Interpreting epidemiological findings [Epidemiology] Flashcards
Name 2 types of literature review.
narrative review
systematic review
What kind of review is this?
- brings published literature into single article
narrative review
What kind of review is this?
- sets out highly structured approach to searching, sifting, including and summarising literature
- underpinning basis for meta-analysis
systematic review
Name some strengths of narrative reviews.
- agile: easier + faster to write
- often more up-to-date than systematic review
- useful for areas of limited research or higher levels of variation in research approaches
- useful when bringing in work from different disciplines for less easily-answerable questions
Name some strengths of systematic reviews.
- aims to collate all available evidence
- implements highly specified protocol
- inclusion criteria
- can take many months to design
Name some limitations of narrative reviews.
- potentially bias (authors can over/under select works)
- can be over-speculative/unbalanced
- important evidence maybe omitted by chance (not-intent)
Name some limitations of systematic reviews.
- only as good as method employed
- only as good as the indices searched
- only as good as evidence incorporated
- very quickly out of date
look at search date, not publication date!
Outline the process of doing a systematic review?
What does a structured search enable?
enables transparency and future researchers to reproduce approach
What is the difference between indices and a registries
- indices: based on published research
-registries: registrations of research yet to be completed or published
Give some exampled of research indices
MedLine, Embase, PsychInfo
What diagram can be used to should the number at each stage when establishing screening/inclusion.
PRISMA diagram
Outline the process of screening/inclusion
- shows how many articles have been found in original search (1000-3000)
- how many articles removed due to duplicates
- process of screening
- full text reviews (eligibility)
- how many studied will be included?(10-30)
What is grey information?
information that is not published in scientific journals
Why do we need to be careful with grey information?
need to be more careful as not peer reviewed
Where can we find grey information?
via search engines: google scholar / open grey
What do the Cochrane Collaboration do?
bring together evidence into a more coherent batch of papers + publishes series of systematic reviews and meta-analysis and keeps them up-to-date
What is meta-analysis?
a quantitive, formal epidemiological study design used to systematically assess previous research studies to derive conclusions about that body of research.
What is this:
a quantitive, formal epidemiological study design used to systematically assess previous research studies to derive conclusions about that body of research.
meta-analysis
Meta-analysis combines the quantitive findings from separate studies into a _____
pool estimate
What does meta-analysis need from the pooled studies in order from them to be pooled?
As such, these pooled studies require ____ ____?
requires pooled studies to be sufficiently similar
critical appraisal
What word do we used to describe the difference between studies included in meta-analysis?
Heterogeneity
Define Heterogeneity in the context of meta-analysis
difference between studies included
What different sources of heterogeneity can between studies? (3)
give examples
Clinical
- patients, selection criteria
Methodological
- study design, blinding, intervention approach
Statistical
- reporting differences
What are fixed and random effects?
How can a difference in this type of weighting in pooled studies affect your meta-analysis?
CONFUSEDDDD
In fixed-effects models, we assume that there is one common effect.
A random-effects model assumes each study estimates a different underlying true effect, and these effects have a distribution (usually a normal distribution).
will give you slightly different weighting and may change findings
Publication bias
Define it.
- studies with positive findings are more likely to be submitted for publication
- studies with positive findings are more likely to be published my journal editors
What type of bias is this:
- studies with positive findings are more likely to be submitted for publication
- studies with positive findings are more likely to be published my journal editors
publication bias
What is a publication funnel plot used for?
used to assess likelihood of publication bias in meta-analysis
What is used to assess likelihood of publication bias in meta-analysis?
publication funnel plot
Why are epidemiologists increasingly looking to use big data / ‘real world evidence’ approaches?
to establishes the efficacy of intervention in the real-world that may not mirror the scientific conditions employed in the original randomised control trials
What is a trial endpoint? Give an example
an outcome that usually describes a clinically meaningful outcome
a common clinical endpoint in cancer trials: survival (at 12 months or five years)
What term to we used to describe this:
- an outcome that usually describes a clinically meaningful outcome.
a trial endpoint
By definition the endpoint should be specified _____.
a priori
We can think about clinical trial outcomes as being about:
_______– how well a therapy works in achieving a desired outcome.
_______ – how well a therapy works in not causing adverse events.
efficacy
safety
What is a primary endpoint and a secondary endpoint?
1ary: endpoint for which study has been powered
2ary: common for a study to examine a slightly different endpoint in addition to the primary endpoint.
Give an example of a secondary endpoint.
ie, while a study seeks to examine survival (i.e. alive or dead) another – often ‘softer’ - measure such as recurrence of disease or hospital admission might also be measured.
