meta-analysis: sports medicine canadian academy of sport medicine lacadémie canadienne de médecine...

60
Meta-Analysis: Sports Medicine Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine Canadian Academy of Sport Medicine L’Académie Canadienne de Médecine du Sport L’Académie Canadienne de Médecine du Sport Ian Shrier Ian Shrier MD, PhD, Dip Sport Med, FACSM MD, PhD, Dip Sport Med, FACSM Centre for Clinical Epidemiology and Centre for Clinical Epidemiology and Community Studies, SMBD-Jewish General Community Studies, SMBD-Jewish General Hospital and McGill University Hospital and McGill University Past-president, Canadian Academy of Sport Past-president, Canadian Academy of Sport Medicine Medicine

Upload: brandon-cantrell

Post on 27-Mar-2015

221 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

Meta-Analysis: Sports MedicineMeta-Analysis: Sports Medicine

Canadian Academy of Sport MedicineCanadian Academy of Sport MedicineL’Académie Canadienne de Médecine du SportL’Académie Canadienne de Médecine du Sport

Ian ShrierIan Shrier MD, PhD, Dip Sport Med, FACSMMD, PhD, Dip Sport Med, FACSM

Centre for Clinical Epidemiology and Community Studies, SMBD-Centre for Clinical Epidemiology and Community Studies, SMBD-Jewish General Hospital and McGill UniversityJewish General Hospital and McGill UniversityPast-president, Canadian Academy of Sport MedicinePast-president, Canadian Academy of Sport Medicine

Page 2: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

Meta-Analysis: Sports MedicineMeta-Analysis: Sports Medicine

Canadian Academy of Sport MedicineCanadian Academy of Sport MedicineL’Académie Canadienne de Médecine du SportL’Académie Canadienne de Médecine du Sport

Ian ShrierIan Shrier MD, PhD, Dip Sport Med, FACSMMD, PhD, Dip Sport Med, FACSM

Centre for Clinical Epidemiology and Community Studies, SMBD-Centre for Clinical Epidemiology and Community Studies, SMBD-Jewish General Hospital and McGill UniversityJewish General Hospital and McGill UniversityPast-president, Canadian Academy of Sport MedicinePast-president, Canadian Academy of Sport Medicine

Page 3: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

OBJECTIVESOBJECTIVES

• How do we think?How do we think?

• A Workshop Example: Does Stretching A Workshop Example: Does Stretching Prevent Injury?Prevent Injury?

• What parameter are you estimating?What parameter are you estimating?

• RCT vs Obs studies in meta-analysesRCT vs Obs studies in meta-analyses

• It’s all about Bias!It’s all about Bias!

Page 4: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

INTERPRETATIONSINTERPRETATIONS

““It’s a rather interesting It’s a rather interesting phenomenon. Every time I phenomenon. Every time I

press this lever, the press this lever, the graduate student breathes graduate student breathes

a sigh of relief”a sigh of relief”

Page 5: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

INTERPRETATIONSINTERPRETATIONS

• Shrier, Platt, Steele. Mega-trials vs. meta-analysis: Precision vs. heterogeneity? Contemp Clin Trials 2007

• Shrier et al. Should Meta-Analyses of Interventions Include Observational Studies in Addition to Randomized Controlled Trials? A Critical Examination of Underlying Principles. Am J Epi 2007

• Shrier et al. The interpretation of systematic reviews with meta-analyses: an objective or subjective process? BMC Med Inform Dec Making 2008

Page 6: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

INTERPRETATIONSINTERPRETATIONS

I believe magnesium has now been shown to be beneficial for I believe magnesium has now been shown to be beneficial for patients during the post-MI periodpatients during the post-MI period (SD-SA) (SD-SA)

Rev 8Rev 8

Rev 7Rev 7

Rev 6Rev 6

Rev 5Rev 5

Rev 4Rev 4

Rev 3Rev 3

Rev 2Rev 2

Ag

DA

Ag

Ag

DA

DA

SA

Ag

DA

-

Ag

Ag

DA

-

SA

Ag

SD

-

Ag

Ag

DA

-

Ag

AgRev 1Rev 1

II22

Rand. ORRand. OR

Fixed ORFixed OR

Ag

DA

Ag

Ag

SD

SD

Ag

59%59%

0.75 (0.61-0.92)0.75 (0.61-0.92)

