week 3 Flashcards
How can we decide if a risk factors causes a disease?
Solid body of evidence (strength of evidence)
checklist that helps us decide
Bradford hill criteria
provide ways of examining whether cause and effect is a reasonable inference
How many points does the bradford-hill criteria has?
9
what are the bradford hill criteria
- temporal relationship (essential)
- strength (effect size)
- consistency
- Analogy
- Specificity
- Reversibility
- Dose-response relationship (not essential)
- Plausibility
- Coherence
what is temporal relationship?
risk factor must occur/be present BEFORE the disease.
Strength - effect size
A strong association is more likely to be causal (but the reverse is NOT true)
Consistency (reproducibility)
similar results in different populations with different study designs.
Analogy
similarities with other well-established cause effect relationships
Specificity
t rarely occurs, as most diseases have multiple causes and most exposures, multiple effects
Reversibility
If the removal of a possible risk factor results in reduced risk of disease,
Dose-response relationship
e.g. Smoking, increased exposure increases risk of lung cancer
Plausibility
Must be consistent with knowledge from other sources (e.g. animal experiments) & should make
biological sense
Coherence
suggested cause-effect should be consistent with the natural history and biology of the disease
Causality
“Both practical and ethical considerations mean that causality cannot, in general, be proved in
human studies. Rather, it must be induced from demonstrated associations between and
exposure and health outcomes.
Challenges of causality
outcomes having multiple component causes
Distinguishing which of these are necessary or sufficient is central to preventive efforts
how much should it be pursue
When can links between exposures and outcomes/disease can be considered causal?
once full consideration has been taken of epidemiological noise -chance, -bias -confounding
External validity
generalisation to the entire poplation
The degree to which the study conclusions can be applied to other samples
Internal validity
measurements and sample.
two underlying concepts of external validity
- generalisability
2. applicability (to a particular sample within a population)
charact. of external validity
inclusion and exclusion criteria for study (age, health…)
Individuals agreeing to participate in research studies are different from those who don’t
Can recruited sample be 100% representative?
No, never!
Judgements to consider for external validity
- Age, sex, severity of disease, comorbid conditions
- Similar drugs, other doses, timing, route of administration
- Other outcomes (not assessed), different duration of treatment
“As the intervention was implemented for both sexes, all ages, all types of sports, and at different levels of
sports, the results indicate that the entire range of athletes,
Internal validity
Degree to which the investigator draws the correct conclusion about what actually happened in
the study
what could the design methods and conduct of the trial/study could do for the analysis?
- reduce likelihood of errors and chance findings
- eliminate possibility of bias
- minimise impact of toher factors (cofounding , interaction)
what can you do to minimise ‘chance’ findings? (false positive)
✓ Strong a priori rationale ✓ Plausible ✓ Adequate sample size ✓ Correct statistical analysis ✓ Replication of findings
Bias
A systematic error in the design, recruitment, data collection or analysis that results in a mistaken
estimation of the true effect of the exposure and outcome
What does bias does to studies?
limits validity and generalizability of study results
rarely eliminates during analysis.
Selection bias:
systematic error in the selection or retention of participants
Information (misclassification) bias:
systematic error due to inaccurate measurement or
classification of disease, exposure or other variable
Selection bias in case control studies
inherent in design ( non comparable between cases and controls)
problematic in hospital settings
Why is selection bias particularly problematic in hospital settings?
Cases: hospital will fail to enrol severe cases that died before reaching hospital. ( not representative of all cases in hospital)
controls: overrepresentation in the control group could underestimate association.
Selection bias in cohort studies.
less problematic.
exposure status identified prior to outcome occurring
can occur when loss to follow up / non-response rate differes between expose groups (e.g. heavy drinkers not responding the survey or family might be more likely to participate).
Will the cohort prevalence and incidence rates will be the same from the general population?
No, they will be different mostly due to selection bias.
when is it less likely to have selection bias in RCT?
If:
- randomisation performed correctly
- Sufficiently large sample
- blinding to treatment allocation
Types of information bias:
- Non-differential classification
- Differential classification
- observer bias
- interviewer bias
- reporting bias
- recall bias
What is non differential classification>?
it referst o when misclassification is randon and all individuals have the same probability of being misclassified.
what is differential classification?
misclassification of disease status is dependent upon risk factors (or vice versa).
