Appraisal of other study types Flashcards

1
Q

What questions do you need to think about for appraisal of a meta-analysis, in the selection bias bit?

A

Database used
Data abstraction (min 2 reviewers for inclusion criteria)
Publication bias

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

Why is publication bias relevant

A

because studies are more likely to be published if large sample size and strong positive findings > may lead to overestimation of effect

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

What graphic will show publication bias

A

A FUNNEL PLOT

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

What is a funnel plot?

A

A scatter plot of the intervention effect estimates of each study against that study’s size (n) and precision (CI)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

What bit of data abstraction is important?

A

That each study included is reviewed by TWO reviewers at least
They should each enter it into data abstraction form
Any discrepancies between forms should be assessed by a third person

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

What concepts matter when pooling results?

A

HOMOGENEITY vs HETEROGENEITY

conceptual homogeneity > statistical homogeneity (results of different studies should not be combined if not conceptually similar)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

What two statistical models can be used to combine results in meta analayses

A

Random effects model VS fixed effects model

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

Explain the random effects model

A

Assumes HETEROGENEITY between studiies (i.e. that they may have different populations with different treatment responsess)

Is MORE CONSERVATIVE, thereby less likelly to show significant results

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

Explain the fixed effects model

A

Assumes HOMOGENEITY between studies! I.e. that all studies have the same population and the same response to treatment, and that any difference is simply due to random error

MORE LIKELY TO SHOW STAT SIG RESULTS

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

Explain a forest plot

A

Used in meta analyses

Summarises heterogeneity and pooled results of the studies

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

What do box and diamond represent in a forest plot?

A
BOX= estimate for a single study (horiz line = confidence intervall) 
DIAMOND = pooled estimate (width of diamond = confidence interval)
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

When is it good to use a cohort study over an RCT?

A

when randomisation of exposure is NOT possibe > either RCT sample will be too large or unethical

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

When is it good to use a case cotrol study?

A
If follow up time period is very long 
Rare outcome (so need for huge sample size otherwise)
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

What are cross-sectional studies useful for?

A

Quick and inexpensive > useful in generating and exploring hypothesis that will be subsequently investigated using other study types

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

What is surveillance bias, and what kind of study would it occur in?

A

RCT / prospective cohort study

= going to look for disiease that may have gone unnoticed as you think there is a risk

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

Which two extra biases should you consider for a retrospective study?

A

Recall bias

Interviewer bias

17
Q

Explain the statistics you should include in a meta analysis

A

Estimate the summary effect size in the form of an ODDS RATIO
Use both FIXED and RANDOM EFFECTS MODEL
Construct a FOREST PLOT

+ include a funnel plot for publlication bias

18
Q

How can you assess the internal validity of each study included in a meta analysis

A

Using the GRADE criteriia - look at whether the observed effect is liklely similar to the true effect