1.01 (0.96-1.07)1.01 (0.96-1.07)

69,50569,505

61%61%

0.65 (0.48-0.87)0.65 (0.48-0.87)

1.02 (0.96-1.08)1.02 (0.96-1.08)

63,04763,047

-

DA

Ag

Ag

Ag

-

SA

Ag

14%14%

0.66 (0.53-0.81)0.66 (0.53-0.81)

0.64 (0.52-0.79)0.64 (0.52-0.79)

3,6853,685

0%0%

0.38 (0.21-0.66)0.38 (0.21-0.66)

0.40 (0.28-0.61)0.40 (0.28-0.61)

597597

0%0%

0.40 (0.18-0.86)0.40 (0.18-0.86)

0.40 (0.19-0.83)0.40 (0.19-0.83)

415415NN

1-231-231-201-201-101-101-51-51-31-3# RCTs# RCTs

Page 7: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

HOW DO WE THINK?

* * * * *

Clue: wantClue: want crave, covet, yearn, fancycrave, covet, yearn, fancy

(Vandenbroucke et al, 2001)(Vandenbroucke et al, 2001)

Page 8: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

HOW DO WE THINK?

J U G

U

I

C * * * *

E

Clue: wantClue: want crave, covet, yearn, fancycrave, covet, yearn, fancy

(Vandenbroucke et al, 2001)(Vandenbroucke et al, 2001)

Page 9: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

HOW DO WE THINK?

J U G

U

I

C * * * *

E H I S

QQuueeuueess

Clue: wantClue: want crave, covet, yearn, fancycrave, covet, yearn, fancy

(Vandenbroucke et al, 2001)(Vandenbroucke et al, 2001)

Page 10: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

HOW DO WE THINK?

J U G L

U I

I N

C * * * E

E H I S

QQuueeuueess

Clue: wantClue: want crave, covet, yearn, fancycrave, covet, yearn, fancy

(Vandenbroucke et al, 2001)(Vandenbroucke et al, 2001)

Page 11: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

HOW DO WE THINK?

J U G L

U I

I N

C R A V E

E H I S

L

E

QQuueeuueess

Clue: wantClue: want crave, covet, yearn, fancycrave, covet, yearn, fancy

(Vandenbroucke et al, 2001)(Vandenbroucke et al, 2001)

Page 12: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

OBJECTIVESOBJECTIVES

• How do we think?How do we think?

• A Workshop Example: Does Stretching A Workshop Example: Does Stretching Prevent Injury?Prevent Injury?

• What parameter are you estimating?What parameter are you estimating?

• RCT vs Obs studies in meta-analysesRCT vs Obs studies in meta-analyses

• It’s all about Bias!It’s all about Bias!

Page 13: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

Does Stretching Prevent Injury?

(adapted from Shrier, Evidence-Based Sports Medicine 2007)(adapted from Shrier, Evidence-Based Sports Medicine 2007)

Page 14: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

AnalysisAnalysisAnalysisAnalysis

YesYes No No

I will now tell my patients to I will now tell my patients to

stretch to prevent injurystretch to prevent injury

YesYes No No

I will now tell my patients not to I will now tell my patients not to

stretch to prevent injurystretch to prevent injury

YesYes No No

I will now tell my patients that I have no idea I will now tell my patients that I have no idea

whether they should stretch to prevent injurywhether they should stretch to prevent injury

Does Stretching Prevent Injury?

Page 15: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

Review: StretchingComparison: 01 Injuries Outcome: 04 Stretching and Injury All Studies

Study OR or RR or HR (random) OR or RR or HR (random)or sub-category log[OR or RR or HR] (SE) 95% CI 95% CI Year Quality

