Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
7/27/2014
1
Bias and Confounder
Asst. Prof. Mathuros Tipayamongkholgul, PhD
Department of Epidemiology,
Faculty of Public Health
Mahidol University
Do you believe in this statement?
• Smokers has higher risk of liver cancer than non-smoker by 4 times.
• Lack of physical activity increased risk of CHD by 10 times.
It would be probably true if the study has validity or
NO BIAS and NO CONFOUNDER.
Objectives of Epidemiologic Research
To examine magnitude of the problems
– Prevalence of CKD in DM patient
– Proportion of CKD with appropriated care
To identify magnitude of association between factors and outcome
– Smoking increase risk of Lung cancer
– Different treatment cause different survival time
Study Validity
Distortion
Observed value
True value
7/27/2014
2
Validity in Epidemiologic research
• External validity (Generalization)
• Internal validity (how close study findings are to the TRUTH )
To identify association between DM and CKD
Effect
• CKD• Disease• Death• Intermediate outcomes
• CD4+ count• Underlying disease
Cause
• DM• Exposure• Risk factor• Covariate
• Age• Gender• Smoking
Threats to Internal Validity
A study’s internal validity can be compromised by three things….
I. Chance/Random error
II. Bias
III. Confounder
Chance error
• Measurements are based on a sample.
• Number of selected samples is important for the validity of the measurements
• Number of sample == Validity
7/27/2014
3
Sample 1: prevalence = 0%
Sample 2: prevalence = 20%
Sample 3: prevalence = 40%Population: prevalence = 20%
I. Chance error: Sampling I. Chance: Precision of CI
Larger sample size provides more narrow interval
N=1000 : 30% (95%CI 25 - 36%)
N=100 : 30% (95% CI 9 - 58 %)
30%
N = 1000
N = 100
I. Chance: P-value
• For comparison of 2 or more
• Tell the probability (percentage) of a wrong conclusion of study result
• Association between
– Number coffee drinking per day
– Developing CA stomach
– RR = 3.6
- c2 , P value = 0.35
II. BiasBIAS Systematic deviation of results or inferences from
truth.
Processes leading to such deviation. An error in the conception and design of a study—or in the collection, analysis, interpretation, reporting, publication, or review of data—leading to results or conclusions that are systematically (as opposed to
randomly) different from truth (Portal, 2008 A dictionary of
Epidemology 5th ed.)
Note: Chance is random error -- not systematic.
Therefore, tests exist based on probability
There is no way to correct for bias later!
7/27/2014
4
II. Bias
Systematic error is caused by 2 main steps:
I. Selection of study sample (Selection bias)
II. Inaccuracy of exposure or outcome measurements (Information bias)
II. Bias
• Wrong selection of comparison groups
• Selecting in different ways
• Give you the wrong answer for the association
• A mistaken estimate of an exposure’s effect on the risk of a disease
Selection Bias:Unrepresentative nature of sample
Cancer of the pancreas VS Coffee consumptionExample:
Cases: Histological confirmed diagnosis of
pancreatic cancer
Controls: Hospitalized patients admitted during
the same time as cases
OR of drinking coffee > 5 cups/day = 2.6
Do you agree with the results? Why?
Selecting Controls
• Coffee and pancreatic cancer, MacMahon B et al. NEJM 1981
– Coffee consumption was associated with pancreatic cancer
– Controls were selected from other patients admitted to the
hospital by the same physician as the case, often
gastroenterologist
– This specialist would admit patients with other diseases
(gastritis or esophagitis) for which he or the patient would
reduce coffee intake
– Controls intake of coffee not representative of population at
risk
7/27/2014
5
Pancreatic CA No Pancreatic CA
Coffee Consumption
Unrealistically Odds Ratio (BIAS)
NO Coffee Consumption
Case Control
Exposed
Not Exposed
A B
C D
Information Bias
• Interview mothers about the history of x-ray exposure comparing between 2 groups
• A mother who had a child with a birth defect often tries to identify some unusual event that occurred during her pregnancy with that child.
• While a mother with a normal child do not try that much.
Malformation+ No Malformation
Hx of radiation
No Hx of radiation
Unrealistically Odds Ratio (BIAS)
Case Control
Exposed
Not Exposed
A B
C D
III. Confounder
• An observed significant association may be
attributable to another unconsidered factor.
• Alcohol Drinking VS Lung Cancer
• The third variable Cigarette smoking
7/27/2014
6
III. Confounder
• The third factors affected the association;
– Confounder
– Effect modifier or Interaction
Alcohol VS Lung Cancer
CA Lung+ CA Lung-
Alcohol+ 71 52
Alcohol- 29 48
100 100
OR = 71x48 = 2.352x29
•This effect is from an unbiased Case-Control study•It is a true effect from the study But is it OK to report this result?
Alcohol VS CA Lung
• Is it possible that the obtained association was actually due the effect of the other factor?
• In the reality Alcohol may not lead people to develop CA Lung.
• Alcohol is just a coincidence of the other true association!!!!
