Lactose Intolerant 1. 2 1. 0 to 1. 4

Why Not Pool Estimates Like With Confounding? Effect modification is a true biological or physical phenomenon and is not a form of bias Is not to be adjusted away or controlled for but rather an inherent feature of the way the world works Report associations for different groups, may even want to target specific groups separately in design

Confounding and Effect Modification

Can You Think of Something that Could Confound This Relationship?