What happens if a primary endpoint is not proven but the secondary endpoint is?
If the secondary endpoint is proven but the primary endpoint is not, then the findings of the study may still contribute to the understanding of disease.
Name 4 different types of end points.
Primary
Secondary
Safety
Composite
Give an example of a safety endpoint?
How do we deal with them?
anaphylaxis / direct mortality associated with the therapy
major issues should usually be detected early in the trial process (before its rolled out to large numbers of patients
Often ‘safety endpoints’ are more nuanced than ‘direct mortality associated with a therapy’.
Explain this further.
measuring commonly observed adverse events (AEs) and grading them into a hierarchy of significance.
A large proportion of patients reporting AEs will require investigation.
Often ‘safety endpoints’ are more nuanced than ‘direct mortality associated with a therapy’.
Explain this further.
measuring commonly observed adverse events (AEs) and grading them into a hierarchy of significance.
A large proportion of patients reporting AEs will require investigation.
What is a composite endpoint?
could potentially describe any endpoint
‘composite’ = multiple potential endpoints have been added together
When do we use composite endpoints and give an example.
when an outcome is uncommon
- ie, one might combine myocardial infarction + ischemic stroke to give a new composite endpoint of ‘cardiovascular event’.
What graph can we used to analyse survival?
Kaplan Meier plot
How many criteria does the bradford hill criteria have?
9
The 9 Bradford Hill criteria:
strength
define it
a stronger association increases the confidence that an exposure causes an outcome
The 9 Bradford Hill criteria:
consistency
define it
tends to rule out errors and fallacies
The 9 Bradford Hill criteria:
temporality
define it
exposure must precede outcome
The 9 Bradford Hill criteria:
biological gradient
define it
a dose-response effects is compelling
The 9 Bradford Hill criteria:
plausibility
define it
relationship should be biologically plausible where the science is ‘understood’
The 9 Bradford Hill criteria:
coherence
define it
consistent with existing theory and knowledge
The 9 Bradford Hill criteria:
experiment
define it
evidence from experimentation should be supporting of the proposed link
The 9 Bradford Hill criteria:
analogy
define it
drawing upon analogous findings, we may make inference on the relationship
The 9 Bradford Hill criteria:
specificity
define it
an association between specific causes and specific effects
(more criticised proposal)
Define internal validity
findings are valid within experiment group
(but unsure if findings will be seen outside of experiment group)
define external validity
finds can be extended to individuals outside of experiment group
- generalisability
real life impact of a certain findings beyond the specific setting where research was conducted
What is bias
any trend in the collection, analysis, interpretation, publication or review of data that can lead to conclusion that are systematically different from the truth.
What can result in a bias?
- a systematic error in the design or conduct of a study can result in bias
- observe results different from the truth
- critical to make distinction between random and systematic error
- with a large enough sample, random error often cancel each other out, systematic ones do not
When can a valid inference be made?
- an inference is valid when there is no bias
What is selection bias?
when an individual’s chance of being selected for the study is related by both the exposure and the outcome
what is this:
when an individual’s chance of being selected for the study is related by both the exposure and the outcome
selection bias
What type of study is particularly susceptible to bias?
case control studies
What is this bias called?
- when the sample is taken not from the general population, but from a subpopulation.
- first recognised in case control studies when both cases and controls are sampled from a hospital rather than from the community
Berkson’s bias
What is this bias called?
- seen in observational studies of occupational exposures with improper choice of comparison group (usually general population)
- occurs because less healthy individuals are more likely to be unemployed than are healthy individuals (i.e., less likely to seek, gain, or retain employment).
What is Berkson’s bias?
- when the sample is taken not from the general population, but from a subpopulation.
- first recognised in case control studies when both cases and controls are sampled from a hospital rather than from the community
What is the healthy worker effect?
- seen in observational studies of occupational exposures with improper choice of comparison group (usually general population)
- occurs because less healthy individuals are more likely to be unemployed than are healthy individuals (i.e., less likely to seek, gain, or retain employment).
How can we avoid selection bias?
- controls representative of target population
- minimise non-response
- when many people decline to participate, it becomes more likely that some bias is introduced
- compare respondents with non-respondents
What is information bias?
where there is misclassification of the exposure, disease status, or both
Why may information bias occur?
can happen due to study variables not being properly defined or due to flaws in data collection
What type of bias does this cause?
where there is misclassification of the exposure, disease status, or both
information bias?
