Critical Appraisal Flashcards
List the hierarchy of research design
RCT double blind
> Randomized Controlled Studies
> Cohort Studies
> Case Control Studies
> Case Series
> Case Reports
> Ideas, editorials, opinions
> Animal research
> In vitro ('test tube') researchRCT considered ‘best’ study design to confer causality b/c of:
Cofounders assigned equally to all groups.
How do you assess exposure status:
Case control, Cohort or RCT?
Case control
- you start w/ the outcome (dx) and assess exposure status
- less time, good for rare dx
- less$ but introduces bias
What is a strength of the cohort design? Con?
start w/ exposure (exposed vs. unexposed)
assess outcome over time
(+): Tx not withheld , compare risks/rate in both groups
(-): introduce bias, time, expense
What is the “gold standard” in research design?
RCT
- assign to groups randomly and compare rates of outcomes
Pro and Con of RCT:
(+)
- best evidence for causality
- minimize cofounder
- greater chance of blinding
(-)
- $$/ time consuming
- difficult for rare events
- may be unethical
How do you calculate relative risk?
Incidence (intervention group)/ incidence (control)
RR> 1= incidence higher in tx group
RR= no difference
RR< 1= incidence lower in tx group
Does the course help pass the exam. 2/20 fail in PR. 10/20 if cooking. Calculate relative risk?
incidence intervention= 2/10= 0.1
Incidence control= 0.5
0.1/0.5= 0.2
Relative risk reduction (RRR)
1- RR
‘attending the course decreases relative risk by failing by x.’
If RR= 0.2
1-0.2= 0.8
80%.
Course decreases relative risk of failing by 80%
AKA how many times intervention or exposure increase/decrease risk of outcome.
Number needed to treat=
NNT= inverse of ARR
ARR= Incidence in Control - Incidence in Intervention
Absolute risk reduction (Risk Difference)
Difference in incidence between tx and control
ARR= Incidence in Control - Incidence in Intervention
= Or risk of control x RRR
New drug. Ad states drug decreases failure rate by 25%. Baseline failure rate in control 40%. How many needed to tx to prevent one failure.
NNT= inverse of ARR
ARR= risk of control x RRR
ARR= 0.4 x 0.25 = 0.1 NNT= 1/0.1= 10
Explain what RR, RRR, ARR are.
RR= incidence higher or lower in intervention after tx= DIVIDE incidence (intervention/control)
RRR= doing tx reduces relative risk by %.= 1- RRR
ARR= diff in risk btwn tx and control = SUBTRACT incidence (control- intervention)
New drug. Ad states drug decreases failure rate by 25%. Baseline failure rate in control 40%. What is the 25% and what is the 40%?
40%= risk in control group.
RRR= 25%
ARR= risk control x RRR
NNT= 1/ ARR
Which is adv. of case control study over cohort?
- retrospective
- ideal for short lag btwn exposure and outcome
- accurately estimate RR
- ideal for rare dx/outcome
Ideal for rare dx/ outcomes
List two pro and con of cohort study:
Pro:
- tx not withheld
- generally cheaper than RCT
- can estimate RR (incidence)
Con:
- control hard to get
- does not deal with anticipated cofounders
- blinding more a challenge
- $$
- challenge for rare dx
T or F: Case control is the reverse of Cohort studies?
True.
- Start with group based on OUTCOME and then go back in the to look at historical rates of exposure
- can match potential cofounders
- give ODDS RATIO
List (+) and (-) of case control design?
Pro:
- cheap, quick
- good for RARE dx or long-lag from exposure to outcome
Con:
- recall bias
- choice of control can be issue
- does not deal w/ anticipated cofounders
- can’t tell prevalence
- no RR or incidence
You are studying population of 100 people of which 30 have dx. What is the incidence?
Don’t know.
But can say prevalence is 30%
Define sensitivity + specificity=
SNOUT
- SeNsivity high
- Negative test
- rule OUT dx
i.e CX for UTI
SPIN
- Specificity high
- Positive test
- rules IN dx
i.e. bx CA
** remember columns!
Write are the equations for sensitivity and specificity?
THINK COLUMN OF TABLE
Sensitivity
= True +/ (True+ and False (-))
Specificity = True (-) / True (-) + False (+)
Define positive and negative productive value.
(+) Predictive value= if test positive likely dx presents.
(-) predictive value= if test negative, likely dx absent
T or F: (+) and (-) predictive values vary with prevalence.
True.
Versus . SNOUT and SPIN do not .
What are the equations for positive and negative predictive value?
THE ROWS!
PPV= True positive/ (true positive + false positive)
Negative PV= true negative/ (true neg. + false negative)
Define Null hypothesis
no difference between groups being compared.
Define P value
= probability that difference in observed study simply due to chance (type 1 error)
Define Type 2 Errors
based on ‘true difference’ what is chance difference not statistically significant
Define Power
1- type 2 errors
= what’s the strength that I have to find differences in our study
How do you critically appraise an article?
Results Valid
- trial address focused issue
- design consistent w/ question (i.e. RCT if want cause-effect, versus rare dx case-cohort)
- i.e. blinded well versus cases recruited well
- cofounders considered
What are Results
- measure really measure what they should (i.e. imperfect proxy)
- how large is the effect
- how precise (p interval, confidence interval)
Will they help
= External validity
- help locally?
- can study be generalized?
What is an RCT
randomized control trials exposure assigned randomly to group and followed over time for outcomes
What are experimental design and what are observational:
RCT= experiment
Rest= Observation
- Cohort
- Case-control
- Cross-sectional etc.
Cross sectional study IS?
design that examine pop’n at ONE point in time.
- selected w/out exposure or dx status in mind
- measure dx prevalence
What is a systematic review? Versus Meta-Analysis
Systematic
= literature reviewed prepared using systematic approach to minimize bias and random error
- ID and critique relevant research studies
- discuss factors that may explain heterogeneity
- synthesize info
- May or may not include meta-analysis
Meta-analysis:
- statistical pooling of results of individual studies
via defined protocol
- statistic goal to produce signal estimate of tx effect
Study with minimal bias. Cofounder distributed unbiased. Can do cause-effect. =
RCT
Can study multiple outcome for one exposure. Can look at DIRECTION of cause of effect.
Cohort
Good for studying rare dx. Can examine multiple exposures. Cost effective.
Case-Control
Best for getting prevalence of disease or risk factor. Cheap and simple.
Cross-sectional survey
= relationship btwn dx (and other traits) as exist in defined pop’n at ONE point in time.
The specificity of a screening test best described as proportion of person:
- condition who test (+)
- condition who test (-)
- without condition who test (+)
- without condition who test (-)
Without condition who test negative.
**alway think it’s opposite of (+) predictive value.
On the true positive and true negative table- what row and column is what?
True positive= PPV row; Sensitivity Column
True negative= (-) PV row; Specificity Column
100 patients screened for colon CA. 30 (+) screen. 5 found to have colon cancer. What is positive predictive value:
True positive/
True positive + false positive
= 5/30
= 16.6%
If newborn screen TOO sensitive - unacceptable # of:
- false (-)
- false (+)
- inconclusive test
- true (+)
- true (-)
False (+) test
T or F: (+) screen is dx for condition
False.
Require dx testing.
