W3: Cohort and Case-Control Studies (Design and Analysis) Flashcards
What if we can’t follow people over time?
Retrospective study. Less time consuming, but information may be limited and may not be able to determine if exposure preceded outcome.
Propective study
More time consuming, but can collect detailed information about exposure and outcome. Less subject to bias; easier to determine if exposure preceded outcome
What if there is migration in or out of the population during the study?
Can happen in open cohort (eg people who live in Berkeley, CA; people under 18 who smoke; RNs who work in San Francisco).
Fixed cohort
Population is based on inalterable characteristics (eg people who had a certain surgery, or WWII atomic bomb survivors in Hiroshima). People cannot enter/exit cohort. No loss to follow up. Unexposed can serve as counter-factual.
Strengths of cohort design
Efficient for rare exposures. Can evaluate multiple effects on an exposure. Can directly measure disease incidence or risk. Prospective cohort study: good info on exposures, less vulnerable to bias, clear temporal relationship b/w exposure and outcome. Retrospective: efficient for diseases with long induction and latent periods.
Weaknesses of cohort design
Inefficient for rare outcomes. Prospective: expensive and time consuming, inefficient for diseases with long induction/latent periods. Retrospective: poor information on exposures and other key variables, vulnerable to bias.
How to analyze data in cohort study?
Most common to measure incidence in prospective (allows investigators to detect new cases of disease). Also possible to measure prevalence using info at beginning of f/u period to determine who already has disease.
Induction period
After causal action (exposure), etiologic process begins. Eg cancer cells beginning to form. Next is disease initiation
Latent period
After disease initiation, before clinical symptoms show.
Differential loss to followup based on exposure
When loss to follow up differs between exposed and unexposed groups. Could also occur based on disease status.
Case control study
Sample based on disease status rather than on exposure status.
Study base principle
Members of the study base are people who would have been identified as cases over the study period if they had acquired disease. Controls must be selected from the study base that gave rise to the cases. Eg. cases could be childhood leukemia cases in CA between 2010-2017; controls could be samples children w/o leukemia in CA between 2010-2017.
Risk set/density sampling
Good for time-dependent exposures. Controls are sampled from person-time at the time the case becomes a case. Cases and controls are matched on time (special analysis needed). Sample people who are at risk of becoming a case; follow cohort over time; people contribute person-time until they become a case. Sometimes called nested case-control study.
Case-base/Case-cohort sampling
Not ideal for time-dependent exposures. Controls are sampled from the study base that gave rise to the cases. Controls are sampled regardless of time contributed to the study and regardless of later disease development.
Cumulative/survivor sampling
All cases have occurred before study begins. Controls are selected from those who did not have disease at time of study. Greater risk of bias may occur if people who are exposed are lost to followup (differential loss to follow up may cause inaccuracy in measure of association).
Traditional vs modern view of case-control studies
Traditional view: case-controls are backwards, because they compare exposure of disease and non-diseased (reverse of cohort study).
Modern view: method of sampling a population to identify cases and controls in order to compare disease rates in exposed and unexposed. Control group provides information on relative proportion of exposed and unexposed in source population.
Case-crossover design
Cases serve as their own controls (they’re controls until they develop disease, then become a case). Difference between this and crossover experimental study: investigator does not assign exposure, they observe it. Cases and controls are usually identical, except for exposure. Efficient because it requires fewer people. Eg. cell phone and MVC study. Appropriate when a brief exposure (eg cell phone use) causes short term change in risk of rare/acute disease (eg MVC).
Does study design dictate which measures of association can be measured?
YES! Cannot validly estimate all measures of association under every study design.
What can we estimate in a cross-sectional study?
All the measures of association we’ve learned thus far.
What can we estimate in a cohort study?
In a cohort study that samples on exposure status, can only estimate MoA that condition on exposure status. Can estimate RR, RD, APe.
Experimental studies
Can only estimate MoA that condition on exposure status.
Measures of association: RD, RR, APe.
Measures of disease/exposures: Re, Ru
Case-control
Trickiest because sample is based on disease status. Odds ratio for disease approximates the relative risk when the disease is rare, so this is what we use to approximate relative risk.
Ecologic study
Measure exposure and outcome at the population level (no sampling). Can’t fill interior of 2x2 table at all. Can estimate relative risk using group-level data. Never as useful as individual-level relative risk. Can try to estimate individual using statistical models (but beware ecologic fallacy)