Includes instrumentation errors, misdiagnosis, missing data.
observer bias
prior knowledge of expected outcomes influences the way inofmration is collected, measured or interpreted.
Interviewer bias
leasing questions
Reporting bias
: individuals may selectively suppress or reveal information
Recall bias
when the information provided on exposure differs between exposed & unexposed
when is Recall bias particularly problematic?
in case -control and retrospective cohort studies.
e.g. individuals with cancer may be more likely to recall exposure to toxic chemicals than controls
Hawthorne effect
Individual’s change in behaviour due to awareness of being observed. (e.g. with placebo)
Why could misclassified information happen?
ambiguous questions
instruments developed in one setting not appropriate to another setting.
Incriminating or personal embarrassing questions
multiple interviewers
self-reported, telephone and face to face survey can provide different results
Key questions to identify selection bias
Is the study population defined?
▪ Is it representative of the target population?
▪ Are the definitions of disease and exposure clear?
▪ Is the case definition precise?
▪ What are the criteria for inclusion/exclusion of participants?
▪ In case controls studies, are the controls representative of the population from which the cases
came?
▪ Could exposure status have influenced the selection of cases or controls?
▪ Are the cohorts comparable except for exposure/intervention status?
▪ Were losses to follow up kept to a minimum?
key questions to identify information bias:
Are the measurements as objective as possible?
▪ Were the observers/interviewers rigorously trained?
▪ Is the study blinded as far as possible?
▪ Are levels of follow up adequate & is it equal for all cohorts?
▪ Was the appropriate analysis performed?
▪ Were the variable groups defined a priori
▪ Is the interpretation supported by evidence?
▪ Were clearly written procedures used to standardise procedures?
▪ Were study participants randomised to observers / interviewers?
▪ Was self-reported information validated against any existing records?
what is a confounder
An extraneous variable that wholly or partially accounts for the observed effect of a risk factor
on disease status
Masks the true effect of an exposure on an outcome
Effect modifier (interaction)
A variable that differentially (positively or negatively) modifies the observed effect of a risk
factor on disease status
The effect of an exposure on an outcome is different for different groups
What is mediator?
on the causal pathway between exposure and outcome.
Confounding and interaction
NOT the same
The influence of 3rd factor which leads to an incorrect estimate of the association
between the exposure and outcome
what could confounding be?
an explanation for an association between exposure & outcome.
What happen to the association between two variables with a 3rd factors?
the association is distorted because of the third factor.
it can over or under estimate a true association.
where does the confounder must be associated and where not?
associated with both exposure and outcome and not on the causal pathway.
How to control confounding at the design stage?
- randomisation:
- restriction
- matching
- identified at design based on previous knowledge.
Where is randomisation possible?
only in clinical trial
what is restriction when controlling cofounders?
limits study participants to those with similar confounders ( e.g. only smorker) but limits generalibilizity.
What is matching?
refers to individuals in casee and control groups are as similar as possible. (mostly in case control studies).
must be decided in advace.
How do you control for confounding at the analysis stage?
- stratification
- multivariable analysis
- standarisation
- residual confounding
what is stratification?
examines the association between exposure and outcome withing different strata of the cofounding factors (e.g. seperate in smokers and non smokers).
Multivariable analysis
statistical modelling to control for 1+ cofounders
which method is the most common one used for cohort studies?
Multivariable analysis.
what does standardisation conveys?
using a standard population for reference to help negate the effect of cofounders.
what is residual confounding?
distortion that remains after controlling for confounding in the design and/or analysis of a study.
only adjusts for things we know about and have measured.
when does the 3rd variable ‘modifies’ a relationship?
When the association between exposure & outcome differs by a ‘level’ of another (3rd) factor
Does modifies a relationship means causality?
NO
what happens When the association between exposure & outcome differs by a ‘level’ of another (3rd) factor?
exposure has a different effect on outcome in different subgroups (stratified analysis)/
the overall estimate of the association between outcome and exposure is misleading
In the association between the flu and dying what factor can be an effect modifier?
age, health status, gender?
Is an association or effect existent if the ffect is statistically significant?
not necessarily.
Should you conclude about the scientific importance based on statistical significant?
NO :)
