New cases over time.
Measures the occurrence of new outcomes among people at risk.
Side 42
Epidemiology
A study of health patterns in populations. Epidemiology measures who experiences an outcome, when and where it occurs, which exposures differ between groups, and how confidently those differences support causal or preventive conclusions.
population→exposure→outcome→comparison→inference
06core measures
05study designs
06bias threats
42Side
Population health begins with denominator discipline.
Counts alone do not reveal risk. Epidemiology relates events to the population and time at risk.
Existing cases at a point or period.
Reflects both incidence and how long the condition persists.
Probability over a defined interval.
Requires a clearly defined population initially at risk.
Events per person-time.
Useful when follow-up time differs across individuals.
Deaths relative to population or person-time.
Cause-specific mortality depends on consistent classification of deaths.
Deaths among identified cases.
Depends heavily on who is diagnosed and how cases are defined.
Study design determines what comparison is available.
Different designs trade speed, cost, control and susceptibility to bias.
| Design | Starts with | Strength | Main caution |
|---|---|---|---|
| Cross-sectional | Population snapshot | Prevalence and associations | Weak temporal ordering |
| Cohort | Exposure groups | Incidence and temporal sequence | Confounding and loss to follow-up |
| Case-control | Outcome status | Efficient for rare outcomes | Selection and exposure recall |
| Randomized trial | Assigned intervention | Strong causal identification | Generalizability, compliance, ethics |
| Ecological | Group-level exposure/outcome | Population patterns | Group associations may not hold for individuals |
Association needs both relative and absolute scale.
A large relative effect can correspond to a small absolute change when baseline risk is low.
Risk ratio
Compare probabilities.
Risk in exposed divided by risk in unexposed.
Risk difference
Compare absolute probability.
Exposed risk minus unexposed risk shows absolute excess or reduction.
Rate ratio
Compare event rates.
Useful when groups contribute different amounts of person-time.
Odds ratio
Compare odds.
Common in case-control studies; can diverge from risk ratio when outcomes are common.
Attributable burden
How much population outcome relates to exposure?
Depends on both effect size and exposure prevalence.
Relative risk is not absolute risk.
Communicating both helps prevent dramatic percentage changes from obscuring a small baseline probability.
Epidemiologic bias often enters before modeling.
The central question is whether compared groups differ in ways that distort the exposure–outcome relationship.
Entry or retention depends on relevant factors.
The analyzed sample may no longer represent the intended comparison.
A third factor influences exposure and outcome.
The crude association can mix causal effect with baseline differences.
Exposure or outcome is labeled incorrectly.
Differential error can distort estimates in unpredictable directions.
Memory differs by outcome status.
Cases may remember prior exposures differently from controls.
One group is observed more intensely.
More testing can produce more diagnosed cases without more underlying disease.
Conditioning creates an association.
Restricting on a common effect of exposure and outcome-related causes can induce bias.
Outbreak investigation is inference under time pressure.
The goal is to characterize spread, identify likely sources and interrupt transmission while evidence is still incomplete.
Is there truly excess disease?
Observed versus expected.
Changes in testing or reporting can mimic an outbreak.
What counts as a case?
Person, place, time, criteria.
Case definitions trade sensitivity against specificity.
Who, where and when?
Epidemic curve + map.
Descriptive patterns generate hypotheses about source and transmission.
Which exposures distinguish cases?
Cohort or case-control analysis.
Analytic studies test hypotheses generated by the descriptive phase.
Which intervention reduces spread?
Act before perfect certainty.
Public-health control often proceeds alongside continued investigation.
Read the design before the headline.
Effect estimates mean little without understanding the population, comparison and sources of bias.
“Exposure X doubles disease risk.”
Ask for baseline risk, absolute difference, confidence interval, exposure definition, confounder control and whether the design establishes temporal order.
“Cases rose 40% this month.”
Check testing volume, reporting changes, seasonality, population size and whether the comparison month is representative.
“A screening program found more disease.”
More detection can reflect earlier or more intensive diagnosis rather than increased underlying incidence; lead-time and overdiagnosis may matter.
“The trial was randomized.”
Still inspect allocation concealment, attrition, adherence, outcome definition, blinding where possible, and whether the trial population resembles the population of interest.
Modern EpidemiologyRothman, Greenland & Lash · methods
EpidemiologyGordis · introductory foundation
Causal Inference: What IfHernán & Robins · causal epidemiology
Field Epidemiologyoutbreak investigation and applied practice