Observational Study Design/Sources of Bias

case report

description of patient clinical experience

case series

multiple case reports

usually no statistical tests used

case series: pros/cons

pros:
- available clinical data
- rapid hypothesis generation
- natural history of presentation

cons:
- cases may not be representative of all cases
- no comparison group
- case identification can be slow

cross-sectional studies

examine a sample of population of interest at given point in time
- usually the present

outcome and exposure of interest determined at the same time
- usually single data point

can provide estimates of prevalence or exposure of a condition

cross-sectional studies: pros/cons

pros:
- rapid analysis if data gathering proceeds quickly
- accurate estimates of population prevalence if sample appropriate
- can generate strong hypotheses

cons:
- unclear temporal direction (causality)
- cannot determine whether exposure occurred before outcome

cohort studies: overview

participants classified as "exposed/unexposed" to a given risk and followed to determine whether exposure status predicts outcome

can be prospective or retrospective

design parallels RCTs but without random assignment to exposure

useful when unethical or impossible to conduct intervention

cohort studies vs case control

cohort methodology stronger than case control
- more rigorous observational design
- exposure known to occur before outcome

role of cohort studies

detection associations between health outcomes and exposures

evaluation of interventions in usual practice

estimate risk or rate of some outcome

cohort studies: pros/con

pros:
- clear temporal relationship
- least susceptible to some forms of bias
- allow direct measurement of incidence in exposed and unexposed

cons:
- no control over exposure
- inefficient for rare outcomes
- loss to follow-up threatens validity
- more expensive than other observational designs

cohort studies: variations

retrospective cohort studies
- ascertain historical control
- ascertain historical outcomes status

pros:
- data already collected, no wait

cons:
- not feasible without accurate historical records

case-control studies

participants identified based on outcome

prior exposure determined after outcome status

generate odds ratios (ORs)

main comparison = rates of exposure in case vs control

case control studies: pros/cons

pros:
- fewer resources
- for rare/multiple outcomes
- for outcomes with long latency

cons:
- inefficient for rare exposures
- hard to identify controls
- can be hard to obtain exposure history
- prone to bias

sources of bias in observational research

selection bias

information bias

confounding

selection bias

selection of participants or data such that the sample is not objectively represented in one or more groups

affects internal validity while sampling bias affects external validity

types of selection bias

self-selection

referral bias = referred based on exposure so overstates association between exposure and positive test/diagnosis

controlling for selection bias

-random sampling

- systematically recruiting to prevent self-selection

- increase frequency of contact/communication to minimize loss

- well-defined patient accrual/enrollment procedures

- use propensity scores to match exposed/unexposed groups

information bias

errors in exposure or outcome info collected about study subjects

misclassifying:
- exposed as unexposed
- deceased as living

types of information bias

recall bias = cases more likely to recall exposures than controls

detection bias = exposed participants followed more closely than unexposed

controlling for information bias

- must be handled during design phase

- use blinding if appropriate

- standardize measurement process

- minimize recall bias by using other information sources

confounding bias

"third variable"

confounder:
- has effect on outcome
- associated with exposure
- cannot be an effect of exposure

controlling for confounding

identify potential confounders

design:
- random allocation
- restriction
- matching

analysis:
- stratification
- regression modeling
- propensity scores
- sensitivity analysis

common confounders

- age
- sex
- principal diagnosis and its severity
- extent and severity of comorbidities
- physical functioning status

comorbidities

most common source of difference

all other conditions that exist in a patient in addition to condition under study