clinical trials Flashcards
case study or case series
no randomization
no comparison with untreated group
may provide info about side effects
studies with a comparison (4 types)
- historical controls
- simultaneous non-randomized controls
- randomization
- stratified randomization
historical controls
data used from comparison group from past
disadvantage: can be diff in base pop., disease rates, disease def, disease treamtnet, quality of data collected
simultaneous non-randomized controls
can be a problem if assignment system can be predicted
EX: patients on even day are treament group, odd are controls
may lead to more admittance on even days
intervention studies
test efficacy of preventive or therapeutic exposures
1. controlled clinical trials
2. community interventions
controlled clinical trial characterisitcs
outcomes in treated compared with outcomes in control
both enrolled, treated, followed over same time period
randomized controlled clinical trials characteristics
rigorous design, greatest control
randomized to intervention or control
double blind
start with
clinical trials
study pop
then randomize
then see if they improve or don’t improve
outcomes
clinical trials
endpoints
relative risk
hazard ratio
odds ratio from logistic regression
clinical trials pros
clinical trials
no selection bias
certain exposure, must monitor
compliance
gold standard
clinical trial cons
clinical trials
expensive
long
not generalizable
ethical issues
selection of subjects
clinical trials
clear critera written in advance
can impact generalizability
what does randomization help with?
clinical trials
decreases bias
establishes balance at baseline of confounders
randomization issues
clinical trials
intervention group could have more loss to follow up (ITT)
if 1 or more arm is noncompliant, it will underestimate effectiveness
goal of randomization
clinical trials
increase probability that there is an even distribution of observable and unobservable potential confounders between intervention and control
single blind
subject unaware
double blind
subject and experimenter doesn’t know group assignment
why use a placebo
placebo effect exists: rates vary 0 to 35%
helps control unplanned crossover
even with placebo, many assume they are on the treatment
non-compliance
patients agree to randomization, but don’t comply with treatment
intentio to treat (ITT)
clinical trials
good bc intervention effect is diluted by crossover
if we don’t use ITT we violate randomization and introduce confounding
variables of interest
clinical trials
cumulative incidence
to measure degree of assoication
clinical trials
risk ratio
rate ratio
special RCT
- crossover
- factorial design
crossover RCT
patients may request crossover
crossover may be planned so subjects can be their own control
factorial RCT
test 2 drugs or 2 nutrients
we can use same pop if:
1. outcomes are diff
2. mode of action are independent
FDA required clinical trials for new meds
phase 1: clinical pharmacology for toxicolog and efficacy
phase 2: clinical investigations for efficacy and safety
phase 3: large scale RCT for effectiveness and safety
phase 4: post marketing surveillance
clinical trials pros
clinical trials
best control over:
amount of exposure
timing and frequency of exposure
period of observation
randomization reduces likelihood that groups differ significantly
clinical trials cons
clinical trials
artificial setting
limited scope of potential impact
adherence is difficult to enforce
ethical dilemmas
community trials
determine benefit of new policy and programs
community is a defined unit
community trials design
community randomized overtime
outcomes measured at baseline, then at follow-up
community trials pros
only way to directly estimate change in behavior or modifiable exposure on incidence of disease
community trials cons
inferior to clinical trials (worse control, monitoring, intervention delivery)
less options to randomize (decreased comparability)
affected by population dynamics, secular trends, nonintervention influences
type I error
probability = alpha
we conclude treatments are different, but they really aren’t
type II error
probability = beta
we conclude treatments aren’t different, but they are
number needed to treat (NNT)
NNT = 1 / (mortality rate in untreated - mortality rate in treated)