Banishing bias Flashcards
what is the heirarchy of evidence
top - least bias
meta-analyses, systematic review
RCT
cohort study
case-control
cross-sectional study
animal trials
case reports
bottom - most bias
why is RCT ranked so highly on hierarchy of evidence
RCT effectively eliminate many sources of bias
protects against confounding by randomisation
by using baseline data and then measuring change after intervention
RCT test temporal precedence of exposure to outcome
what is RCT study design
participants randomly allocated between intervention + control group
randomisation involves clinical equipose, so ethical
clinical equipose = uncertainty in medical community whether one treatment more effective than other (so not knowingly assigning someone a treatment less effective than the other)
what type of study is RCT
analytic - testing hypothesis, to identify/quantify risk or exposure
experimental - as manipulating circumstances
experimental studies offer most protection against confounding
what is internal + external validity in RCT
internal = exposure causes outcome in study
external = whether study’s findings can be generalised to whole population at different times/situations
what is bias + does sample size/statistical significance affect bias
bias = tendency of an estimate to deviate in 1 direction from true value
NO - bias is independent of sample size or statistical significance
what error occurs in RCT
systematic error = data collected consistently changes in 1 direction from true value, so effect is not reduced when observations are averaged
how is bias reduced
masking/blinding allocations to intervention
how does intention-to-treat + per-protocol analyses differ
intention to treat = analyse everyone’s results regardless of whether they adhered to treatment
(treats more conservatively, better representation of reality)
per-protocol = only analyse those who adhered to treatment
(better test of intervention’s actual effectiveness)
what biases occur with internal validity
internal validity = exposure causes outcome in study
selection, performance, attrition, observer/information bias
selection bias - whether sample representative of wider population, systematic differences between baseline of both groups
performance bias - systematic differences between groups in care or exposure to other factors
attrition bias - systematic differences in number of pts dropping out
observer/detection/information bias - systematic differences in how outcome is measured, outcome measure doesn’t accurately measure outcome of interest
what are issues with external validity
external validity = whether study’s findings can be generalised to whole population at different times/situations
intervention effects are not generalisable
study population are unrepresentative
what are strengths + weakness of RCT
strength:
establishes safety + efficacy of intervention
best single-study evidence for causal association (as temporal precedence and identifies necessary dose/delivery of intervention)
weakness:
time-consuming, expensive
internal validity issues - selection, performance, observer, attrition biases
external validity issues - sample not generalisable, intervention effects not generalisable