01 MenAmako (2003) -0.3011 (0.1976) 0.74 [0.50, 1.09] 2003 DBixler (52) Not estimable 1992 DCross (1999) -0.7134 (0.2372) 0.49 [0.31, 0.78] 1999 DEkstrand (1983a) -1.7047 (0.3536) 0.18 [0.09, 0.36] 1983 DEkstrand (1983b) Not estimable 1983 DHartig (1999) -0.5551 (0.2180) 0.57 [0.37, 0.88] 1999 DHilyer (1990) -0.1985 (0.1654) 0.82 [0.59, 1.13] 1990 DMacera (1989) 0.0953 (0.1582) 1.10 [0.81, 1.50] 1989 DMcKay (2001) -0.9637 (0.4509) 0.38 [0.16, 0.92] 2001 DPope (1998) -0.0834 (0.2855) 0.92 [0.53, 1.61] 1998 DPope (2000) 0.0392 (0.1255) 1.04 [0.81, 1.33] 2000 DWalter (1989 never) 0.1398 (0.2823) 1.15 [0.66, 2.00] 1989 DWalter (1989 some) -0.4463 (0.1796) 0.64 [0.45, 0.91] 1989 DWalter (1989 usual) 0.2231 (0.1917) 1.25 [0.86, 1.82] 1989 Dvan Mechelin (1993) 0.2311 (0.2844) 1.26 [0.72, 2.20] 1993 D

Subtotal (95% CI) 0.77 [0.61, 0.96]Test for heterogeneity: Chi? = 46.20, df = 12 (P < 0.00001), I? = 74.0%Test for overall effect: Z = 2.27 (P = 0.02)

02 WomenMacera (1989) 0.4700 (0.3994) 1.60 [0.73, 3.50] 1989 DWalter (1989 never) 0.1655 (0.5831) 1.18 [0.38, 3.70] 1989 DWalter (1989 some) -0.5798 (0.3445) 0.56 [0.29, 1.10] 1989 DWalter (1989 usual) 0.0488 (0.3536) 1.05 [0.53, 2.10] 1989 D

Subtotal (95% CI) 0.98 [0.61, 1.57]Test for heterogeneity: Chi? = 4.27, df = 3 (P = 0.23), I? = 29.8%Test for overall effect: Z = 0.09 (P = 0.93)

Total (95% CI) 0.80 [0.65, 0.98]Test for heterogeneity: Chi? = 50.92, df = 16 (P < 0.0001), I? = 68.6%Test for overall effect: Z = 2.15 (P = 0.03)

0.1 0.2 0.5 1 2 5 10

Favour Stretching Favour No Stretching

Does Stretching Prevent Injury?Does Stretching Prevent Injury?

(adapted from Shrier, Evidence-Based Sports Medicine 2007)(adapted from Shrier, Evidence-Based Sports Medicine 2007)

Page 16: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

AnalysisAnalysisAnalysisAnalysis

YesYes No No

I will now tell my patients to I will now tell my patients to

stretch to prevent injurystretch to prevent injury

YesYes No No

I will now tell my patients not to I will now tell my patients not to

stretch to prevent injurystretch to prevent injury

YesYes No No

I will now tell my patients that I have no idea I will now tell my patients that I have no idea

whether they should stretch to prevent injurywhether they should stretch to prevent injury

Does Stretching Prevent Injury?

Page 17: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

70% 80% 90% 100% 110% 120% 130%

% Unstretched Condition% Unstretched Condition

Acute Stretching: Force (MVC/1RM)Acute Stretching: Force (MVC/1RM)

(adapted from Shrier, Clin J Sport Med 2004)(adapted from Shrier, Clin J Sport Med 2004)

Page 18: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

70% 80% 90% 100% 110% 120% 130%

Hortobagyi (38): Knee extension

Handel (29): Knee

Wilson (34): Bench Press

Dintiman (35): Knee extensionSprint & Weight Training

Sprint Training

Pure Concentric Bench Press

Rebound Bench Press

Extension

Flexion

% Unstretched Condition% Unstretched Condition

Regular Stretching: Force (MVC/1RM)Regular Stretching: Force (MVC/1RM)

PNFPNF

StaticStatic

(adapted from Shrier, Clin J Sport Med 2004)(adapted from Shrier, Clin J Sport Med 2004)

Page 19: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

Acute Stretching: Force (Isokinetic)Acute Stretching: Force (Isokinetic)

80% 85% 90% 95% 100% 105% 110% 115% 120%

% Unstretched Condition% Unstretched Condition

Slow (30-60 deg/s)Slow (30-60 deg/s)

Fast (>180 deg/s)Fast (>180 deg/s)

(adapted from Shrier, Clin J Sport Med 2004)(adapted from Shrier, Clin J Sport Med 2004)

Page 20: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

Regular Stretching: Force (Isokinetic)Regular Stretching: Force (Isokinetic)

