Epi Flashcards
Which study design controls all con founders?
RCT
Stratification
Analyses patient subgroups separately and then weighted average
Multivariable regression
Takes into account a number of confoundeds at the same time
Single estimate of stratification
Mantel haenzel
Ecological fallacy
Average characteristics of a population
What can you measure in cross sectional?
Prevalence
NOT incidence
What do we calculate with case control?
Odds ratio
Bias in case control
Reverse causality
Selection bias
Measurement- recall and interviewer
Bias in ecological
Selection
Measurement
Reverse causality
Trend test
Statistical
Presence of a linear increase or decrease in risk associated with increase in exposure
Binary
Trend test 2 effects
Dose response effect
Threshold effect
Cohort bias
Reverse causality Selection Loss to follow up Recall Interviewer
Inclusion or exclusion criteria in rct causes
Poor external validity
Good chance of detecting a clinically significant effect
Power more than 80%
Not achieving planned sample size
High risk of missing a clinically important effect
Can only be published if it proves evidence of an effect
Internal validity
The intervention caused the outcome or an observed outcome
Construct validity
If what you observed is what you wanted to observe
Or what you did is what you wanted to do.
Minimum effect size
Should be big enough to detect the smallest effect that is clinically important
The probability of correctly rejecting the null when the treatment has an effect
Power
Outcome reporter bias
Form of publication
Only present things that support
Contamination
Cluster rcts
Interim analysis
If study over years
Data monitoring committee
Disadvantages of interim analyses
Open to abuse
Over estimate treatment effect
Completed by confidential committee independent of study researchers
Number needed to harm
Round down
Number needed to benefit
Round up
Bias in rct
Selection
Performance
Detection
Attribution
Concealment is not
Blinding
Attrition bias
Use an itt
Patients analyses to groups they originally allocated not on whether they completed
Only unbiased up confounded estimate of effectiveness
Reflects reality
Public health impact
Missing data can be assessed with
Sensitivity analysis
Sensitivity analysis
Primary analysis
Then repeated with missing data filled in (assumed)
If results same as analysis then they are robust
If different then must use caution
Itt minimizes
Attrition bias
Consort framework
Framework for reporting trials
Forest plot boxes
Draw attention to studies with greatest weight
Forest plot diamond
Overall summary estimate
Vertical unbroken line forest plot
Null wave
Data extraction done by
2+ independent observers
Prisma statement
Guidance on what to include in systematic review
Examples of fixed effect
Hanzel
Basics of fixed effect
Assumes one true effect weighted average
Any deviation is due to chance or sampling error
Only looks at variation within samples
Examples of randome effect
Dersimonian
Basics of random effects
Assumes heterogeneity
Within study variance and between studies
Wider range
Weighted average
Bigger weight to bigger studies
Weights use the inverse of the variance of treatment effect
Between study variance
Tsquared
Derived from q
Fixed effect weight
W=1/v
Random effects weight
Includes inter study variance
Random effect weights are
Smaller and closer to each other than fixed
Fixed effect
Assumes studies are all measuring same treatment effect
Time-trade-off
Respondents are asked to choose between remaining in a state of ill health for a period of time, or being restored to perfect health but having a shorter life expectancy.
Standard gamble
Respondents are asked to choose between remaining in a state of ill health for a period of time, or choosing a medical intervention which has a chance of either restoring them to perfect health, or killing them.
Visual analogue scale
Respondents are asked to rate a state of ill health on a scale from 0 to 100, with 0 representing being dead and 100 representing perfect health. This method has the advantage of being the easiest to ask, but is the most subjective.
the cost effectiveness plane
Cost on y and effective on x
Dominant
more effective and less costly (South-East)
Dominated
expensive and cheaper (North-West)
Incremental cost-effectiveness
difference in cost divited by difference in effectiveness.
up to reader to decide cost-effective.
ICER can be misleading unless…
one intervention is more expensive and more effective.
