Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
85 Cards in this Set
- Front
- Back
Lecture 1 Epi definition |
The study of the distribution and determinants of health-related states or events in specified populations over specified periods of time and the application of this study to control and prevent health problems. |
|
Three primary aims of epi research |
Describe, explain, predict ...disease in populations |
|
How does epi differ from other research fields and clinical medicine? |
Focus on etiology and prevention in humans |
|
Descriptive vs analytic |
Descriptive = distribution of disease Analytic = risk factors (determinants) for disease |
|
Purpose of descriptive studies |
-Prevalence (burden) of disease -compare subgrops -basis for planning and prevention -generate hypotheses -calculate measures of disease frequency |
|
Purpose of analytic studies |
-determinants of disease -test etiology or preventative hypotheses -suggest potential for health promotion or disease prevention |
|
Populations: Target, Source, Eligible, Study |
Target: population to which results can be applied Source: population from which subjects will be drawn Eligible: all eligible to participate Study: those contributing data to the study |
|
Lecture 2 Two key concepts about epi studies |
1) Epi is the study of distribution and determinants of disease in populations 2) Epi is quantitative - measures of frequency and association |
|
Measures of disease frequency characteristics (4) |
1) calculated from descriptive studies 2) used to enumerate occurence of outcome in specified pop over specified time 3) measured for either prevalent or incident cases 4) expressed as count or relative measure |
|
Measures of disease association characteristics (4) |
1) calculated from analytic studies 2) reflect strength of relationship between exposure and outcome 3) reflect excess cases attributable to or preventable by the exposure in specific pop over specific time 4) compares disease frequency between 2 or more groups at different exposure levels |
|
Disease counts uses and importance |
-Used to monitor occurence of outbreaks (ex. 1 zika case is low risk, but needs to be known) -Important to health planners (CDC) |
|
Risk vs rate basic definition |
Risk = probability of developing disease over specified period of time (dimensionless) Rate = speed at which people develop disease (time dimension) |
|
Cumulative incidence definition |
CI is a RISK. Prob of developing a disease, conditional on survival throughout interval. Range 0 to 1. |
|
Incidence density definition |
ID is a RATE. Instantaneous occurrence of new case relative to population at risk. Range 0 to infinite. |
|
Simple CI method equation |
# new cases during f/u time / population at risk f/u time |
|
Simple CI method assumptions (3) |
-No/few losses to f/u --> underestimate risk -period of risk is short -closed population all followed for full period |
|
Advantages of a Life Table |
Can calculate interval based risk and overall survival probability. Helps to deal with losses to f/u and changing rate of disease during f/u |
|
Assumption for synthetic fixed cohort |
Calendar time does not matter. If something sig happend drug f/u (ex. new drug), must split into multiple cohorts |
|
Life table assumptions (4) |
-Rate of disease of withdrawals is same as rate for those who remain in f/u -On avg, withdrawals happen midway through interval -Ppts survive at risk for entire f/u (risk conditional on survival) -No secular trend in risk |
|
Kaplan-Meier method differences from Life Table (3) |
-Don't organize into intervals - unable to answer risk questions about specific intervals -Risk is estimated at time each case occurs -No assumption about uniformity of withdrawals |
|
Censoring assumptions (2) |
-Censored ppts have same survival prospects as those who remain in the study -Cause of censoring cannot be related to the study (can investigate by looking at those lost) |
|
Definition of survival analysis |
Statistical procedures where outcome of interest is time until event occurs, not just whether it occurs |
|
Incidence density method equation and use |
CI = incidence rate x time period (time units cancel) Used when there is a dynamic cohort with rates available and usually with long period of risk |
|
ID method assumptions (4) |
-Constant rate of disease -Closed population -No competing risks -Low risk (small number of events) Otherwise, tends to overestimate CI |
|
Lecture 3 Incidence density definition and equation |
Incidence per person-time (RATE) ID = I / PT |
|
Assumptions for ID estimation with aggregate data when exact PT is not known |
-Events and censorings occur uniformly -Those who have events of withdraw can be considered to contribute PT for half the interval -Open or dynamic cohort and steady state (ex. no massive death or migration) |
