Upload
michael-stokes
View
223
Download
0
Tags:
Embed Size (px)
Citation preview
Understanding heterogeneity Understanding heterogeneity in systematic reviews and in systematic reviews and
met-analysismet-analysis• meta-analysis generates a single meta-analysis generates a single best estimate of effectbest estimate of effect– what are the underlying assumptions?what are the underlying assumptions?
• how to judge consistency of how to judge consistency of resultsresults– 4 strategies4 strategies
• what to do if inconsistencywhat to do if inconsistency
The right questionThe right question• all cancer therapy for all cancersall cancer therapy for all cancers
• all antiplatelet agents for all all antiplatelet agents for all atheroembolic events (heart, head, leg)atheroembolic events (heart, head, leg)
• all aspirin doses for strokeall aspirin doses for stroke
• 30 to 300 mg. for stroke30 to 300 mg. for stroke
• what is guide about when right to pool?what is guide about when right to pool?
What were your What were your criteria?criteria?
• what made you decide some were OK and some what made you decide some were OK and some were not?were not?
• across range of across range of – patientspatients– interventionsinterventions– comparatorscomparators– outcomesoutcomes
effect more or less sameeffect more or less same
• if notif not– big effect in severe patients, no effect in mildbig effect in severe patients, no effect in mild– big effect in high dose, no effect in lowbig effect in high dose, no effect in low– big effect in short term, none in long termbig effect in short term, none in long term
Are you happy Are you happy
pooling?pooling?
10.5
Relative Risk (95% CI)
0.73 (0.49, 1.07)
0.74 (0.59, 0.94)
0.76 (0.51, 1.12)
0.71 (0.56, 0.90)
0.73 (0.61, 0.88)
Are you happy pooling?Are you happy pooling?
10.5
Relative Risk (95% CI)
0.44 (0.30, 0.65)
0.45 (0.36, 0.60)
1.25 (0.84, 1.84)
1.17 (0.92, 1.49)
0.73 (0.61, 0.88)
What criteria were you What criteria were you using?using?
• similarity of point estimatessimilarity of point estimates– less similar, less happyless similar, less happy
• overlap of confidence overlap of confidence intervalsintervals– less overlap, less happyless overlap, less happy
-40 -24 -8 8 24 40 56
RRR (95% CI)
HomogenousHomogenous
10.5
Relative Risk (95% CI)
0.73 (0.49, 1.07)
0.74 (0.59, 0.94)
0.76 (0.51, 1.12)
0.71 (0.56, 0.90)
0.73 (0.61, 0.88)
test for heterogeneity what is the p-value?
what is the null hypothesis for the test for heterogeneity?
Ho: RR1 = RR2 = RR3 = RR4
p=0.99 for heterogeneity
HeterogeneousHeterogeneous
10.5
Relative Risk (95% CI)
0.44 (0.30, 0.65)
0.45 (0.36, 0.60)
1.25 (0.84, 1.84)
1.17 (0.92, 1.49)
0.73 (0.61, 0.88)
p-value for heterogeneity < 0.001
test for heterogeneity what is the p-value?
HomogenousHomogenous
10.5
Relative Risk (95% CI)
0.73 (0.49, 1.07)
0.74 (0.59, 0.94)
0.76 (0.51, 1.12)
0.71 (0.56, 0.90)
0.73 (0.61, 0.88)
p=0.99 for heterogeneity
I2=0%
What is the I2 ?
HeterogeneousHeterogeneous
10.5
Relative Risk (95% CI)
0.44 (0.30, 0.65)
0.45 (0.36, 0.60)
1.25 (0.84, 1.84)
1.17 (0.92, 1.49)
0.73 (0.61, 0.88)
p-value for heterogeneity < 0.001I2=89%
What is the I2 ?
II2 2 InterpretationInterpretation
No worries 0%
25%Only a little
concerned
50%Getting concerned
75%Very
concerned
100%Why are
we pooling?
HeterogeneousHeterogeneous
10.5
Relative Risk (95% CI)
0.44 (0.30, 0.65)
0.45 (0.36, 0.60)
1.25 (0.84, 1.84)
1.17 (0.92, 1.49)
0.73 (0.61, 0.88)
p-value for heterogeneity < 0.001I2=89%
If this result, what next?
