Epi Terms Flashcards
Accuracy
Ability to give a true measure of the substance being measured.
Analytical Epi
examines how an exposure relates to a disease.
Attack rate
used in epidemics- risk of becoming affected ( #of people sick / # of people at risk )
Risk difference / Attributable risk
RD = risk of O in E+ minus risk of O in E-
RD: p ( D+ I E +) - P (D + I E- )
Indicates the increased risk of outcome ifyou are exposure positive ,beyond.
baseline
the absolute difference btwn E+ and E - groups
Interpretation: for every 100 exposed )# had it due to exposure
Measures of Effect
The impact of a risk factor (E) on a disease
expressed using absolute effect and is computed as the difference btwn 2 measures of disease freq.
more closley relates to the #of cases exposure causes (or prevents) than MoAs
Effect of E on O (literal #of cases)
Baseline level of risk
Incidence of O in non-exposed group
Attributable risk of being exposed (Afe)
isthe proportion of O in the exposed group That is due to the exposure , assuming causal relationship
Interpretation: 88 percent Of diseased cases among exposed are due to exposure
Afe also for vaccine efficacy and you treat non-vaccinated as exposure positive
Afe: RD / P(D + | E + )
OR
Afe= (RR -1 ) / RR
What do Afe and RD quantify?
The effect of an exposure in an exposed grOup but do not reflect effect of E on pop.
example: there may be a strong association btwn IV use and HIV (MOE exposed) but if IV use is rare in a Specific pup, then it will not contribute much to HIV prevalence (MOEpop)
MOE pop
used in public Health
A relatively weak risk factor that is
common may be more important in determinants of Disecise in public pop
population Attributable Risk
Analogus to RD- Simple difference btwn 2 groups
PAR isthe increased risk of outcome in entire pop due to exposure
Is influenced by strength of association and frequency of exposure to risk factor
PAR= P(D + ) - P (D+ I E- )
PAR= RD* P (E + )
interpretation: For every 100 people in population 13.2 have HIV due to exposure (IV use)
population fraction Risk
Analogous to Afe
Reflects the effect of The disease in entire pop rather than just E+ group
AFP is the proportion of disease in pop that would be avoided if exposure were removed
Afp =PAR /P (D+)
interpretation: 66 percent of HIV Cases in pop are due to exposure
What MOEs and MoAs cannot be used in case control?
Relative risk, Incidence risk, Risk difference, population Attributable risk
AFP and AFe can be estimated
Standard Error
precision of estimate oF a sample mean as a measure of the pop mean
A measure of accuracy ofthe estimate
SE= SD/ sqr n
SE tells us how accurate the mean of any gives sample from that population is likely to be, compared the the the mean. when SE increases it becomes more likely that any given mean is an inaccurate representation of the the mean.
Standard Deviation
tells us how spread out thedata is
It is a measureof how far each observation is from the mean
SD= sqr variance
confidence Intervals
a range of values which contain a population parameter with a given probability that is likely to encompass the valve
The level of uncertainty of an estimate
p value
probability of obtaining The observed result (or more extreme) if null- hypo is true, based on a predetermined Cutpoint
Why should we use caution when using p-values?
knowing a result was significant does not provide any information . about the magnitude of The effect observed
Necessary cause
precedes the outcome and must always be present for disease to occur.
le : exposure will always be present if disease occurs
Sufficient cause
proceeds the outcome , is often multifactorial , and will always produce disease
le: if the exposure is present, the outcome will follow.
causal complements
Additional factors that combine with exposure to form sufficent cause.
What can a component cause Model not tell us?
confounding and intervening variables
Strength of association
Bias
reason why study sample and true population differ, beyond random variation.
cause
Any factor that produces change in severity or frequency of an outcome.
Descriptive epi
examines distribution or patterns of Disease in pop
useful for hypothesis generating surveillance Implementation of control (prevention ) Strategies,
Determinant
Any factor that when altered produces a change in freq or characteristics of disease
Diagnostic test
used to confirm or deny or classify disease . used to guide treatment or aid in prognosis
Incubation period
period between exposure to Clinical disease
Latent period
period between exposure to detectable pathological changeldisease
lead time Bias
screening bias that makes it look like it improves survival only because we detected disease eany , but the person would have lived same amount of time
Makes screening program look better than it is
Negative predictive value
probability of truly being disease negative given a negative test result
p (D- I T - )
d/ (Ctd)
positive predictive value
probability of truly being disease positive given a positive test result
p( D + I T + )
a/a+ b
non-differential Misclassification
Magnitude and direction of the misclassification are similar in 2 groups being compared.