Net monetary benefit
required to know how much NHS is able to pay per QALY
Statistical variability in economic evaluation
due to small sizes, high variability of costs and missing data
Cost effectiveness acceptability curve
a sensitivity analysis in economic evaluation
one way sensitivity analysis
estimates for each uncertainty varied one at a time to investigate the impact on the results.
scenario analysis
best case
worst case
sensitivity
probability of a + test in people with the disease
specificity
probability of a - trest in people without the disease
SnNout
test has a high sensitivity-
a neg result would rule out the disease
SpPin
test has high specificity-
a pos result would rule disease in.
when 2 tests are equally costly and convenient we can use the
Likelihood ratio
NPV
probability of being disease free if test result is negative
spectrum bias
diagnostic test only finds barn door cases from the controls
work-up bias
gold standard is expensive, risky and unpleasant
cases who test + have gold standard then we underestimate the false -ves and overestimate the true positives
likelihood ratio (+)
sensitivity/(1-spec)
likelihood ratio (-)
(1-sensitivity)/spec
likelihood ratios
the further away from the null (1) the more informative the test.
LR=1
equal to chance
LR=1.5
greater than chance
when 2 tests are equally costly and conveneient we can use the
Likelihood ratio
deductive
quantitative
inductive
qualitative
depth
qualitative
dependability
data reliability and dependability.
codes?
independent coding?
triangulation?
qualitative data collection methods
observation
interviews
focus groups
snowballing
qualitative
max variation sampling
sample for heterogeneity
the researcher selects a small number of units or cases that maximize the diversity relevant to the research question.
negative/deviant case sampling
This involves searching for and discussing elements of the data that do not support or appear to contradict patterns or explanations that are emerging from data analysis.
Deviant case analysis is a process for refining an analysis until it can explain or account for a majority of cases.
saturation
now themes no longer arise
QALY less than 0?
Worse than death
credibility
plausible and trustworthy
been analysed? and grouped.
reflexivity
awareness of researchers contribution to the construction of meaning throughout the research
(can’t remain ‘outside’ when conducting qualitative)
dependability
data reliability and dependability.
codes?
independent coding?
triangulation?
CASP criteria
to help appraise research
qualitative and quantitative
Critical Appraisal Skills Programme assessment criteria
DIE CIDRE
Do nothing Inform Enable choice Change default policy Incentives Disincentives Restrict choice Eliminate choice
QALY questionairres
EuroQoi
HUI
SF6D
CPSPC
Concieved Performed Submitted. Published Cited
QALY 0
Dead
QALY less than 0?
Worse than death
Discounting leads to
down-weighting the cost/benefit in future years
Compound discount per year?
3.5%
Positive time preference
good things now
Technical efficiency
increases survival now
questionable effect on quality of life
Allocative efficiency
Which is better? A or B?
Small study effect
Small studies show bigger effects than larger
(due to pub/reporting bias)
type of heterogeneity
CPSPC
Concieved Performed Submitted. Published Cited
Sensitivity analysis for heterogeneity
get rid of low quality ones
subgroup analysis for heterogeneity
does the effect differ across the sub groups?
Measure of heterogeneity
Q
evidence for heterogeneity
What does the Q mean?
Nothing on its own
Need the P value.
p will disprove the null hypothesis that there is no heterogeneity
I^2 illustrates
the MAGNITUDE of the heterogeneity
Examples of fixed effect
Manetl
Peto
Inverse variance
Examples of random effect
Dersimionian L.
Culmulative effects life course
As you age you get more risks
gender inequality
NOT explained by inherent physiological changes
RII
assumes linear relationship between poverty and mortality
RII=
magnitude of the inequality
SII=
absolute difference
RII=1
no RELATIVE difference
SII=0
no ABSOLUTE difference
Deprivation indicator
small geographical areas
ecological measure
based on census- derived variables.
4 inequalities
gender
SEP
ethnicity
geography
Population strategy
treat everyone and protect those that are low risk
High risk strategy
only treat those at high risk
might miss some who present slowly.