• Let’s think about the effect of Smoking on this association
CA Lung+
CA Lung-
Alcohol+
71 52
Alcohol-
29 48
100 100
OR = 71x48 = 2.352x29
CA Lung+
CA Lung-
Alcohol+ 63 36
Alcohol- 7 4
70 40
OR = 1
SMOKE NON-SMOKE
CA Lung+
CA Lung-
Alcohol+ 8 16
Alcohol- 22 44
30 60
OR = 1
7/27/2014
7
III. Confounder
• Smoking Alcohol Drinking
• Smoking CA Lung
• Smoking is a confounder for the association between Alcohol and CA lung
III. Confounder
• Association between CHD and Smoking
smokers nonsmokers
CHD 305 58 363
no CHD 345 292 637
650 350 1000
• RR = 2.8
III. Confounder
• Gender M/F might be a confounder of the association
• Gender CHD
• Gender Smoking
What is the RR among Men?
What is the RR among Women?
Calculate stratum-specific measures of association...
STRATUM 1: MEN
smkrs nonsmkrs
CHD 300 50 350
NO CHD 300 150 450
600 200 800
STRATUM 2: WOMEN
smkrs nosmkrs
CHD 5 8 13
NO CHD 45 142 187
50 150 200
RR = 2.0
RR = 1.9
7/27/2014
8
IS THERE A CONFOUNDER?
CRUDE RR for smoking and CHD =2.8
STRATUM-SPECIFIC RR for smoking and CHD with gender as a potential confounder...
MEN RR = 2.0
WOMEN RR = 1.9
Do Breslow- Day tests
Gender confounds the association between smoking and CHD because the crude RR of 2.8 is NOT the same as the stratum-specific RR’s of approx. 2.0
roughly the same
Smoking VS CHD
• Smoking CHD : RR = 2.8 (Crude RR)
• After adjusting for gender
• Smoking CHD : RR ~ 2 (Adjusted RR)
• Gender confound the association between smoking and CHD
• Gender or Sex increases the effect of smoking
CONTROL FOR CONFOUNDERS
• Review for the potential confounder of the association you are going to find.
• Restriction
• Matching (in case-control study)– Matched by Smoking status
– Matched by Gender
• Stratified analysis Mantel Haenszel Method
• Multivariate analysis
Criteria of Confounder
• Related to the disease outcome
• Related to the exposure
SMOKING CA Lung
OR or RR > 1
SMOKING Alcohol
OR or RR > 1
7/27/2014
9
Criteria of Confounder
• Not be an intermediate of the association between the exposure and the outcome
CHDHigh Fat Diet
Serum Cholesterol
IV. Effect Modification
• In the process of stratification to look at the RR or OR in each subgroup, what is the explanation when the RRs or ORs in each subgroup are not the same?
IV. Effect Modification
• You were going to find the association between Aflatoxin and liver cancer
• You conducted a cohort study
• Crude RR of Aflatoxin = 5.8
• You would like to know if HBsAG was a confounder for the association or not
• You performed STRATIFIED ANALYSIS
CA Liver+
CA Liver-
Aflatoxin +
Aflatoxin -
1000 1000
RR = 5.8
OR = 3.4
HBsAg-
CA Liver+
CA Liver-
Aflatoxin +
Aflatoxin -
OR = 8.3
HBsAg+
CA Liver+
CA Liver-
Aflatoxin +
Aflatoxin -
7/27/2014
10
IV. Effect Modification
• Crude RR; Aflatoxin:CA Liver = 5.8
• Stratum specific: HBs Ag +– RR = 3.4
• Stratum specific: HBs Ag -– RR = 8.3
• HBs Ag is not the confounder of the association (3.4 ≠ 8.3 ; not homogeneous)
• HBs Ag is the effect modifier of the association• No adjustment required• Report stratum specific RR
IV. Effect Modification
Hbs Ag
CA LiverAflatoxin
Exercise
• You are interested in the association of non-regular exercise during young adult and MI
• What is your method to identify the association?
• Design?
• Outcome of interest?
• Exposure of interest?
• Measurement of association of interest?
Exercise
• The Crude OR of non regular exercise on the development of MI = 3.5
• What do you need to do before reporting the result?
• What is the potential confounders?
– Smoking
– Diet
– Age
– Gender
MINon Reg Excs
7/27/2014
11
Exercise
• Crude OR = 3.5 (Non-reg Excs VS MI)
• Stratified Analysis for Smoking status
• Among Smoke+; OR = 1.80
• Among Smoke-; OR = 1.78
• What is your conclusion?
SMOKING
MINon Reg Excs
ORMH = 1.79
Exercise
• What is the potential confounders?
– Smoking
– Diet
– Age
– Gender
Exercise
Crude OR = 3.5 (Non-reg Excs VS MI)
Stratified Analysis for Smoking status
Among Men; OR = 4.8
Among Women; OR = 1.7
What is your conclusion?
Exercise
• The effect of Non-regular exercise during young adult on the development of MI was different in each gender
• The effect was higher among Men than Women.
• Among men, those who had a history of non-regular exercise during young adult had 4.8 times more likely to develop MI than those who had a history of regular exercise
• Among women, those who had a history of non-regular exercise during young adult had 1.7 times more likely to develop MI than those who had a history of regular exercise
7/27/2014
12
CONTROL FOR CONFOUNDERS
• Review for the potential confounder of the association you are going to find
• Restriction.
• Matching (in case-control study)– Matched by Smoking status
– Matched by Gender
• Stratified analysis Mantel Haenszel Method
• Multivariate analysis