Confounding

1
Confounding

• Dr. Sunita Dodani
• Assistant Professor
• Family Medicine, CHS
• The Aga Khan University
• Pakistan

2
Learning objectives

• To understand the role of confounders in a study
• To learn relationship between an exposure,
  disease and potential confounding factors
• To understand difference between confounding and
  effect modification
• To learn methods to control confounding in study
  designs and in data analysis

3
Performance objectives

• After this lecture the student will be able to
• Differentiate the role of a confounder and a
  exposure in a study
• Use methods to control effects of confounders in
  research projects

4
Confounding

• Confounding occurs when two factors are
  associated with each other, or travel together
  and the effect of one is confused with or
  distorted by the effect of the other.
• A confounder is a variable which is associated
  with the exposure, and independent of that
  exposure is a risk factor of the disease

5
Confounding

• Examples
• Study one found an association with smoking and
  loss of hairs.
• The study was confounded by age
• Study two found improved outcome for maternal
  centers when compared to hospitals
• Study might be confounded by highly motivated
  volunteers that may have selected these centers
  as an option

6
Confounding

• Confounders are generally correlates of other
  causal factors
• HSV-2
• Sexual activity
• HPV Cervical cancer
• A confounder cannot be an intermediate link
• in the causal pathway between exposure and
  disease

7
Confounding

• Delete sample documenticons and replace with
  working document icons as follows
• From Insert Menu, select Object...
• Click Create from File
• Locate File name in File box
• Make sure Display as Icon is checked
• Click OK
• Select icon
• From Slide Show Menu, Select Action Settings
• Click Object Action and select Edit
• Click OK

• In other words, confounding is a variable that is
  associated with the predictor variable and is a
  cause of the outcome variable
• Aside from bias, confounding is often the likely
  alternative explanation to cause-effect and the
  most important one to try to rule out.
• In contrast to bias, confounding can be
  controlled at several levels of a study

8
Effect modification

• Effect modification is a type of interaction
• When the strength of the relationship between two
  variables is different with respect to some third
  variable called effect modifier.

9
Effect modification

• EXAMPLES 1
• relationship between dose of thiazide and risk
  of sudden death.addition of K sparing drug
  modifies the effect at several doses.
• effect modifier.. K sparing drug

10
Effect modification

• Example 2
• People who take monoamine oxidase inhibitors
  (MAOI) are at risk of stroke if they eat certain
  foods such as cheese.
• effect modifier. MAOI
• MAOI is not associated with eating cheese. This
  is not a confounder

11
Coping with confounders

• In the design phase
• Investigators should be aware of confounders and
  able to control them
• First list the variables (like age sex) that
  may be associated with the predictor variable of
  interest as well as cause of the outcome

12
Coping with confounders

• Two design phase strategies
• Specification
• Matching
• Both sampling strategies
• Specification
• Design inclusion criteria that specify a value
  of the potential confounder and exclude everyone
  with a different value
• e.g In coffee and MI , only non smokers could be
  included in the study.if an association observed
  b/w coffee and MI, it obviously could not be due
  to smoking

13
Coping with confounders

• Specification
• Advantages
• Easily understood
• Focuses only on subjects for the research
  question at hand
• Disadvantages
• Limits generalizability
• May make it difficult to acquire adequate sample
  size

14
Coping with confounders

• Matching (mostly in case control studies)
• Selection of cases and controls with matching
  values of the confounding variable
• Pair wise matching
• e.g in coffee drinking study as a predictor of
  MI, each case (a patient with MI) could be
  matched with one or more controls that smoked
  roughly the same amount as the case (10-20
  cigarettes/day)

15
Coping with confounders

• Matching
• Advantages
• Can eliminate influence of strong confounders
• Can increase precision (power) by balancing the
  number of cases and controls in each stratum
• May be sampling convenience making it easier to
  select controls

16
Coping with confounders

• Matching
• Disadvantages
• Time consuming
• Requires early decision as to which variables are
  predictors and which are confounders
• Requires matched analysis
• Creates the danger of over matching( matching on
  a factor which is not a founder, thereby reducing
  power)

17
Coping with confounders

• In the Analysis
• Stratification
• Adjustment
• Stratification
• Ensures that only cases and controls with similar
  level of a potential confounding variable are
  compared.
• It involves segregating the subjects into strata.

18
Coping with confounders

• Stratification
• Advantages
• Easily understood
• Flexible and reversible
• Can choose which variable to stratify upon after
  data collection

19
Coping with confounders

• Stratification
• Disadvantages
• Number of strata limited by sample size needed
  for each stratum
• Few co variables can be considered
• Few strata per co variable leads to less complete
  control of confounding

20
Coping with confounders

• Statistical Adjustment
• Several statistical techniques are available to
  adjust for confounders.
• These techniques model the nature of the
  associations among the variable to isolate the
  effects of predictor variables and confounders
• This require software for multivariate analysis

21
Coping with confounders

• Statistical Adjustment
• Advantages
• Multiple confounders can be controlled
  simultaneously
• Information in continuous variables can be fully
  used
• Flexible and reversible

22
Coping with confounders

• Statistical Adjustment
• Disadvantages
• Model may not fit
• Inaccurate estimates of strength of effect (if
  model does not fit predictor-outcome
  relationship)
• Results may be hard to understand
• Relevant co variables must have been measured