80% 85% 90% 95% 100% 105% 110% 115% 120%

Knee flexion

Extension

Knee extension

Flexion

% Unstretched Condition% Unstretched Condition

Slow (30-60 deg/s)Slow (30-60 deg/s)

Fast (>180 deg/s)Fast (>180 deg/s)

(adapted from Shrier, Clin J Sport Med 2004)(adapted from Shrier, Clin J Sport Med 2004)

Page 21: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

90% 95% 100% 105% 110%

% Unstretched Condition% Unstretched Condition

Acute Stretching: Jump HeightAcute Stretching: Jump Height

StaticStatic

CMJCMJ

(adapted from Shrier, Clin J Sport Med 2004)(adapted from Shrier, Clin J Sport Med 2004)

Page 22: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

90% 95% 100% 105% 110%

HunterStretch

Pow er & Stretch

% Unstretched Condition% Unstretched Condition

Regular Stretching: Jump HeightRegular Stretching: Jump Height

StaticStatic

CMJCMJ

(adapted from Shrier, Clin J Sport Med 2004)(adapted from Shrier, Clin J Sport Med 2004)

Page 23: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

Stretching and Force

• Overstretching occurs with as little as 20% stretchOverstretching occurs with as little as 20% stretch

• ProtocolProtocol Skinned bullfrog muscle fibers stretched and released Skinned bullfrog muscle fibers stretched and released

to different lengthsto different lengths

(adapted from Higuchi et al, J Mus Res Cell Motil 1988)(adapted from Higuchi et al, J Mus Res Cell Motil 1988)

• ResultsResults

Page 24: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

Regular Stretching: Force

• ProtocolProtocol Weights attached to left wing of Japanese Quail x 30 Weights attached to left wing of Japanese Quail x 30

daysdays Animals killed and ant. lat. dorsi. placed in vitroAnimals killed and ant. lat. dorsi. placed in vitro

(adapted from Alway, J Appl Physiol 1984)(adapted from Alway, J Appl Physiol 1984)

Con StretchMass 25.7 66.8Force 55.0 120.0Velocity 2.7 4.1

• ResultsResults

Page 25: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

Study OR or RR or HR (random) Weight OR or RR or HR (random) or sub-category 95% CI % 95% CI Year

Hilyer (68) 43.93 0.82 [0.59, 1.13] 1990 Hartig (67) 25.29 0.57 [0.37, 0.88] 1999 Am ako (6) 30.78 0.74 [0.50, 1.09] 2003

Total (95% CI) 100.00 0.73 [0.59, 0.90] Test for heterogeneity: Chi? = 1.71, df = 2 (P = 0.42), I? = 0% Test for overall effect: Z = 2.92 (P = 0.003)

0.1 0.2 0.5 1 2 5 10 Favour Stretching Favour No Stretching

Regular Stretching: Injury

(adapted from Shrier, Evidence-Based Sports Medicine 2007)(adapted from Shrier, Evidence-Based Sports Medicine 2007)

Page 26: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

Study OR or RR or HR (random) Weight OR or RR or HR (random) or sub-category 95% CI % 95% CI Year Men Ekstrand (52) 5.52 0.18 [0.09, 0.36] 1983 Ekstrand (64) Not estimable 1983 M acera (58) 8.98 1.10 [0.81, 1.50] 1989 W alter (59 never) 6.69 1.15 [0.66, 2.00] 1989 W alter (59 som e) 8.59 0.64 [0.45, 0.91] 1989 W alter (59 usual) 8.36 1.25 [0.86, 1.82] 1989 Bixler (51) Not estimable 1992 van M echelin (61) 6.65 1.26 [0.72, 2.20] 1993 Pope (63) 6.63 0.92 [0.53, 1.61] 1998 Cross (49) 7.51 0.49 [0.31, 0.78] 1999 Pope (62) 9.53 1.04 [0.81, 1.33] 2000 M cKay (7) 4.24 0.38 [0.16, 0.92] 2001 Am ako (6) 8.25 0.74 [0.50, 1.09] 2003

Subtotal (95% CI) 80.94 0.78 [0.59, 1.02] Test for heterogeneity: Chi? = 42.89, df = 10 (P < 0.00001), I? = 76.7% Test for overall effect: Z = 1.83 (P = 0.07) W om en M acera (58) 4.87 1.60 [0.73, 3.50] 1989 W alter (59 never) 3.02 1.18 [0.38, 3.70] 1989 W alter (59 som e) 5.66 0.56 [0.29, 1.10] 1989 W alter (59 usual) 5.52 1.05 [0.53, 2.10] 1989