|
Important assumptions about ID (5) |
-independence between censoring and survival -lack of secular trends -Risk is approx constant over study period -Steady state -ID for individual and aggregate data will be about equal when withdrawals and events occur uniformly (common disease --> underestimate true rate since denominator is inflated) |
|
Stratification of PT |
Method for rectifying the changing rates of disease over time, which violates the assumption for ID calculation. Can use stratified rates over smaller intervals to calculate ID for periods of time with relatively constant rates |
|
Lecture 4 Ratio vs absolute risk |
Ratio: strength of association (contributes to causal relationship) Absolute: excess risk, preventing disease (practical to PH policy) |
|
Assumptions for absolute differences (2) |
-Relationship is causal -everyone is at risk |
|
Distinct characteristics of OR |
OR of exposure = OR of disease OR event is reciprocal of OR nonevent |
|
OR built in bias |
Always away from the null |
|
Equations for attributable risk (2) |
AR = CI(ex) - CI(unex) -- dimensionless IDD = ID(ex) - ID(unex) -- per unit time |
|
Definition of percent attributable risk in the exposed |
Proportion of total risk in the exposed group attributable to that exposure (ex. What percent of cancer among smokers is attributable to smoking) |
|
AR% Equations (3) |
CI(ex) - CI(unex) / CI(ex) CIR - 1 / CIR ----- Can use OR or IDR ID(ex) - ID(unex) / ID(ex) ALL MULTIPLIED BY 100 to get % |
|
Population attributable risk definition |
PAR: what proportion of risk in the population being studied is attributable to the exposure NOT generalizeable to all populations because freqency of exposure varies |
|
Using %AR for intervention |
Group receiving intervention is considered "non-exposed", which hopefully lowers incidence of disease. %AR considered equivalent to efficacy within intervention trials |
|
When we use relative differences vs absolute |
Relative = determinants (strength; establish causality) Absolute = PH policy (once causality established) |
|
Lecture 5 Distinguishing characteristic of an observational study? |
Subjects assign themselves to exposure groups |
|
Descriptive vs analytic |
Descriptive: natural hx of dz, disease frequency, disease disparities Analytic: test hypotheses, measures of association |
|
Distinguishing characteristic of cohort design? |
Ppts non-diseased at baseline |
|
Advantages of cohort design? |
-Less prone to selection bias improving validity -Temporal relation can be established -Post marketing surveillance to identify adverse events -Many outcomes can be studied -Provides direct estimates of incidence and changing rates over time |
|
Disadvantages of cohort design |
-Costly, slow -losses to f/u threaten validity -inefficient for rare diseases or induction periods |
|
Population-based cohort? |
Ppts identified based on probability sampling of the general pop or subgroup. Ppts selected without regard to exposure status |
|
Advantages of population-based cohorts? |
Generalizability due to random sampling -Strong internal validity -Multiple hypotheses (exposures) can be tested |
|
Disadvantages of population-based? |
-expensive -logistically complex -can have large losses to f/u -Inefficient if exposure is rare |
|
Special-exposure cohort design |
Ppts identified based on exposure status |
|
Advantages of special-exposure cohort? |
-ensures exposure of interest (good for rare) -Minimized losses to f/u -may have relevant info on exposure and confounders in records |
|
Main disadvantage of special-exposure? |
May lack generalizability |
|
Key issues to consider when measuring exposure? |
-intensity and duration (dose-response) -induction and latency periods -time-variable exposure -categorization of exposure |
|
Induction vs latency period |
Induction (biologic) - exposure to biologic initiation of disease Latency (screening) - time from initiation of disease to diagnosis |
|
Important characteristic of induction period from the standpoint of epi studies? |
A person is NOT at risk for a disease during the induction period since they have already been exposed - must exclude from the person-time calculations (Ex. 3 year induction period for radiation to cause cancer. That 3 years is not included in PT)
|
|
Time-variable exposures |
Exposure changes over time. A person can contribute f/u to several exposure rates. (Ex. smoking 1 pack/day for 5 years then 1 pack per week for 5 years) |
|
Categorization of exposure |
Grouping exposure into categories to make them more clinically relevant (ex. Blood Pressure - one point doesn't matter, but normotensive vs hypertensive) |
|
Lecture 6 Distinguishing characteristic of a case-control design? |
