Questions and lessons can't be printed or saved as a PDF.
What problem across the pooled studies does this test, plot or model deal with?
More, if you want it
- Tell from by whether the method adds missing studies or counts them.
- Tell from the pooled confidence interval by whether the number tests .
- Name when the pooled analysis is re-run with different quality choices.
Which problem does the patient's pooled study evidence show when the doctor reviews it?Tap a term: its definition, what you can do with it, and how often the exam asks it.
You can: Tell I-squared from fail-safe N by whether the number measures disagreement between studies or missing studies.
Definition: a pooling method that allows true effects to vary between studies.
You can: Tell a random effects model from a fixed effects model by whether true effects may vary between studies.
Definition: a plot of z score against precision that shows heterogeneity and outliers.
You can: Tell a Galbraith plot from a funnel plot by whether the plot shows outliers or missing studies.
Definition: a plot of each study result and the pooled estimate.
You can: Tell a forest plot from a Galbraith plot by whether the plot lists each study and the pooled result.
Definition: a plot of study size against effect that shows publication bias as asymmetry.
You can: Tell a Galbraith plot from a funnel plot by whether the plot shows outliers or missing studies.
Definition: a method that adds the missing studies and corrects the effect size.
You can: Tell trim and fill from fail-safe N by whether the method adds missing studies or counts them.
Definition: the number of unpublished null studies needed to cancel the pooled result.
You can: Tell fail-safe N from the pooled confidence interval by whether the number tests publication bias.
Definition: a repeat of the meta-analysis with different quality choices.
You can: Name sensitivity analysis when the pooled analysis is re-run with different quality choices.
Each card is one term. The grey line is what the patient says, which is what the question stem gives.
Summary
- I-squared: the percentage of variation between studies due to real differences rather than chance.
The I-squared statistic gives the percentage of variation across studies due to heterogeneity rather than chance.
The I-squared statistic gives the percentage of variation across studies due to heterogeneity rather than chance. Heterogeneity means the studies give real differences in results. An I-squared of about 50 to 90% is substantial heterogeneity. A random effects model is a pooling method that allows the true effect to vary between studies.
A high I-squared means the studies disagree beyond chance. An adapted chi-squared test can also test heterogeneity. A significant chi-squared result means the studies differ more than chance allows. Small design differences can make studies give widely different results. The heterogeneity test comes before the model choice. With uncertain heterogeneity, run both fixed and random effects analyses and compare them. A good fixed result with a poor random result means the finding is only marginally significant.
Kwame is a research fellow pooling antidepressant trials at a Cochrane review group. Kwame says, "The I-squared is 66%, so these trials really disagree."
A assumes one common true effect, so it suits only studies that agree. An I-squared of 66% is substantial heterogeneity. A random effects model suits the analysis when heterogeneity is present.
5 more sections on Heterogeneity and publication bias in meta-analysis
Log in free to read the rest: the full mental model, exam recognition, and the look-alikes that trip people up. Your daily session is free, always.
mrcpsych.co.uk · An independent study tool · not affiliated with the Royal College of Psychiatrists