Subtotal (95% CI) 19.06 0.98 [0.61, 1.57] Test for heterogeneity: Chi? = 4.27, df = 3 (P = 0.23), I? = 29.8% Test for overall effect: Z = 0.09 (P = 0.93)

0.1 0.2 0.5 1 2 5 10 Favour Stretching Favour No Stretching

Acute Stretching: Injury

Excluding multiple co-intervention studiesExcluding multiple co-intervention studies (adapted from Shrier, Evidence-Based Sports Medicine 2007)(adapted from Shrier, Evidence-Based Sports Medicine 2007)

Page 27: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

Study OR or RR or HR (random) Weight OR or RR or HR (random) or sub-category 95% CI % 95% CI Year Men Ekstrand (52) 1983 Ekstrand (64) Not estimable 1983 M acera (58) 8.98 1.10 [0.81, 1.50] 1989 W alter (59 never) 6.69 1.15 [0.66, 2.00] 1989 W alter (59 som e) 8.59 0.64 [0.45, 0.91] 1989 W alter (59 usual) 8.36 1.25 [0.86, 1.82] 1989 Bixler (51) Not estimable 1992 van M echelin (61) 6.65 1.26 [0.72, 2.20] 1993 Pope (63) 6.63 0.92 [0.53, 1.61] 1998 Cross (49) 1999 Pope (62) 9.53 1.04 [0.81, 1.33] 2000 M cKay (7) 4.24 0.38 [0.16, 0.92] 2001 Am ako (6) 2003

Subtotal (95% CI)

W om en M acera (58) 4.87 1.60 [0.73, 3.50] 1989 W alter (59 never) 3.02 1.18 [0.38, 3.70] 1989 W alter (59 som e) 5.66 0.56 [0.29, 1.10] 1989 W alter (59 usual) 5.52 1.05 [0.53, 2.10] 1989

Subtotal (95% CI) 19.06 0.98 [0.61, 1.57] Test for heterogeneity: Chi? = 4.27, df = 3 (P = 0.23), I? = 29.8% Test for overall effect: Z = 0.09 (P = 0.93)

0.1 0.2 0.5 1 2 5 10 Favour Stretching Favour No Stretching

29.41 0.97 [0.79, 1.19] Test for heterogeneity: Chi? = 13.27, df = 7 (P = 0.07), I? = 47.2% Test for overall effect: Z = 0.27 (P = 0.79)

Excluding multiple co-intervention studiesExcluding multiple co-intervention studies

Acute Stretching: Injury

(adapted from Shrier, Evidence-Based Sports Medicine 2007)(adapted from Shrier, Evidence-Based Sports Medicine 2007)

Page 28: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

AnalysisAnalysisAnalysisAnalysis

YesYes No No

I will now tell my patients to I will now tell my patients to stretchstretch before exercisebefore exercise to prevent injury to prevent injury

YesYes No No

I will now tell my patients not to I will now tell my patients not to stretch stretch before exercisebefore exercise to prevent injury to prevent injury

YesYes No No

I will now tell my patients that I have no idea whether I will now tell my patients that I have no idea whether they should they should stretch before exercisestretch before exercise to prevent injury to prevent injury

Does Stretching Prevent Injury?

Page 29: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

AnalysisAnalysisAnalysisAnalysis

YesYes No No

I will now tell my patients to I will now tell my patients to stretch stretch regularlyregularly to prevent injury to prevent injury

YesYes No No

I will now tell my patients not to I will now tell my patients not to stretch stretch regularlyregularly to prevent injury to prevent injury

YesYes No No

I will now tell my patients that I have no idea whether I will now tell my patients that I have no idea whether they should they should stretch regularlystretch regularly to prevent injury to prevent injury

Does Stretching Prevent Injury?

Page 30: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

OBJECTIVESOBJECTIVES

• How do we think?How do we think?

• A Workshop Example: Does Stretching A Workshop Example: Does Stretching Prevent Injury?Prevent Injury?

• What parameter are you estimating?What parameter are you estimating?

• RCT vs Obs studies in meta-analysesRCT vs Obs studies in meta-analyses

• It’s all about Bias!It’s all about Bias!

