Epidemiology
study of distribution and determinants of health related states, AND application to control health problems
Morbidity
diagnosed with disease
Mortality
dying from disease
Study designs
Descriptive
Analytic
What (4 things) can epidemiology provide
describes health status of a population (descriptive epidemiology)
offers way to figure out causations of disease
evaluation of treatments
identify natural history and prognosis
Descriptive study questions
when is it?
who is it?
where did it happen?
what is it?
analytic study questions
How did it happen?
why did it happen?
Incidence rate
NEW cases in a period of time
measure of disease burden
# new cases in 1 year/avg number of people in the population in the same year
# new cases in a specified period/total person-time when people were at risk of getting the disease
Prevalence
TOTAL cases AT A POINT in time
measure of disease risk
one point in time
number of people with the disease at one time/total number of people in the population at that time
incidence rate x average duration of disease
Standardized rates
allows for comparison between groups
standardized between age, gender
where do we get data?
census
vital statistics
disease registries
health records
surveys
Experimental/analytic study designs
randomized controlled trial
Observational study designs
case control
cohort
cross sectional
Case control study
Identify case, match with controls
Disadvantage: recall bias, difficult to select controls, reverse causality
Advantage: good for rare/uncommon disease, quick, economical, multiple exposures
gives ODDS RATIO
Case control + cross sectional selection bias
volunteers
low response rate
ascertainment
Case control + cross sectional measurement bias
interviewer
recall
cohort + RCT measurement bias
interviewer
immortal time bias
misclassofication of variable
cohort + RTC selection bias
loss of follow up
ascertainment
healthy worker effect
Cohort study
Follow group of people over time
disadvantage: time consuming, expensive, loss of follow up
advantage: no recall bias, no reverse causality, explore exposures which change over time, direct measure of incidence, test multiple effects
gives RELATIVE RISK
Cross sectional study
Looks at a situation in one point in time
advantage: one point in time, less time consuming, little ethical problems, study multiple exposures
disadvantage: reverse causality (did x cause y or did y cause x), recall bias
gives ODDS RATIO
Randomized controlled trials
best way to ensure groups are exchangeable
strengths: best to avoid confounding, no reverse causality, direct measure of incidence
weaknesses: long time, high cost, ethics
Ethics principles
Do no harm
do good
autonomy (informed consent)
justice/equity/fairness/impartiality
Odds ratio
<0.5 = less outcome when exposure increased
1 more outcome when exposure increased
Association
statistical measure of risk
NOT causation
3 factors to consider before accepting as truth
random sample error
measurement bias
selection bias
Statistical power
having enough people to find a true association
however, having a huge sample every time is not feasable
complete sample size calculation to find sample size large enough to find true association
Measurement bias
Biases arising from method of measurement
prevent using standardized instruments (i.e., standardized questionnaires)
CBC
chance
bias
confounding
alternative explanations for study findings, must be addressed before accepting findings
Confounder
mixing or muddling effects when the relationship is confused by the addition of another factor
illusory/biased association between exposure and outcome, which may be due to a 3rd factor
how to control for confounders
good matching
restriction (do not include confounders)
randomization
How do adjust for confounders
done during analysis
stratification
multivariable modelling
3 features of confounders
NOT on causal pathway
risk factor for disease
associated with the exposure of interest
Census
information from every person in a population
vital statistics
birth records
death records
marriages
Disease registries
i.e., cancer registry
data from practitioners and pathology labs
most diseases are poorly reported, only cover minority of conditions
Family doctor data
electronic medical record
hospital data
admissions data
ER visits
surgery data
ONLY WHEN ADMITTED
surveys
collects data from a sample of a larger population
Canadian community health survey
Information found on census
age
sex
education
location
family structure
Odds ratio definition
ratio of disease to non disease
can get from case control and cross sectional study
(exposed with disease x unexposed without disease)/(exposed without disease x unexposed with disease)
Relative risk
Ratio of disease incidence in exposed vs unexposed
(exposed with disease/(all exposed))/(unexposed with disease/(all unexposed)
incidence propotion
proportion of a defined population that develops the outcome of interest in a specific time period
number of people who develop the disease in a period/number of people at risk at the start of the period
ecological study
compares levels of exposure/disease across populations, not individuals
Crossover trial
participants serve as own controls
removes variability that cannot be eliminated by randomization
N of 1 trial
single patient trial
one individual receives both experimental and control treatments in a random order
cluster randomized controlled trials
people randomized together
used when intervention cannot occur at individual level
community trials
cluster trial, intervention is implemented at community level
Confidence interval
range placed around a point estimate, in which the true result likely lies
95% CI will likely contain the true value 95% of the time
narrower CI = less sampling error
Selection bias
selection of the study sample differs from the population in a non-random directional way, and relates to the exposure
How to minimize selection bias
clearly define study population
study sample represents population of interest
High response rate of cases and controls
Measurement bias
systematic or non-random error in the collection of data
ROAR: recall bias, observer bias, attention bias, response bias
how to reduce measurement bias
blinding RCTs
standardized instruments
quality control
Statistical significance
unlikely to arise by chance
Practical significance
useful practically
statistical vs practical significance
in large study: small differences that are not clinically useful come up as statistically significant
in small study: a clinical use may arise, but is not statistically significant, so hard to know if result was by chance
How to control for confounders through study design
SMR
matching
restriction
How to control for confounders through analysis
ASM
stratification
multivariable modelling