Does Vitamin D prevent Does Vitamin D prevent non-vertebral fractures? non-vertebral fractures? • systematic review and meta-analysis
• patients: over 60
• intervention: Vitamin D (cholecalciferol or ergocalciferol)
• outcome: non-vertebral fractures– follow-up at least a year
• methods: blinded RCTS
Relative Risk with 95% CI for Vitamin D Non-vertebral Fractures
Chapuy et al, (2002) 0.85 (0.64, 1.13)Chapuy et al, (2002) 0.85 (0.64, 1.13)
Pooled Random Effect Model 0.82 (0.69 to 0.98)
p= 0.05 for heterogeneity, I2=53%
Chapuy et al, (1994) 0.79 (0.69, 0.92)Chapuy et al, (1994) 0.79 (0.69, 0.92)
Lips et al, (1996) 1.10 (0.87, 1.39)Lips et al, (1996) 1.10 (0.87, 1.39)
Dawson-Hughes et al, (1997) 0.46 (0.24, 0.88)Dawson-Hughes et al, (1997) 0.46 (0.24, 0.88)
Pfeifer et al, (2000) 0.48 (0.13, 1.78)Pfeifer et al, (2000) 0.48 (0.13, 1.78)
Meyer et al, (2002) 0.92 (0.68, 1.24)Meyer et al, (2002) 0.92 (0.68, 1.24)
Trivedi et al, (2003) 0.67 (0.46, 0.99)Trivedi et al, (2003) 0.67 (0.46, 0.99)
Favours Vitamin D Favours Control
Relative Risk 95% CI
Relative Risk with 95% CI for Vitamin D (Non-Vertebral Fractures, Dose >400)
Chapuy et al, (1994) 0.70 (0.69, 0.92)
Dawson-Hughes et al, (1997) 0.46 (0.24, 0.88)
Pfeifer et al, (2000) .48 (0.13, 1.78)
Chapuy et al, (2002) 0.85 (0.64, 1.13)
Trivedi et al, (2003) 0.67 (0.46, 0.99)
Pooled Random Effect Model 0.75 (0.63 to 0.89)
p= 0.26 for heterogeneity, I2=24%
Favours Vitamin D Favours Control
Relative Risk 95% CI
Relative Risk with 95% CI for Vitamin D (Non-Vertebral Fractures, Dose = 400)
Lips et al (1996) 1.10 (0.87, 1.39)
Meyer et al (2002) 0.92 (0.68, 1.24)
Pooled Random Effect Mode 1.03 (0.86 to 1.24)p = 0.35 heterogeneity, I2=0%
Favours Vitamin D Favours Control
Relative Risk 95% CI
Heterogeneity• look for explanation
• patients • interventions
• outcomes
• risk of bias
• No good explanation? What to do?
• decrease confidence in effect estimates
Beta-blockers in non-Beta-blockers in non-cardiac surgery: Strokecardiac surgery: Stroke
1 5 10 50 1000.5 0.1
Study Year Overall Event Rate Relative Risk (95% CI)
Wallace 1998 5 / 200 3.06 (0.49 to 19.02)
Pobble 2005 1 / 103 2.63 (0.11 to 62.97)
DIPOM 2006 2 / 921 4.97 (0.24 to 103.19)
MaVS 2006 6 / 496 1.83 (0.39 to 8.50)
Zaugg 2007 1 / 119 2.97 (0.12 to 72.19)
POISE 2007 60 / 8351 2.13 (1.25 to 3.64)
Random Effects Estimate 2.22 (1.39 to 3.56)
p=0.99 for heterogeneity, I²=0%
p=0.99 for heterogeneity
I2= 0%
Vitamin D + Calcium vs CalciumVitamin D + Calcium vs Calcium
10.5 0.1
Study Year Treatment Control Relative Risk (95% CI)
Komulainen 1998 8/116 14/116 0.57 (0.25, 1.31)
Pfeifer 2000 3/70 6/67 0.48 (0.12, 1.84)
Flicker 2005 25/313 35/312 0.71 (0.44, 1.16)
Grant 2005 179/1306 185/1311 0.97 (0.80, 1.18)
Random Effects 0.86 (0.67, 1.09)
Vitamin D + Calcium vs CalciumVitamin D + Calcium vs Calcium
10.5 0.1
Study Year Treatment Control Relative Risk (95% CI)
Komulainen 1998 8/116 14/116 0.57 (0.25, 1.31)