Ppts selected based on their disease status |
|
Advantages to case-control? |
-PRIMARY: Efficient (money, time, # of ppts)
-OR is flexible -Good for rare disease -Good for long induction or latency periods -Can study many exposures |
|
Disadvantages of case-control |
Particularly susceptible to bias: selection bias (represent source pop, selective survival), information bias (accurate collection of exposure), confounding |
|
OR approximates RR when...? |
Disease is rare (<10%) OR controls are selected to represent the source population, not just the non-cases |
|
Valid subject selection for controls within case-control requires...? |
Controls selected so that their exposure distribution reflects the exposure distribution in the source population (i.e. the underlying cohort). |
|
Both cases and controls must be... |
selected INDEPENDENT of exposure status (i.e. same sampling rate for exposed and unexposed) |
|
Types of case selection within case-control |
-Incident or prevalent (incident better) -cross-sec or longitudinal (long is better otherwise we have selective survival) |
|
Two types of sampling strategies to select controls? When are they typically used? |
-Traditional (cumulative) sampling: used in traditional case-based case-control design (primarily used with disease is rare, since selecting unbiased controls can be difficult) -Incidence density sampling: used within a hybrid case-control design. OR approximates RR without rarity assumption because controls selected from baseline cohort |
|
Case-cohort design |
Cases selected as they occur (density sampling). Controls selected by a random sample of the total cohort at baseline. Controls can become a case. OR approximates CIR. |
|
Nested case-control design |
Cases selected as they occur (incidence density sampling) and matching control selected concurrently. OR approximates IDR. |
|
Lecture 7 Distinguishing characteristic of ecologic study design? |
Unit of analysis is at the aggregate level |
|
Types of subject groupings |
-Multiple-group design (by place) -Time-trend design (by time) -Mixed-design (by place and time) |
|
Advantages of ecologic design |
-Low cost, convenient -No measurement limitations of individual level studies -No design limitations of individual level studies like limited variability -Interested in ecologic effects |
|
Disadvantages of ecologic design |
-Collinearity -Temporal ambiguity -Confounder control -Ecologic fallacy |
|
Purposes of ecologic studies |
-Generate etiologic hypotheses -Study broad or cultural factors -Alternate to collecting expensive or sensitive data from individuals (ex. income) -Testing impact of group wide interventions |
|
Distinguishing characteristic of cross-sectional design? |
Anyone can be included - doesn't matter the disease, exposure status etc. Simply a random survey. No f/u. |
|
Uses of cross-sectional study |
-Measure disease burden -Generate hypotheses -Health planning, services |
|
Lecture 8 Purpose of crude data analysis |
Inform multivariate analysis to determine potential confounders, interactions. Must clean data first (outliers, ~N, missing/incorrect data)
|
|
Three basic parts of data analysis |
-Point estimate -Interval estimate -Hypothesis testing |
|
Point estimate purpose |
Reflects the strength of association (RR, OR) |
|
Interval estimate purpose |
Reflects the precision and variability of the point estimate (Confidence interval) |
|
Hypothesis testing purpose |
Examines the role that random error may have played in our measure of association or impact using P-value |
|
Three components of hypothesis testing step |
Error, P-value, Power |
|
P-value definition |
Probability, assuming the null hypothesis is true, that the data will show an association by chance alone. |
|
Best way to 'think about' p-value |
P value is the compatibility between the data and the null hypothesis |
|
The P-value is NOT... |
-Level of significance -Probability that the null is true -Alpha -Indicative of the magnitude of association -Probability of the observe data under the hypothesis |
|
Power definition |
Probability of rejecting the null when it is false. Stated another way, probability of finding an association is significant when in truth it is significant. |
|
Presenting data: table 1 |
Table 1: Descriptive stats: general characteristics, demographics of study pop along with univariate analysis. Include n's for each group. |
|
Presenting data: table 2 |
Crude data analysis with point and interval estimates and hypothesis testing results for all vars
|
|
Presenting data: table 3 |
Adjusted multivariate analysis controlling for vars of significance in table 2 |
|
Elements of interpretation |
Risk (point estimate), exposure, group, comparison group, disease, CI, significance, precision, controlled vars |