usually bias towards null aslong as Sn +Sp >1 → less likely to detect a difference of there is one
ex. Scale is offby 1 Ib ) will be same in both g groups
prevalence
number of cases (new and existing) in a specified pop at a given point in time .
conceptually prev= Incidence* duration
prev decreases when ppl die or getbetter (Short disease)
“How much disease is present”
measures burden of disease in a pop at any given time
Issues with prevalence
Does not tell us when disease was obtained
Issues with prevalence
temporal issues for causal inference
Magnitude’ May vary drastically since prev is a function of I and D… le if disease freq is low but has long duration, prev will increase
precision
consistency
Sensitivity (Sn)
Ability of a test to detect true diseased Ppl correctly
a /a+ c
p(T + I D + )
Snout → False neg fraction = l-sn
specificity (Sp)
Ability of a test to correctly classify non-diseased people
Spin-> false positive fraction =1- Sp
p( T- I D - )
Sp= b/ b+ d
If you want to confirm disease …
use test with high Sp because there would be few false positives
ex. scary test
if you want to rule out disease …
Use test with high sn because There would be few false negatives
ex. cost for false negative is high, ie. highly infectious disease ‘ you want to minimize false negatives
* decrease catpoint
Why are predictive valves not a good measure of test performance?
pVs vary with prevalence increased prev = increased predictive value of test
ROC curve
isa plot of Sn 1-SP( false positive fraction)
Area under curve measures overall ability of test
left corner best for Sp and Sn
multiple test- parallel
test positive with one or both and you are considered disease positive
high sn , low sp
Multiple test-series
Test positive to both tests to be desease positive
nigh Sp, low sn
Name 2 memods to compare tests on a continuous scale
limits of agreement plot: Aka bland-Altman Plot :
concordance correlation coefficient : perfect agreement is perfect 45 degree line
Name a test to compare tests on a dichotomous scale… .
KAPPA: removes agreement that would happen by chance alone. O would mean no agreement 0.8-1is almost perfect.
proportion
numerator is a subset of the denominator
ex: 188 ppl are tested for COVID and 82 test positive , me eqn is 82/188
prev and risk are proportions
point prevalence
number of cases at one point in time / .
period prevalence
Of people who are identified as cases during specified time period
Risk
numerator is #of new cases over a period of time
is a proportion
probability that an individual will contract anatcome in a defined time period
since it’s a probability it’s dimensionless
only first case counts in numerator since it’s a proportion
AKA: culminative incidence
used in studies where we want an individual prediction
closed populations with short disease period
Rate
of new cases taking into account time at risk (person time at risk)
not a proportion
probability Of new cases occuring in a time period.
Has dimensions
AKA: incidence density
Rate focuses on cases
open pop with long risk period.
incidence
#of NEW events in a defined population within a specific time period. associated with studies about BECOMING ill
incidence Risk calculation .
I risk = # of new cases in time period / initial NAR -1/2 W D
Incidence Rate calculation
Exact method:
- used when we know exact amount of person time contributed by each member , preferred but often info is not available
#new cases/ person-time @ risk
estimated method:
- if only I case per person:
I Rate= cases / (#@start - 1/2 sick - 1/2WD - ‘/2 Add) x time
- if multiple cases per person included :
(Rate= cases/ (#@start -1/2 WD-1/2 Add )xtime - if individual movements are unknown:
Irate: cases/ 1/2 (# disease free@ start- # disease free @ end)
case Fatality
deaths among cases /# of cases
proportional Morbidity
casesordeaths from specific disease /cases or deaths from all diseases
Standardization
A way to control confounding by standardizing risk or rates-le dividing pop into Strata based on confounding factors
The goal is to make Inferences about the factors which affect freq of disease
indirect Standardization
using a set of Standard rates from a referent pop
- compute Stratum specific rates of referent pop
- Multiply SSR oF referent pop by # of ppl or proportion in each strata from Study pop (Strata specific rate)
- Add to getexpected rate
- calculateSMR= crude rate/ observed rate
- Multiply SMR by referent pop Crude rate to get standard rate
Rates usedfrom referent pop / no SSR are available or pop is small
SMR
Standardized Mortality Rate
percent increase or decrease immortality relative to referent pop
SMR < 1 means study pop has lower risk
Direct Standardization
used when #of events or rate in each Strata is known, i e we know age Specific Mortality rates in City 1 and 2
STeps :
1. calculate study pop “Stratum specific rate” =SSR ref pop # in each Stratum* study pop Stratum specific rate
2. #Stratum specific rate (study pop) * referent pop proportion (SSR x #ppl in each Strata )
3. Add product for final Standard rate
What are Hills criteria for causation?
