Standardization and Methods - Health - Lecture Slides, Slides of Public Health

Download Standardization and Methods - Health - Lecture Slides and more Slides Public Health in PDF only on Docsity!

Standardization and Methods

to Control Confounding

Learning Objectives:

1. Understand the “design” and “analysis” methods used to control confounding.

--- Randomization --- Restriction --- Matching --- Stratification --- Multivariate analysis.

2. Understand pros and cons of the methods used to control confounding.

CONFOUNDING - REVIEW

DEFINITION: A third variable (not the exposure or outcome variable of interest) that distorts the observed relationship between the exposure and outcome.

• Confounding is a confusion of effects that is a nuisance and should be controlled for if possible.
• Age is a very common source of confounding.

CONFOUNDING - REVIEW E (^) D

CF

Confounding IS present

E ?CF D

Confounding NOT present

CONFOUNDING

POSITIVE CONFOUNDING:

• The confounding factor produces an estimate that is more extreme (positive or negative) than the true association.

NEGATIVE CONFOUNDING:

• The confounding factor results in an under- estimate of the true association.

CONFOUNDING

METHODS TO CONTROL CONFOUNDING:

DESIGN:

1. Randomization
2. Restriction
3. Matching (Analysis also)

ANALYSIS:

4. Stratification
5. Multivariate Analysis

CONTROL OF CONFOUNDING

2. RESTRICTION (Design):

Definition: Study participation is restricted to individuals who fall within a specified category or categories of the confounder.

Pro: Straightforward, convenient, inexpensive

Con: Sufficiently narrow restriction range may severely reduce the number of eligible participants

Con: If restriction criteria are not sufficiently narrow, possibility of residual confounding exists

CONTROL OF CONFOUNDING

• RESTRICTION (cont.):

Con: Does not permit evaluation of the association between exposure and disease for varying levels of the factor.

Note:Although restriction may limit generalizability, it does not affect the internal validity of any observed association between the groups included in the study.

CONTROL OF CONFOUNDING

• MATCHING (cont.):

Con: Does not control potential confounding by factors other than those matched on

Con: Not needed as much as in the past due to alternative techniques (e.g. multivariate analysis)

CONTROL OF CONFOUNDING

INDICATIONS FOR MATCHING:

• Factors for which there would otherwise be insufficient overlap between study groups (e.g. nominal-level variables such as race).
• Small case series in which baseline characteristics are likely to differ between study groups.
• Most often employed in case-control studies.

CONTROL OF CONFOUNDING

• STRATIFICATION (Analysis):

Definition: Evaluation of the exposure/disease association within homogeneous categories or strata of the confounding variable.

Pro: Intuitively appealing, straightforward, and enhances understanding of intricacies of the data

Con: Impractical for simultaneous control of multiple confounders, especially those with multiple strata

CONTROL OF CONFOUNDING

Hypothesis: Sedentary lifestyle is associated with risk of myocardial infarction (cohort study)

D+^ D- E+^40 80 E-^100 750 970

RR = (40 / 120) / (100 / 850) RR = 2.

It appears that persons with a sedentary lifestyle are 2.83 times more likely to experience myocardial infarction compared to persons without a sedentary lifestyle.

BUT WHAT ABOUT SMOKING?

CONTROL OF CONFOUNDING

In general:

If Strata 1 RR < Crude RR > Strata 2 RR OR If Strata 1 RR > Crude RR < Strata 2 RR

then confounding is present.

CRUDE

RRMI = 2.

STRATA 1

RRNS = 1.

STRATA 2

RRSM = 2.

CONTROL OF CONFOUNDING

Now, the question is:

Should the stratum-specific estimates be combined to obtain an unconfounded (adjusted) estimate of the relationship between sedentary lifestyle and risk of myocardial infarction?

CRUDE

RRMI = 2.

STRATA 1

RRNS = 1.

STRATA 2

RRSM = 2.