Give an example of how interviewer bias may occur and how we can prevent it?
interviewer might be more thorough when asking about the exposure status when the disease/outcome is present
- blinding
- collection process carefully standardised
Give an example of how recall bias may occur and how we can prevent it?
- patients might be more detailed in recalling past exposures related to the disease if they have disease/outcome
- using objective way to identify exposures: medical records/biomarkers
Name the different types of misclassification?
non-differential
differential
What is the difference between non-differential and differential misclassification?
non-differential: when when exposure status is misclassified, but equally amongst cases and controls
differential: where there is an error in determining an individual’s exposure status occurs unevenly among cases and controls
Give an example of non-differential misclassification?
error in determining outcome, but this occurs equally among exposed and unexposed individuals
What kind of bias does non-differential misclassification have?
results in bias always towards null
What kind of bias does differential misclassification have?
leads to bias estimate, but cannot predict towards or away from null
What is confounding bias?
the effect of an extraneous variable that wholly or partially accounts for the apparent effect of the study exposure, or that marks an underlying true association
confounding can lead to bias estimate
What is bias estimate?
In statistics, the bias of an estimator is the difference between this estimator’s expected value and the true value of the parameter being estimated
What are the 4 ways in which we can identify a confounding factor?
- Knowledge of subject matter
- the 3 conditions for confounding
- Stratification
- Compare crude and adjusted estimates
There are 4 ways in which we can identify a confounding factor.
Knowledge of subject matter is one of them. Can you explain how we can do this?
- explore literature
- knowledge of plausible biological pathways (not always possible with novel associations)
There are 4 ways in which we can identify a confounding factor.
The 3 conditions for confounding is one of them. Can you explain what these are?
- associated with the exposure in the source population
- associated with the outcome in the absence of the exposure
- not a consequence of the exposure
There are 4 ways in which we can identify a confounding factor.
Stratification is one of them. Can you explain what these are?
- compare stratum specific estimates with the estimate that we get when we analyse the entire set of data in the study
- when the pooled estimate is considerably different that what you would expect from stratum specific estimates, it is very reasonable to think that there is confounding
There are 4 ways in which we can identify a confounding factor.
Comparing crude and adjusted estimates is one of them. Can you explain what this is?
- if the adjusted OR differs from the crude OR by 15% or more, this may indicate confounding
- this number is arbitrary: can even introduce confounding by doing this
(not optimal)
What are Un-adjusted vs. Adjusted Odds Ratios?
In epidemiology, an un-adjusted OR will estimate the relative risk between a certain event in an exposed group with a certain event in an unexposed group.
Adjusted ORs are used to control for confounding bias. The AOR measures the association between a confounding variable and the outcome, and controls for that value.
What are adjusted ORs used for? What does the AOR measure for?
used to control for confounding bias
The AOR measures the association between a confounding variable and the outcome, and controls for that value.
What are adjusted ORs used for? What does the AOR measure for?
used to control for confounding bias
The AOR measures the association between a confounding variable and the outcome, and controls for that value.
What is effect modification?
when the strength of the association varies over different levels of a third variable
What do we call it when the strength of the association varies over different levels of a third variable?
effect modification
How should we deal with effect modification?
-
should not try to control !!!
- just report it in results
- conduct stratified analysis
What are the 3 tests for effect modification?
- Breslow-Day test
- Q test
- Interaction terms in regression models
What do we call it when an effect modifier potentiates the effect of the exposure?
synergism
What do we call it when an effect modifier diminishes the effect of the exposure?
antagonism
Would addressing an exposure where effect modification is apparent ever be useful?
yes
for example:
it may be possible to target an intervention into a more homogeneous pool of participants where a greater impact will be yielded
Would addressing a confounded relationship by addressing an exposure exclusively ever be useful?
very unlikely to yield a gain
What is a crude model?
- the univariate analysis of exposure vs outcome
- impact of the exposure on the outcome – with no consideration of anything else.
What is an adjusted model?
- multivariate analysis of a range of exposures vs. outcome
- multiple potential exposures have been included
- The inference is that the outputs of these analyses mean that holding all other adjusted variables equal, X is the association between exposure and outcome.
What method do we use to identify and account for potential confounding?
adjustment
What method do we use to identify and account for potential confounding?
adjustment
What is Multivariate regression?
a technique used to measure the degree to which the various independent variable and various dependent variables are linearly related to each other.
- highly effective way for us to identify possible confounding
- gives us this adjusted output
What is this?
the assumption that there is not one ‘better’ intervention present (for either the control or experimental group) during the design of a randomized controlled trial (RCT)
clinical equipoise
A true state of equipoise exists when one has no good basis for a choice between two or more care options.
