CAUSAL INFERENCE Flashcards
Making judgment about causality; has process to follow
Causal Inference
Process of using different statistical methods to characterize the association between variables.
Statistical Association
(Statistical dependence between two variables
There is an identifiable relationship bet 2 variables
Association/ relationship is either positive or negative)
Process of ascribing causal relationships to associations between variables
An example of an association
Causal Inference
Factor that plays an essential role in producing an outcome
Event, condition, and characteristics
Presence of this factor should result to an outcome
Cause
Cause = Exposure, Outcome = disease
Ex: great intake of sugar = diabetes
Identifiable relationship between exposure (factor) and disease (outcome)
Cannot tell yet which one is the factor and which one is the outcome
Relationship could be co-existence: bidirectional: Cannot say yet that the exposure is the cause of the disease or that the factor is the cause of the outcome
Association
Cannot say yet that the exposure is the cause of the disease or that the factor is the cause of the outcome
Ex: poor lack of education, intelligence success in life
Presence of mechanism that leads from exposure to disease
Relationship is cause-effect: one direction: unidirectional: causal
There is really a cause that leads to the effect
Cause
Cause must precede the effect; cause first then effect
Ex: Mycobacterium tuberculosis —> TB, x —> y, infectious agent —> disease
Types of Association: (Causal vs Non-causal)
direct: Alteration in the frequency or quality of one event is followed by a change in the other
Direct relationship; one increase/ decreased the other increases/ decreases too
Causal
Types of Association: (Causal vs Non-causal)
indirect: Association is a result of the relationship of both factor and disease with a third variable
There is a relationship between the 2 variable because of the presence of a third/ confounder
Non Causal
Third variable: confounder: only variable that makes the 2 variables related
Associated to exposure and will be the risk factor for the outcome of interest
Confounder should be eliminated to be a causal association/ direct
Process/ Steps of Causal Inference : (Step 1 or Step 2)
Rule out chance, bias, confounding as explanation of the observed association
If ruled out = association is valid
If not ruled out, the association is not valid
Chance: external validity; random errors
Bias and confounding: internal validity; systematic errors
Step 1
Determine the validity of the association
Process/ Steps of Causal Inference : (Step 1 or Step 2)
Consider totality of evidence taken from a number of sources
Step 2
Determine if observed association is causal
Causal Inference: Step 1: Validity of the Association
(Internal vs External)
Bias and confounding
Estimate of the effect measure is accurate. Association should not be due to systematic error
Internal Validity
Validity within the study
Causal Inference: Step 1: Validity of the Association
(Internal vs External)
Chance
Estimate generalizable to a bigger population. Not due to random error
External Validity
Validity beyond the study
TRUE OR FALSE:
The goal of epidemiologic studies is to estimate the value of the parameter (population) with little error
TRUE
Sources of errors:
sampling errors; chance
Difference between population value of parameter being investigated and the estimate value based on the different samples
Inference will be inaccurate due to chance
Random Errors
Generalization about the group on the basis of data from the sample of the group
Chance
Sources of errors: distortion in the estimation of the magnitude of association between Exposure and Disease (over or under estimation) Deviation from the truth Bias: Selection and Information Confounding
Systematic errors
Types of Bias:
non-representative sample; not valid
Sample respondents are not representative of the population or general respondents
Selection Bias
Types of Bias:
inaccurate information collected from sample
Misclassification: Differential (non-random) and Non-differential (random)
Information Bias
Sources of Bias:
anything used to collect the data
Ex: weighing scale that was not calibrated = inaccurate info
Questionnaires that have vague instructions
Instrument
Sources of Bias:
respondents can give inaccurate information
Due to old age, not want to disclose information
Subjects
interactive responses; responses are modified because of the knowledge that they are being studied or observed
Hawthorne effect
deals with instrument or there is a random fluctuations in some of the biological factors of interest
Biologic variability
Sources of Bias:
researchers should not have an influence or prior knowledge before collecting the data; researchers are the one that modifies the answers of the respondents or results
Observers
strategy to avoid observers’ bias wherein researchers are not involved in data collection
Blinding
Mixing the effect of the exposure on the disease with that of a 3rd factor
Associated with the exposure: bidirectional
Risk factor of the disease/ outcome: unidirectional —>
Confounding
Presence of Confounder can lead to over or under estimation of the association
Methods to Control Confounding: Design Stage
aim is random distribution of confounders between study groups
Random assignments of individuals in the study
Can only be done in experimental studies
Randomization
Methods to Control Confounding: Design Stage
restrict entry to study of individuals with confounding factor
Limiting the participants to one category of the confounding variable
Restriction
Methods to Control Confounding: Design Stage
aim for equal distribution of confounders
Mostly used in case-control studies
Confounders are identically distributed among each of the study groups
Matching
Methods to Control Confounding: Analysis stage
confounders are distributed evenly within each stratum
Stratified analysis
Methods to Control Confounding: Analysis stage
analysis of data that takes into account a number of variables simultaneously.
Multivariate analysis
tool that can be used to assess if observed association is causal
Bradford Hill’s 9 Criteria for Causal Inference
HILL’S CRITERIA
-Biostatistician and epidemiologist
Bradford Hill’s 9 Criteria for Causal Inference:
stronger the relationship between the 2 variables the less likely the relationship is due to the confounder
Can check the risks and odds ratio
Strength of association
Bradford Hill’s 9 Criteria for Causal Inference:
exposure must occur first before the outcome
Important and required for establishing causation
If there is no temporality, there is no causal relationship
Temporality or time element
Bradford Hill’s 9 Criteria for Causal Inference:
relationship produces the same results even with different people, circumstances, and methods
Outcome is consistent
Consistency
Bradford Hill’s 9 Criteria for Causal Inference:
knowledge: rational and theoretical basis for a relationship supported by known biological and other factors
There is already an established or existing knowledge, theories, and studies about the factor, and is really the cause of the outcome
Theoretical plausibility
Bradford Hill’s 9 Criteria for Causal Inference:
facts stick together as a coherent whole
Cause and effect interpretation fits with what is known regarding the diseases’ natural history, transmission or patterns
Coherence or Biological plausibility
Bradford Hill’s 9 Criteria for Causal Inference:
in an ideal situation, the effect or outcome should only have one cause for it to be a causal relationship
Specificity in the causes
Bradford Hill’s 9 Criteria for Causal Inference:
direct relationship between the dose and response or exposure and outcome
Dose-response relationship
Bradford Hill’s 9 Criteria for Causal Inference:
strongest and most direct epidemiologic evidence to make judgement about the existence of the cause-effect relationship
Conducted an experiment to see if the factor is really the cause of the outcome
Strongest support for causality if it is done well
Experimental evidence
Bradford Hill’s 9 Criteria for Causal Inference:
weakest and sometimes not considered criteria
Commonly accepted phenomenon in one area can be applied to another area
Analogy