- Strength of association 2. consistency 3. specificity 4. Temporality 5. Biological gradient/dose response 6. Biological plausibility 7. coherence 8. Experiment 9. Analogy
consistency (Hills criteria)
The same results have been found by different persons, in different places, circumstances and time (ie are the results due to random error, chance or fallacy? )
Havethe effects been seen by others?
specificity (Hills Criteria)
one cause one outcome
→ usually not useful as many disease is multifactorial
-→ if specificity exists we may be able to draw conclusions without hesitation
Temporality (Hills Criteria)
Does Exposure preceded outcome?
Biological Gradient /dose response (Hills Criteria)
If the association is one which can reveal a dose response relationship then weshould look closely for this.
→ A change in exposure causes a change in association to outcome
does increase in exposure lead to increase in outcome?
plausibility. (Hills Criteria)
is the relationship possible based on current knowledge?
Does the association make sense ?
coherence (Hills Criteria)
Similar to plausibility_ the cause. effect relationship should not interfere with what is generally known about the natural history of disease
is the association consistent with available evidence?
Experiment (Hills Criteria)
Strongest support for causation can come from experiment
does treatment X really effect the outcome ?
Analogy (Hills Criteria)
when there is strong evidence of a causal relationship bowman exposure and specific atcome, we should be more accepting of weaker evidence that a similar agent may cause a similar outcome.
Is the association similar to others ?
Strength of association (Hills Criteria)
The larger the association btwn exposure and disease the more nicely it is causal.
→we should not disregard smaller associations
measured in Risk ratio or odds ratio
Why doesn’t Hill include Statistical association as a causal criteria?
correlation does not equal causation. correlation may occur statistically due to a counfounding variable that causes a spurious relationship . correlation is fine if yourpurposes are just prediction. However if the goal isto understand Why something happens or Manipulate variables to change outcome then you need to determine causality
poor choice of comparison groups
groups not counterfactual
example: in cohort study E negative group must be comparable to E positive group with respect to risk factors for atcome related to exposure of interest , le CFV’s
example: c-c studies , control group must reflect prevalence of exposure (risk based) or proportion of exposed person-time at risk (rate based) in The non-case members of source pop.
non-response bias
association between E and O differs in responders from non- responders
who isn’t answering and why?
selective Entry/survival bias
“healthy worker effect”
Individuals are highly selected because removal of possible sample Units from Original pop may be highly correlated with exposure and outcome
Follow-up Bias
Differential loss to follow up that is related to exposure status and outcome.
Also includes Hawthorne effect.
Detection Bias
sampling Bias in case control studies : occurs when exposed are screened for disease more frequently than non-exposed.
Misclassification (info) Bias in cohort studies: occurs if those assessing outcome know exposure Status and this Influences how they classify outcome.
Admission Risk Bras
AKA Berkson’s Bias
occurs in C-C studies- secondary base
probability of admission is related to both disease and exposure
results in controls having and excess or deficit of exposure relative to the source pop
How to deal with confounding
- Restricted sampling
- Matching
→ frequency matching : same freq of potential Cfr in both groups
→ pair Matching: one or more controls 1s individually matched ( matched analysis required) - Statistically
→ Mantel-Haenzsel
→Matche 1:1- McNemar’s Chi- Square
→ Regression
simple Antecedent
occurs temporally before exposure and is causally associated with outcome only through exposure variable
when this variable is controlled it does not change association between E and O.
Exposure-Independent variable
Exposure and extraneous variable have independent relationship to outcome but no relationship to each other.
occurs when matching in Cohorts thus do not bias estimate and do not need to be controlled for
Explanatory Antecedent: complete confounding
extraneous variable proceeds and causes (predicts) bom exposure and outcome.
when extraneous variable added to model association between E and O becomes notsignificant because extraneous explains the original association Y
Explanatory Antecedent: Incomplete confunding
extraneous variable causes both E and O but E also causes O
including extraneous reduces residual variance
Intervening variable
is on the causal or temporal path
DO not control for.
Distorter variable
Structurally the same as explanatory antecedent except at least One causal effects is a different sign than the other ( le causal arrows reflect prevention causation).
need to contro l extraneous variable
Distorterscan reverse association ( a significant positive association can become significantly negative).
suppressor variable
Occurs when exposure and extraneous variable are part of a global variable. The effect of exposure onatcome is lost with Other variables
Moderator variable
produces Statistical Interaction The causal structure is that exposure causes atcome but this depends on extraneous variable.
Interaction
False Negative / positive fraction
quantify errors in diagnostic tests
ex: if someone has a disease there is a10% chance that the test will be negative (false negative fraction).