Definitions Flashcards
Clinical Epidemiology
- the basic science of EBM
- study of distribution and determinants of health related states and events in specified populations, and the application of this study to control of health problems
- using scientific methods to make predictions/improve pt outcomes
- epidemiological concepts change over time (biostatistic concepts do not)
The 5 D’s
Death: bad outcome if untimely
Disease: set up symptoms, physical signs, lab abnormalities
Discomfort: symptoms such as pain, nausea, dyspnea, etc
Disability: impaired ability to go about usual activities
Dissatisfaction: emotional reaction to disease and its care
Prevalence vs. Incidence
P: current cases of outcome, proportion of total cases to total pop; burden of disease (how widespread the outcome is)
I: new cases, reflects risk of getting the disease; when time is in numerator is incidence rate
Point Prevalence vs Period Prevalence
Point: time period is instantaneous
Period: longer time periods (but time not in the denominator)
Cumulative Incidence
proportion of group that develops disease over a given period of time
= # new cases/# people at risk of developing disease over defined time
Incidence Rate and Incidence Density
IR: rate at which new disease has occurred in the population at risk per some unit time
ID: refers to IR in dynamic, changing pop in which ppl are under study and at risk for varying periods of time; number of cases over person-years
IR (ID) = # new cases/total time experienced by the pop at risk
Systematic vs. Random Error
systematic: bias, compounding, within the study design, sometimes unavoidable; can differential (misclassification unevenly) or non-differential (bias toward the null)
random: occurs due to variations in people, in their responses; non-differential (ex: misclassification)
Epidemic
Outbreak&Pandemic
(epidemic curve)
increase in incidence of a disease in a community or region
O: small, in limited region, P: crosses many international boundaries
(a plot of the distribution of cases over time)
Endemic
the constant presence of a disease or infectious agent within a geographic area or pop
Absolute Risk, Absolute Risk Difference
AR = Incidence (I) ARD = Iexposed - Iunexposed; describes someone's increased risk for a particular disease
Relative Risk
RR = Iexposed / Iunexposed
- evaluates the strength of an association btwn exposure and disease; relative to all other cases
- aka risk ratio
- value of 1 = no difference, >1 = greater risk,
Random vs Probability Samples
R: every person has an equal chance of being sampled
P: every person has a known, though not necessarily equal, chance of being sampled; can weight the sample toward some low frequency groups of interest
Relative Risk
ratio of incidence in unexposed group to incidence in exposed group
Random Error vs Bias
- random errors likely to cancel each other out as # of measurements increases (i.e. bigger sample size); bias will not
- chance more likely to lead to type ii error; bias more likely to lad to type i
Confounder
- must meet three rules:
1) must be associated with exposure
2) must be independently associated with outcome
3) must not be within a causal pathway btwn the exposure and the disease - distorts the association btwn exposure and outcome; Type I Error if distorts toward strengthening association; Type II Error if distorts toward weakening association
Randomization
attempt to evenly distribute potential confounders; does not guarantee control of confounders
Restriction
ex of smoking as confounder btwn alcohol consumption and lung cancer
- prevents confounding but reduces study size –> could decrease statistical significance
- cannot evaluate effect of excluded variable after restriction
Stratification
- data broken down (stratified) by the potential confounder –> removes the effect of the confounder
- if confounding present: risk ratios in strata will be lower than in the unstratified data
- if risk completely due to confounding: would be no diff in risk within the strata
Matching
- for each subject in exposed group, one or more subject (with or without confounder) is chosen for unexposed group
- eliminates effect of confounder at individual level
- has to be coupled with matched analysis
- practical limitation on number of confounders to be matched on
- once matched, the effect of the variable on outcomes cannot be evaluated
Effect Modification
- effect of confounding factor (ex of birth order, maternal age, and DS in 3D graph)
= “interaction” - if stratum-specific risk ratios are DIFFERENT, its effect modification
-effect modifiers are variables that change the effect of exposure on risk of disease
Multivariable Adjustment
- allows us to look at multiple confounders simultaneously
- use regression analysis; if results are close to the null, know confounders are important
Selection Bias
- often in cohort studies
- selective differences btwn comparison groups that impacts the relationship btwn exposure and outcome
- often results from comparison groups NOT coming from same study base and NOT being representative of their pops
ex: “healthy worker effect”
Self-Selection and Withdrawal Bias
SS: volunteers (ex: asbestos retrospective cohort study)
WB: loss to follow up, differential attrition leads to selection bias (“survivorship bias”)
both in cohort studies
Information Bias
- investigators who know exposure status (ex: radiologist looking at pt who’s a smoker)
- subjects who know exposure status (may be more likely to report potential symptoms)
- remedy: BLINDING
Sensitivity vs. Specificity
Sensitivity: pos when it should be pos = TP/ (TP+FN)
Alpha vs. Beta
A: our willingness to be wrong - our willingness to reject the null when we shouldn’t, to make a Type I Error
- usual convention is p =.05 (so we’re 95% certain; willing to be wrong 1 in 20 times)
B: our willingness to tolerate failure - to make a Type II Error
- usual convention = .1 or .2 (so we’re more willing to make a Type II Error than Type I), but sometimes not specified
- if we want 80% power to reject null, our beta is 20%
Power
- ability to detect or verify a difference that is real, to avoid a Type II Error
P = 1 - beta - if you reject null, then by definition, you cannot lack power (even if small sample size)
- is the sensitivity of our study
T-Test
- compares the diff btwn the two means
- divided by variability in the two samples
T = (meanA-meanB)/(varA+varB)^.5
df = (nA-1) + (nB-1) - if we calculate t at less than the critical value (based on alpha and df), then we fail to reject the null
- in a graph: the wider the curves, the less likely they’ll be stat sig
Gaussian Distribution
“normal curve”
bell shaped, symmetrical about the mean
mean = median = mode
2/3 of observations fall within 1 SD of mean, about 95% within 2 SDs
What are the three criteria for determining whether an observation is abnormal?
- is it unusual?
- is it associated with disease?
- does labeling and treating do more good than harm?
