Epi Flashcards
Prevalence formulae*
P = cases / total [@one moment in time] P = incidence x duration
Cross sectional study:
- purpose
- disadvantage
- prevalence at single time point
- no causality can be attributed
Incidence formula*
I = new cases in given time / at risk population in given time
*must subtract numerator from at risk population
Incidence study type*
Cohort study (follows group over time to see who gets case)
Sensitivity:*
- formula
- utility
= TP/(TP + FN)
-ruling out disease if negative
Specificity
- formula
- utility
=TN/(TN+FP)
-ruling in diagnosis if positive
PPV*
- formula
- interpretation
- affected by*
=TP/(TP+FP)
- probability of having disease if test is positive
- Higher specificity of test or higher prevalence increases PPV
NPV*
- formula
- interpretation
- factors
- TN/(TN+FN)
- probability of being disease free if test is negative
- higher sensitivity of test & lower disease prevalence increase NPV
- LR
- formula
- interpretation
= #diseased with -test/ #non-diseased w -test
=(1-sensitivity) / specificity
-how odds of having disease decrease with negative result
Cohort study*
- design
- other names
- advantages*
- disadvantage
- all non-diseased initially ==> followed over time and exposures compared between those who get and don’t get disease
- longitudinal or incidence study
- prospective is low bias; allows INCIDENCE
- time-consuming /costly
Case-control study*
- design
- advantage
- disadvantage
- Cases are compared to non-diseased controls for exposures (selection of controls affects bias)
- study rare diseases; quick
- can’t assess RR or INCIDENCE* (only OR)
Odds Ratio
- when used*
- valid when*
- formula
- case control study as estimate of RR
- low disease incidence
- ad/bc
Risk formulae*
- general terms
- absolute
- attributable (AR)
- attributable risk percentage*
- relative (RR)
- null value of RR
“probability of getting disease of certain time period”
=incidence = a/(a+b)
AR =incidence(e) - incidence(unexposed)
ARP = AR/incidence(e) = (RR-1)/RR
RR = incidence(e) / incidence(unexposed) = a/(a+b)//c/(c+d)
Null =1
*i don’t get why you aren’t subtracting the diseased from total as instructed with incidence formula…but you dont
Type 1 error
- meaning
- associated value
- chance of false +; that significant difference found erroneously
- p-value: chance of association found by random chance alone
Selection bias:
- not controlled by:
- confident that randomization works if:
Two groups differ in characteristic
- matching!
- baseline characteristics (demographics) are same in both groups
Type 2 error
- meaning
- associated variable & formula
- chance of false negative; no association see despite one existing
- power: chance study finds significant difference if one exists = 1-beta
- as power increases, type 2 error risk decreases
Bell curve, 95% fall:
Within 2 SD +/- from mean
68.5% within 1SD; 99.7% within 2SD
Hazard ratio
- formula
- interpretation
-#events in treatment group / #events in control
- HR>1 = more events in treatment
- HR<1 = more events in control
RCT designs
- parallel
- factorial
- cross-over
- cluster
- normal treatment vs control
- multiple treatments at once
- treatment is given to both arms just at different times
- different groups randomized, not individuals
Attributable Risk Percentage
- meaning
- formulae
-%risk attributable to risk factor
=(riskExposed-riskUnexposed)/riskExposed
=(RR-1)/RR
Confounding meaning
How to reduce
A third variable is associated with outcome & exposure
-matching confounders (same # per group), randomization, restricting, stratifying (only selecting non-confounders), Multivariate analysis
Correlation coefficient
-does not imply
R= linear association from -1 to 1; 0 is no correlation, -1 is negative correlation, +1 is positive correlation
-does not imply causality
NNT *
- meaning
- formula[???]
-NNT to prevent bad outcome
=1/ARR* = 1/(incidenceE-incidenceU)
*ARR should be in decimal form, so 4%ARR = 0.04
Chi-square test
Compares categorical variables in proportion
Confounding vs effect modifier
Confounding: bias variable associated with both exposure & disease ==> risk association between groups removed when stratifying
Effect modifier: non-bias variable that affects a risk factor ==> risk association between groups remains only in one group, other other association goes away
T/Z tests compare:
Two means
ANOVA compares:
Compares 3+ means
Mean
Median
-if even:
Mode
Average
Middle #
-average middle 2 numbers
Most repeating
Statistical significance expressed by:
P<0.05 or 95%CI not overlapping
ROC curves
-axes
- increasing sensitivity (y-axis) vs decreasing specificity (x-axis)
- thus, as sensitivity increases, specificity decreases
Hawthorne effect
Changing behavior because being observed
Lead time bias:
Earlier diagnosis makes survival look increased
Lost to follow up affects
Selection bias
Standard 4 square table setup*
Diseased | Non-diseased
Exposure A B
No exposure C D
