exposure and outcome have common causes − the exposed are not exchangeable with the unexposed − association is not causation [Confounder] Definition of a confounder is: ・ Associated with the exposure (A) in the study base that produce the case ・ Associated with the outcome (Y) in the non-exposed ・ Is not a downstream consequence of exposure or outcome * Not all variables that have these properties are confounders [M-bias] ・ Collider-stratification bias arising from conditioning on a collider that opens a path from exposure to outcome. ・ In the presence of M bias, even if one had stratified on the collider, it may still be possible to estimate a causal effect of exposure on outcome. [Selection bias] ・ Inappropriate selection of controls in a case control study ・ As a result of factors that influence continued participation of subjects in a study. ・ Differential loss to follow-up in longitudinal study ・ Non-response bias, missing data bias ・ Volunteer bias/ self-selection bias ・ Healthy worker bias ・ Conditioning on a common effect − procedures used to select subjects − factors that influence study participation ・ The relation between exposure and disease is different for those who participate compared to all those who should have been theoretically eligible for the study − Any types of study, selection bias can occur
− Case-control: selection of case/controls is dependent on exposure − Cohort: success of tracing subject, selection into the study [Collider-stratification bias]
[Experimental Studies] ・ Investigator assigns exposure Clinical Trial ・ Randomization: Exchangeability (+), Confounding↓ ・ Blinding: ↓Potential for bias Field Trial ・ Primary prevention e.g., vaccine trials Community intervention trial ・ Community intervention studies e.g., fluoride in drinking water [Non-experimental Studies] ・ Observational study: Investigators do not assign exposure Cohort ・ Define 2 or more groups: free of disease, but at risk ・ Measure the occurrence of the disease ・ Start a single group with heterogeneous exposure P-T,IR, CI, OR IRR etc Case- control ・ Identify the cases ・ Controls (sampled independently exposure) ・ Sample from the study base ・ Determine the exposure distribution ・ Estimate relative size of the exposed & unexposed Ratio (OR) Cross- sectional ・ All person in the population ・ At the same time ・ Disease prevalence ・ ×etiologic research preval ence Proportiona l mortality ・ Only dead subjects ・ The Proportion Mortality Ratio (PMR) The proportion of death due to a specific cause among the exposed and among the unexposed PMR Ecological ・ Information on exposure and/or disease on a group level (not individual level) ・ No information on if “exposed” are getting disease ・ Aggregate association may not reflect ind. Assoc. Scatte r plot
[Prospective vs. Retrospective] Prospective: Exposures are measured before the outcome events occur Retrospective: Exposure are measured after the outcome events occur [Person-Time of Observation] ・ Sum of time at risk in the cohort study for each person ・ Can be computed in closed & open cohorts [Closed & Open Cohorts] Cohor t Closed Open How Membership defining event No New member can enter Lose members: Outcome (death) Membership: defined by “state” Can enter and re-enter Can exit Good CI can be measured directly
You've reached the end of your free preview.
Want to read all 13 pages?
- Fall '18