Prevention paradox
contradictory situation where the majority of cases of a disease come from a population at low or moderate risk of that disease, and only a minority of cases come from the high risk population (of the same disease)
Rule of rescue
an ethical imperative to save individual lives even when money might be more efficiently spent to prevent deaths in the larger population
ICER rejected over
30K per QALY
20-30K per QALY?
Only innovative, Proven benefit
Shadow price
threshold which you will pay under
League table
rank everything and most cost effective on top
Order of priority setting
NICE Regional Local commissioning boards PCT Hospitals
Done annually
Explicit priority setting
tell the patient about ALL options even unavailable
Implicit priority setting
only talk about the ones that are available.
3 aims of public health
Protection
Improvement
Service
(PIS easy)
Bradford Hill criteria
Temporal Strength of association Consistency Biological gradient Reversibility Specificity
Ecological fallacy
the average person is not equal to all individuals
error type 1
incorrect rejection of the null
error type 2
failure to reject a false null.
non-differential selection bias
non-generalizable
differential selection bias
over or under estimation
Performance bias
unequal care because dr. knows
Detection bias
The doctor’s views affect the measurements
NNTB round
up
NNTH round
down
selection bias not possible in
cohort
stratification in confounding
estimates association between exposure and disease.
different subgroups then sweighted average.
Multivariable models
a number of confounders at once.
correlation
positive or negative.
correlation coefficient is the gradient of the line.
Regression
mathematical y=mx+c
PH outcome framework
improve determinants of health
improve health
protect health
health care ph and preventing premature mortality.
Joint strategic Needs assessment
demographic changes (Population now)
services appropriately tailored (Provisions now)
any unmet needs? (Shortages now)
pressure for future? (Future)
Power
the strength of the results to be against the null
dependent on sample size
Concealment
allocation sequence
Basic reproduction rate
R0
the larger the value, the more difficult to control
(secondary cases from primary)
Effective reproduction number
Ro x proportion susceptible
Control:
reduce transmission
Eliminate
get transmission near 0
Eradicate
transmission=0
DALYs have reduced with all infections except
HIV and malaria
CIC for an outbreak
control
investigate
communicate
mucosal vaccines are:
live
vaccine adjuvant
enhance response
preservatives
protect from bacteria/fungi
additives
stabilize from heat
Egg vaccines
flu and fever
MMR reaction
febrile convulsion
Hypotonic hyporesponsive episode
whooping cough
herd immunity is when
transmission less than 1 per case
polio transmission
oral- replicates in GI tract
lymph nodes- blood- meninges
replicates in mn and affects muscles.
Measles complications
Pneumonia Otitis media SSPE (fatal) Encephalitis Diarrhoea
Mumps complications
Pancreatitis Oophritis Orchitis Neuro- deaf Nephritis
standardised mortality ratio
indirect SMR=observed deaths/expected*100
killed immunization
DTaP
conjucated vaccine
HiB
Men C
13PVC
cost effectiveness analysis
money differences/ health benefits measured by primary outcome
cost utility analysis
money differences/ health benefits measured in QALYs
cost benefit analysis
money differences/ health benefits valued in money
cost consequences study
money differences/outcomes *benefit or not
accuracy
how representative the sample is of the population (you are near the true value)
precision
amount of variation between samples.
high precision means low variation
familiar aggregation
the tendency for a disease to more common in probands than the public.
Power
Calculated from type 2 error
Non inferiority trial
To be able to do a superiority trial or to get on the market
Equivalence trial
Set delta margins
Margins fall within margins of other drug
Standard deviations
1- 68.3
- 95.4
- 99.7
Work up bias
Gold standard is painful expensive
Only likely to do on worse cases
Consort
22 principles for rcts
Quorma
Meta analysis
Absolute risk reduction
__
Interval properties
To do with qaly
heritability
Proportion of total phenotypic variance attributable to genetic effects (h2);
for phenotypes arising from a large number of genetic loci.
h2= additive genetic variance / total variance
Often expressed as a %.
Can be estimated from extended pedigrees, nuclear families siblings, twins or adoptees.
Only applies to measured population; cannot be used to explain differences between populations