W4 Random sampling error, bias, confounding Flashcards
Today’s Class
-validity of studies in health research
-major threads to validity
–bias
—selction
—infromation (happens due to measurement error, last week)
-Confounding
Epidemiology and Causal Inference
-Goals of Epidemiology and health research…
-Building Blocks…->
Results of research…
-identification of causes and preventions for disease
-measures of Disease Frequency, Various Study Designs
Measures of Association
Once you have calculated a measure of association, you need to determine if the observed association is valid and if it is causal
–assessment of validity is a methodological undertaking
–evaluation of the process of ‘causal inference’ is more a philosophical discussion
Research Evidence
Strong evidence is (1…2…)
1) of the lowest possible random sampling error (a statistically significant exposure/outcome association)
2) Based on a good design
-free of selection and information biases
-under minimal influence of confounding
High validity! Of the study and the measures association
Not the validity of the measure of events or exposure
Internal Validity
do the observed results accurately reflect the true association?
Generalizability (External Validity)
- to whom can results be applied?
- requires internal validity
What are key things to note about internal validity and External Validity (generalizability)?
- If a study lacks internal validity, external validity is irrelevant
- We do not compromise internal validity in an effor to achieve external validity (generalizability)
External validity
-will be achieved by a sample that represents the target population
–also by weighting
—more in survey/data collection methods courses
do not confuse with selection bias( healtheir, younger, educated, etc. more likely to volunteer for research)
How do we determine whether our MEASURED ASSOCIATIONS are true (valid)?
Four hallmarks of Health Studies
1) A research question/plausible theory
2) a well thought design to address the research question
3) MEASUREMENT of exposure and outcome
4) Analysis to COMPARE groups (measure association)
-suboptimal design, imperfect measurement, and influence of other factors threat the VALIDITY of health/ epidemiological studies
- also we assumed there is no other factor except exposure and outcome
If we observe an association
-What are the two apparent associations?
First consider(3) before ensuring it is a true association:
-
-
-
-
If we observe an association
-What are the two apparent associations?
-Exposure + Disease/Outcome
First consider(3) before ensuring it is a true association:
-due to chance… if chance unlikely then:
-due to bias……If no then:
-due to confounding…..If no then:
-true association… YES
Validity is..
- having fewer errors (errors=measured value- True value)
Sources of error:
–chance (random sampling error)
–bias: systemic error in selection of participants and/or measurements
–Confounding
Threats to Validity
Threats to Validity
Random sampling error
1) Random error in measurement
–information bias
2) Randomization ( a process in experimental studies)
-role of chance and statistics
-sample variation, sample to sample differences
Random Sampling (week 2)
-selection techniques wherein the probability of selecting each sampling unit is known
-following laws of probability
–12 participants out of 36 are selected RANDOMLY
–chance of being in the sample is THE SAME for each of the 36 member of the population equal to 1/36
Random Sampling Error
-variability in sampling due to chance
-statistical analysis cant FIX this, reported p. value only shows how much THE OBSERVED RESULTS may be only DUE TO CHANCE IN RANDOM SAMPLING
-a wide confidence interval suggest a high probability of random sampling error
The best way to MINIMIZE random sampling error is to increase the sample size
What is a P. value?
Threats to Validity (check if makes sense)
1) chance
2) bias
Threats to Accuracy(check if makes sense)
Bias
Bias
-refers to a systematic error in the design or conduct of a study
-when bias occurs in a study the OBSERVED association between the exposure and outcome will be DIFFERENT from the TRUE association
Most biases relate to the STUDY DESIGN AND PROCEDURES and can be classified into categories:
-selection bias
-information bias (due to measurement error)
Types of Bias
(1) Selection Bias
(2) Information Bias
Is it possible to have both types of bias in the same study?
(1) WHO is in the 2x2 table
(2) WHERE in the 2x2 table
It’s possible to have both types of biases in the same study.
