Epidemiology Midterm 2

Necessary (causal relationship)

disease will not happen without this factor

Sufficient

with this factor, the disease will exist

Characteristics of a cause

time order (cause then outcome), association (correlation between cause and outcome), direction (change in cause —> change in outcome)

Counterfactual

difference in effect with and without cause

Randomized Control Trial

accounts for inherent missing data of potential outcomes, only possible differences come from intervention, differences after randomization, or random chance

strong RCT

excludability (can exclude other factors), non-interference (treatment of one unit doesn’t affect other units)

How to randomly assign groups

participants assigned to treatment/control groups randomly

stratified randomization

stratify by potentially confounding characteristics

cluster randomization

randomize hospitals/schools/etc.

crossover design

patients switch between treatments/others - need to have effect be temporary

non-inferiority/equivalence designs

test if one treatment is the same or better than an existing one

issues with RCT

non-adherence to treatment, unethical (?), impossible, against policy

to assert an association

accurately measure outcome and exposure, see if outcome is different w/ or w/out exposure

cohort study participants classified based on

exposure

cohort studies look at associations with

incident disease

case-control study participants classified based on

outcome

case-control studies look at associations with

prevalent disease

options in selecting population for cohort studies

select on basis of exposure, or select defined population

prospective cohort studies

select population, sort into who has exposure/follow to see who gets exposure, follow over time to see who develops outcome

prospective pros/cons

pros: know exposure predates outcome

cons: expensive, long time to findings

retrospective cohort studies

identify population with past exposure, identify who has disease

retrospective pros/cons

pros: faster and cheaper

cons: aren’t collecting data yourself, data may not exist, difficult follow-up

relative risk

proportion who develop disease among exposed / proportion who develop disease among unexposed

incidence rate ratio

#of cases of disease per person-year of exposed / #of cases of disease per person-year of unexposed

*used if participants are in study for varying amounts of time

hypothesis testing

null: there is no difference in groups

alternative: there is a difference

(assuming we accounted for other reasons for difference)

statistical vs. clinical difference

with large sample size, small associations can be statistically significant, but not significant enough to address clinically/practically

case-control study design

identify people with and without disease, determine past exposure

case-control, prevalence or incidence cases?

prevalence: easier to find, more bias because factors/exposures could be because of development of disease

choosing control population

should be from same population as cases, and matching on confounders

sources of controls

hospitals, dead controls, friends/neighbors, random population

group matching

match on general characteristics: ex. 25% male cases —> 25% male controls

requires cases to be found before controls

individual matching

one case has x, y, z characteristics —> one control has x, y, z characteristics

challenges with matching

eliminates ability to investigate effect of the matching characteristics (since it is balanced), cases must be found before controls, hard to find individual matches, may accidentally match (if matching race, then accidentally match on geography)

strengths of matching

easy to do, only matching can really manage some factors that are strongly associated

weaknesses of matching

can be impossible, limits how representative of the population the study is, can’t be undone if you accidentally match on variables of interest

case-control pros/cons

pros: cheap, fast, multiple potential exposures, better for rare diseases - don’t have to wait for them to develop

cons: can’t always determine time-order, exposure associated with survival and not incidence, inaccurate recall, comparison groups harder to define

odds

probability of something happening/probability of something not happening

odds ratio

odds of exposure among cases/odds of exposure among controls

(a/c)/(b/d)

odds ratio interpretation

greater than 1: exposure associated with more disease

odds ratio can be estimate of relative risk/prevalence if:

cases representative of all people with disease in the population with regard to exposure history, and disease being studied doesn’t occur frequently among people unexposed

odds ratio with matched pairs

b/c

cross-sectional study design

identify population, identify exposure and disease status

*use relative prevalence

cross-sectional pros/cons

pros: population-representative, repeated shows changes in population

cons: survivor bias, impossible to know time-order

population attributable risk

(incidence in pop - incidence in nonexposed) / (incidence in pop)

attributable risk among exposed

(incidence in exposed - incidence in nonexposed) / (incidence in nonexposed)

guidelines for causality

temporal relationship, strength of association, dose-response relationship, replication of findings, biological/sociological plausibility, consideration of alternate explanations, cessation of exposure, consistency with other knowledge, specificity of the association

dose-response relationship

increase in disease as increase in exposure can support causal relationship

cessation of exposure

if exposure stops, disease stops (not necessary for causal, but can support - like inverse dose-response relationship)

specificity of associations

dumb, one exposure causes one outcome (and one outcome associated with one exposure)

confounding variable

causes outcome of interest, associated with exposure of interest, not in direct causal pathway

real association, but not causal

address confounding

study design (matching, randomization), analysis (stratification, adjustment - like weighted average)

selection bias

systematic error in selection of participants, selections predispose an association

nonresponse bias

groups with particular illness/exposure more likely to participate

information bias

means of collecting information is inadequate

types of information bias

recall, rumination, reporting, wish, interviewer

observer bias

lean one way with ambiguous cases

respondent bias

participant errs in if they have the disease

types of combined selection/information bias

detection (groups more likely to be detected), survival (actually associated with survival not cause), temporal (factor results from disease, doesn’t cause it), lead-time (screened earlier affects survival time)