Pfeifer 2000 3/70 6/67 0.48 (0.12, 1.84)
Flicker 2005 25/313 35/312 0.71 (0.44, 1.16)
Grant 2005 179/1306 185/1311 0.97 (0.80, 1.18)
Random Effects 0.86 (0.67, 1.09)
Vitamin D + Calcium vs CalciumVitamin D + Calcium vs Calcium
10.5 0.1
Study Year Treatment Control Relative Risk (95% CI)
Komulainen 1998 8/116 14/116 0.57 (0.25, 1.31)
Pfeifer 2000 3/70 6/67 0.48 (0.12, 1.84)
Flicker 2005 25/313 35/312 0.71 (0.44, 1.16)
Grant 2005 179/1306 185/1311 0.97 (0.80, 1.18)
Random Effects 0.86 (0.67, 1.09)
p= 0.32 for heterogeneity
Vitamin D + Calcium vs CalciumVitamin D + Calcium vs Calcium
10.5 0.1
Study Year Treatment Control Relative Risk (95% CI)
Komulainen 1998 8/116 14/116 0.57 (0.25, 1.31)
Pfeifer 2000 3/70 6/67 0.48 (0.12, 1.84)
Flicker 2005 25/313 35/312 0.71 (0.44, 1.16)
Grant 2005 179/1306 185/1311 0.97 (0.80, 1.18)
Random Effects 0.86 (0.67, 1.09)
p= 0.32 for heterogeneity
I2= 14%
Vitamin D + Calcium vs Vitamin D + Calcium vs Placebo/ControlPlacebo/Control
10.5 0.1
Study Year Treatment Control Relative Risk (95% CI)
Chapuy 1994 255/1176 308/1127 0.79 (0.69, 0.92)
Dawson-Hughes 1997 11/187 26/202 0.46 (0.23, 0.90)
Chapuy 2002 97/393 55/190 0.85 (0.64, 1.13)
Harwood 2004 3/39 5/37 0.57 (0.15, 2.22)
Porthouse A 2005 34/714 69/1391 0.96 (0.64, 1.43)
Porthouse B 2005 24/607 22/602 1.08 (0.61, 1.91)
Jackson 2006 2102/18176 2158/18106 0.97 (0.92, 1.03)
Random Effects 0.87 (0.76, 1.004)
Vitamin D + Calcium vs Vitamin D + Calcium vs Placebo/ControlPlacebo/Control
10.5 0.1
Study Year Treatment Control Relative Risk (95% CI)
Chapuy 1994 255/1176 308/1127 0.79 (0.69, 0.92)
Dawson-Hughes 1997 11/187 26/202 0.46 (0.23, 0.90)
Chapuy 2002 97/393 55/190 0.85 (0.64, 1.13)
Harwood 2004 3/39 5/37 0.57 (0.15, 2.22)
Porthouse A 2005 34/714 69/1391 0.96 (0.64, 1.43)
Porthouse B 2005 24/607 22/602 1.08 (0.61, 1.91)
Jackson 2006 2102/18176 2158/18106 0.97 (0.92, 1.03)
Random Effects 0.87 (0.76, 1.004)
p= 0.06 for heterogeneity
Vitamin D + Calcium vs Vitamin D + Calcium vs Placebo/ControlPlacebo/Control
10.5 0.1
Study Year Treatment Control Relative Risk (95% CI)
Chapuy 1994 255/1176 308/1127 0.79 (0.69, 0.92)
Dawson-Hughes 1997 11/187 26/202 0.46 (0.23, 0.90)
Chapuy 2002 97/393 55/190 0.85 (0.64, 1.13)
Harwood 2004 3/39 5/37 0.57 (0.15, 2.22)
Porthouse A 2005 34/714 69/1391 0.96 (0.64, 1.43)
Porthouse B 2005 24/607 22/602 1.08 (0.61, 1.91)
Jackson 2006 2102/18176 2158/18106 0.97 (0.92, 1.03)
Random Effects 0.87 (0.76, 1.004)
p= 0.06 for heterogeneity
I2= 50%
SummarySummary• starting assumption of pooled estimate
– across pts, intervention, outcome, similar effect
• broad criteria for meta-analysis desirable– maximize precision– maximize generalizability– can check out assumption
• is there excessive heterogeneity?– point estimates too variable– confidence intervals non-overlapping– low heterogeneity p-value– high I2
• if so, look for explanation– patients, intervention, outcome, methodology