Threats to Validity (check if makes sense)
Selection and information
-information Bias that can occur if the collected information from or about study participants during MEASUREMENT is ERRONEOUS
Selection Bias
refers to a systematic error in the way particpants are SELECTED or RETAINED in a study
- it occurs when individuals have different probabilities of being included or retained in the study according to the EXPOSURE and/or OUTCOME/disease
KEY: Participation differs on exposure AND disease
Types of Selection Bias (3)
- Inappropriate Control Selection (control-selection Bias)-> case-control
2.Differential Participation->case-control, cohort - Differential loss to follow up-> cohort, experimental
Vitamin D cheese study
go over
Main types of Selection Bias(4)
- Volunteer bias: volunteers are more health-conscious or from a different socio-economic group..differential exposure (effect of interventions for enhancing physical activity in older adults
- Non-response bias: those suffering from a disease with a particular belief….differential outcome
- Membership bias: healthy worker effect..service in Vietnam reduced mortality rates
4.Loss to follow-up bias: in clinical trails or longitudinal studies the sickest usually leave the study early
2 year long randomized controlled trail of medication use and weight loss in adults with obesity. Three treatment (exposure) groups: Metformin, Orlistat, or Placebo
-particpants enrolled in this study with the intention and hope of losing weight
-during the 2 year intervention, 10% of the Metformin group dropped out, 25% of the orlistat group dropped out and 35% of the placebo group dropped out
-The LOSS TO FOLLOW-UP also differed by
WEIGHT LOSS (OUTCOME) and SIDE EFFECTS OF THE DRUGS (related to EXPOSURE)
Oral Contraceptive Pills (OCP) and Deep Vein Thrombosis (DVP)
Backround: OCP increases the risk of DVP
answer the question and go over
Reducing Selection Bias
-little (or nothing) can be done to FIX selection bias once it has occurred
–cannot be “controlled” in the analysis
–Unlike random sampling error, selection bias CANNOT be reduced by increasing sample size
-selection bias must be avoided through careful study DESIGN and CONDUCT
-define TARGET population
-CLEAR ELIGIBILITY CRITERIA/inclusion and exclusion criteria
-Maximize response rates/follow-up
Another threat to Validity is…
Confounding
A multivariate world
-the simple exposure outcome model is never a reality
-relationships are multi-variable; really!
the third variable can act as a: CONFOUNDER, effect modifier, or mediator (on the pathway)
Confounding
-mixed together; in which two effects are not separated
Confounding: Distortion of the actual association due to a mixing of effects between the exposure and an incidental variable(s) known as the confounder
-all part of an observed effect is due to a factor other than the primary exposure of interest
–SYSTEMATIC ERROR that threatens the internal validity of the study
Why does confounding occur?
-because the ‘exposed’ group and the ‘unexposed’ group are not EXCHANGEABLE
-they differ by factors OTHER THAN THEIR EXPOSURE STATUS
-confounders are particularly a problem for observational studies
Confounding example: Birth Order and Risk of Down Syndrome
Maternal Age as a Potential Confounder of the Association between Birth Order and Down syndrome
Criteria for defining a confounding variable
- Causally associated with the outcome (a true risk factor)
- Noncausally or causally associated with the exposure
- Not an intermediate in the causal pathway between exposure and outcome
How to Identify a Confounding Variable
-Literature review of comparable studies
-consult experts
-statistical tests
Example: Association of Physical Activity with Coronary Heart Disease (CHD)
Assessing Confounding
Step 1 + Step 2
Step 1: conceptualize the relationship
Step 2: think of variables with criteria q and 2
age, sex, education,blood pressure, lipid profile,diabetes, inflammatory markers , stress, diet/smoking/drinking
Assessing Confounding
Step 3+step4+step 5
Step 3: exclude those on the pathway (criterion 3)
Step 4: identify true confounders
Step 5: DEAL with confounders
How to deal with confounding
A) At the design stage
-Restriction..
-Matching..
-Randomization
-Restriction:limit study inclusion criteria with respect to confounding factor(s); study only men or women
-Matching: produce case and control (exposed/non-exposed) groups that have similar characteristics
-Randomization: experimental studies
How to deal with confounding
(B) At the analysis stage
-Standardization
-Stratified analysis
-Including confounding factor(s) in a ______ regression model
-Standardization: age-standardization is in fact ‘adjustment’ for age
-Stratified analysis
-Including confounding factor(s) in a multivariate regression model
Is the association between Physical Activity and CHD CONFOUNDED by sex?
review im tired asf its 523 am!
Does coffee drinking cause CHD?
Does coffee drinking cause CHD? Cont.
First step: identify who drank coffee regularly
Does coffee drinking cause CHD? Cont.
Smoking causes CHD- what if smokers drink more coffee than non-smoker?
Does coffee drinking cause CHD? Cont.
Smoking causes CHD- what if smokers drink more coffee than non-smoker?
-Identify smokers and sort (stratify) by smoking status:
calculate the RR for the effect of coffee SEPARATELY for smokers and non-smokers
Does coffee drinking cause CHD?
The big 3(CBC) threats to validity
-chance
-bias (selection, information
-confounding
can you differentiatie them??
review tut for this week and write down answers
