Galton.uchicago.edu

Meta-Analysis
(or Overview)
Clinical Scenario
• Mrs BW 64 y.o.
• Hysterectomy
• Dx of osteoporosis (Spinal BMD < 2SD below
Dx of osteoporosis (Spinal BMD < 2SD below
• ? aminobisphosphonate (etidronate,
? aminobisphosphonate (etidronate,
alendronate)
• Pt concern: cost
Osteoporosis
• Affects 30-40% of postmenopausal women
• 1.5M fractures annually
• Vertebral fractures most common
• Nonvertebral fractures (esp. hip) major causes
Nonvertebral fractures (esp. hip) major causes
of morbidity, mortality, cost
• Risk of hip fracture (50yo white woman): 1/6
• Risk of wrist fracture (50yo white woman): 1/6
Postmenopausal
Postmenopausal
osteoporosis
• Imbalance of bone formation and resorption
• Progressive decline in bone mass, + risk of
Progressive decline in bone mass, + risk of
fracture
• Aminobisphosphonates: potent, specific
Aminobisphosphonates: potent, specific
inhibitor of osteoclast-mediated bone
inhibitor of osteoclast-mediated bone
resorption
• Eg: etidronate, alendronate
Alendronate
• + Bone mineral density
• —But is new bone weaker?
—But is new bone weaker?
• Mineralization unaffected at therapeutic does
• —But is there a clinical effect?
—But is there a clinical effect?
• - Vertebral fracture incidence
• —But most morbidity from hip, wrist
—But most morbidity from hip, wrist
• Does it reduce nonvertebral fractures?
Meta-Analysis is
• A formal method for research synthesis
• A quantitative and potentially more objective
A quantitative and potentially
more objective
alternative to the expert review paper.
Aspirin and MI
• How strong is the effect of aspirin in reducing
How strong is the effect of aspirin in reducing
post-MI mortality?
• Answer: about 10% reduction
• How do we know?
• Many studies “inconclusive”
Finding small effects
Small effects can be important
A 10% reduction is hard to find
(reducing 20% mortality to 18%)
—Without a long, huge (thus expensive) study
How large a study?
Need to have 95% CI narrower than roughly
Need to have 95% CI narrower than roughly
0.20 ± 0.02
This gives 50-50 chance of detection
Thus, need se(difference) < 0.01
SE(difference of two proportions) =
0.8 x 0.2 x
Setting this equal to 0.01 and solving for n we
Setting this equal to 0.01 and solving for n we
have n 800 in each group
Combine evidence
Combine evidence
from many studies
• Assess whether there really is any effect
• Estimate the size of the effect
Reasons for doing
Reasons for doing
meta-analysis
• Increase power by increasing sample size
Increase power by increasing sample size
(pooling)
• Resolve conflicting reports (consensus)
• Improve estimates of effect size
• Answer new questions
Meta-analysis
• Combines features of multicenter trials and
Combines features of multicenter trials and
retrospective studies
• Multicenter trials similarities—we are
Multicenter trials similarities—we are
combining studies with
• Common questions
Common questions
• Similar study designs
Similar study designs
• Simultaneous controls within each study unit
Simultaneous controls within each study unit
• Similarities to retrospective studies
• Investigators choose which studies to include and exclude
Investigators choose which studies to include and exclude
• Susceptible to selection biases
Susceptible to selection biases
• Susceptible to ascertainment biases (Are “negative”
• Susceptible to ascertainment biases (Are “negative”
studies as likely to be published?)
studies as likely to be published?)
Alendronate
Alendronate
meta-analysis
• Karpf, et al, JAMA (April 9, 1997)
• Five studies
• RCT
• Placebo controlled
• >2 yr duration
Meta-analysis Issues
• Were criteria for including and excluding
Were criteria for including and excluding
studies clearly defined in advance?
• Were criteria for evaluating the results from
Were criteria for evaluating the results from
studies clearly defined in advance?
• Does the overview use patient-level data?
➔ • How objective was the review?
➔ • Does the overview assess heterogeneity?
How objective was the
How objective was the
• Review by more than one reader
• Procedures for resolving disagreements
Procedures for resolving disagreements
between readers
• Blinded review
• Use of prepared data-extraction forms
Does the overview
Does the overview
assess heterogeneity?
• Variation in the validity of the studies being
Variation in the validity of the studies being
• Homogeneity of the studies regarding study
Homogeneity of the studies regarding study
• Homogeneity of the studies with respect to the
Homogeneity of the studies with respect to the
size of measured effects?
• Possible factors accounting for heterogeneity?
Reader's Guide:
All articles
• Are the results valid?
• What were the results?
• Will the results help me in caring for my
Will the results help me in caring for my
patients?
Are the results valid?
• Did the review address a focused clinical
Did the review address a focused clinical
question?
• Were the criteria used to select articles for
Were the criteria used to select articles for
inclusion appropriate?
• Secondary guides
• Is it unlikely that important, relevant studies were missed?
Is it unlikely that important, relevant studies were missed?
• Was the validity of the included studies appraised?
Was the validity of the included studies appraised?
• Were assessments of studies reproducible?
Were assessments of studies reproducible?
• Were the results similar from study to study?
Were the results similar from study to study?
What are the results?
• What are the overall results of the review?
• How precise were the results?
Will the results help me in
Will the results help me in
caring for my patients?
• Can the results be applied in my patient care?
• Were all clinically important outcomes
Were all clinically important outcomes
considered?
• Are the benefits worth the harms and costs?

Source: http://galton.uchicago.edu/~thisted/courses/315/lectures/2297.pdf