Epidemiological study designs Causation

Epidemiological study designs Causation
Epidemiology
• Descriptive - ecological • Cross-sectional • Cohort studies - Prospective - Retrospective • Case-Control • Experimental (intervention)
Epidemiological-Clinical • Human subjects • Clinical - focus on individual disease cases • Experimental - controlled exposures/treatments • Descriptive studies (simply describe) – Longitudinal (historical - follow over time) – Cross-sectional (snapshot in time) • Analytical epidemiology (does a link exist?) – Cohort (start from exposure/treatment and look for disease) – Case-control (start with disease and compare exposures/treatments) • Ecological or cluster investigations – Geographic, time period, certain population
Cohort studies
• Prospective (exposure now - disease to follow up in future) • Retrospective (historical) (exposure in past - disease since then) • Directly measure risk of a disease (illness rate, risk ratio or relative risk) • Large groups needed (100’s, 1000s) - costly • Can assess many risk factors together • Occupational more often • Variant: proportional morbidity (cause of interest), best for uncommon diseases
Cohort study
Case-control studies
• Powerful and accurate • Economical (population size and time) • Estimate odds ratios • Acute, chronic, long latency, rare diseases • Variant: nested design (second phase, narrowed down)
Case-control study
Ecological studies
• Unit of study is population group not individuals • Most publicity • Likely to be incorrect, misunderstood, or by chance • Causation still necessary to prove, but … • Exposed and sick may not be the same individuals • Useful in generating new hypotheses • Inexpensive (existing data can be used) • Exploratory, time-trend, space-time, multigroup, mixed
Rates of disease and disease ratios
(Incidence Rate)
(RR = IRE/IRU)
Measures
• Rate of disease:
• Risk ratio:
• Attributable fraction:
• Odds ratio*:
# cases # total at risk rate in exposed* (E) rate in unexposed (U)
A/C B/D
E - U E
I - U I
or
* treated/untreated* Used in case -control studies
Mortality measure
• Standardized mortality ratios (SMR) Observed deaths Expected deaths x100
Causation
• Not simply an association in numbers but the plausibility that risk factor leads to disease (toxicological basis)
• Lab studies required to supplement epidemiological studies • Large enough study (2x increase, n=300, for ?=5%, ?=20% or 80% chance to detect a true effect)
Does association mean the exposure caused the effect?
Statistical concepts • Central tendency (average value or mean, median) • Accuracy (bounds of likely values, ie. within +/- x%) – Confidence interval (likely range of values) – Depends on # cases, population size, variability, • Errors – Alpha (?): observing the effect by chance alone (see an effect when there is none) – Beta (?): not being able to detect a true effect (lack of sensitivity in detection of effect) • Power: being able to detect a true effect (1-?) (size of study and variability the most influential along with rarity of outcome)
How many people do we need so that we can detect a doubling of incidence?