OSU Epi: Confounding

what are the conditions of a confounder

• associated with the disease

• associated with the exposure

• not on causal pathway between exposure and disease

what type of confounding is this

bias is up and away from the null

what type of confounding is this

bias is down and toward the null

what type of bias is this

bias is down and away the null

what type of bias is this

bias is up and towards the null

what is the null value for risk ratios vs disk differences

• risk/rate ratio

  • null = 1.0

• risk/rate difference

  • null = 0.0

what are the steps in assessing a confounding variable

• create a criteria (ex: magnitude of confounding is 10%)

• calculate crude estimate

• calculate adjusted estimate

• calculate magnitude and change

• compare estimates to criteria

how do you calculate magnitude of confounding

(RR crude - RR adjusted)/ RR crude

how do you calculate change of confounding

• ln (RR Adjusted / RR Crude)

what type of confounding is this

down and across confounding

why is it important to control for confounding

• alters causal pathway

  • makes association appear

  • makes association disappear

  • changes strength of association

how can you control for confounders in the study design

• randomization

• matching

• restriction

how can you control for confounders in analysis

• restriction

• stratified analysis

• multivariable analysis

describe using randomization as a way to control for confounders

• used in interventional studies to control for known and unknown confounders

  • strengths

    • accounts for unmeasured confounders

  • limitations

    • severely limits study design

describe using matching as a way to control for confounders

• exposed and unexposed groups have equal distribution of confounding variable (ex: gender)

  • strengths

    • easy to implement

  • limitations

    • cannot study this confounder

    • may have residual confounding

describe restriction as a way to control for confounding

• restricting eligibility of recruited participants (restricting smokers)

  • strengths

    • fairly easy to implement

  • weakness

    • loss of generalizability

    • loss of sample size

describe adjustment in analysis as a way to control for confounding

• estimate association by controlling for confounding

  • (standardization, mantel-haenszel adjustment, MV regression)

    • strengths

      • flexible and straightforward

    • limitations

      • residual confounding

describe direct standardization

• applying stratum-specific rates from study population to national population to get expected outcome

describe indirect standardization

• applying stratum-specific rates from national population to study population stratum sizes

describe mantel-haenszel adjustment

• calculates odds ratio through different strata

what is effect measure modification

• association between an exposure and outcome differs across a third variable

  • not a confounding variable

  • reveals that an effect is stronger/weaker in specific subgroups

describe how confounding variables effect strata

• both strata are similar to one another

• strata specific estimates are different from weighted average

describe how effect measure modification variables effect strata

• strata estimates are different from one another

• strata specific estimates are similar to weighted average

solve this

!