Page 31: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

Clinical Trials and Meta-Analysis 1994;29:41–47 Clinical Trials and Meta-Analysis 1994;29:41–47

Effect of RCT on Outcomes

Page 32: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

RCT vs. Observational: TheoreticalRCT vs. Observational: Theoretical• HughesHughes

Balanced placebo/traditional designBalanced placebo/traditional design RCT Informed: pts told Nicotine or PlaceboRCT Informed: pts told Nicotine or Placebo Balanced placebo: Pts randomized to be told Balanced placebo: Pts randomized to be told

Nicotine or Placebo, but random 50% given what Nicotine or Placebo, but random 50% given what they were told (4 groups)they were told (4 groups)

# D

ays

# D

ays

Sm

oke

dS

mo

ked

NicNicoo

PlaPlacc

BlinBlindd

% C

om

ple

te

% C

om

ple

te

Ab

stai

nm

ent

Ab

stai

nm

ent

NicNicoo

PlaPlacc

BlinBlindd

NicoNico

PlacPlac

Psychopharmacology 1989Psychopharmacology 1989

• ITT: Patient wants to know effect of intervention conditional on them ITT: Patient wants to know effect of intervention conditional on them receiving the intervention (per protocol?)receiving the intervention (per protocol?)

Page 33: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

OBJECTIVESOBJECTIVES

• How do we think?How do we think?

• A Workshop Example: Does Stretching A Workshop Example: Does Stretching Prevent Injury?Prevent Injury?

• What parameter are you estimating?What parameter are you estimating?

• RCT vs Obs studies in meta-analysesRCT vs Obs studies in meta-analyses

• It’s all about Bias!It’s all about Bias!

Page 34: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

RCT vs. Observational: Evidence

• Linde: Linde: Observational studies about 10-20% better for Observational studies about 10-20% better for acupuncture/headacheacupuncture/headache (J Clin Epi 2002)(J Clin Epi 2002)

• Concato: Concato: No difference in well-designed studiesNo difference in well-designed studies (NEJM 2000)(NEJM 2000)

• MacLehose: MacLehose: discrepancies for high quality studies were small but discrepancies for high quality studies were small but discrepancies for low quality studies were largediscrepancies for low quality studies were large (HTA 2000) (HTA 2000)

• Benson: Benson: No difference after 1984No difference after 1984 (Am J Opthalmol 2000)(Am J Opthalmol 2000)

• Britton: Britton: “Non-randomized overestimated magnitude of effect”“Non-randomized overestimated magnitude of effect” (HTA 1998)(HTA 1998)

• Linde: Linde: Observational studies about 10-20% better for Observational studies about 10-20% better for acupuncture/headacheacupuncture/headache (J Clin Epi 2002)(J Clin Epi 2002)

• Concato: Concato: No difference in well-designed studiesNo difference in well-designed studies (NEJM 2000)(NEJM 2000)

• MacLehose: MacLehose: discrepancies for high quality studies were small but discrepancies for high quality studies were small but discrepancies for low quality studies were largediscrepancies for low quality studies were large (HTA 2000) (HTA 2000)

• Benson: Benson: No difference after 1984No difference after 1984 (Am J Opthalmol 2000)(Am J Opthalmol 2000)

• Britton: Britton: “Non-randomized overestimated magnitude of effect”“Non-randomized overestimated magnitude of effect” (HTA 1998)(HTA 1998)

more extrememore extreme

Page 35: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

RCT vs. Observational: Theoretical

• All StudiesAll Studies Adjust for known confoundersAdjust for known confounders

• All StudiesAll Studies Adjust for known confoundersAdjust for known confounders

• Unknown confounders likely to be Unknown confounders likely to be equally distributedequally distributedRCTRCT

ConConProProDesignDesign

Concealed randomization specifically removes the possibility of selection Concealed randomization specifically removes the possibility of selection bias or confounding in RCTs, i.e. any differences between the groups are bias or confounding in RCTs, i.e. any differences between the groups are attributable to chance or to the intervention, all else being equal.attributable to chance or to the intervention, all else being equal.

Deeks et al, Health Tech Asess 2003Deeks et al, Health Tech Asess 2003

Based on assumption of randomization in infinite population, or opposite Based on assumption of randomization in infinite population, or opposite distribution of confounders if many trials examineddistribution of confounders if many trials examined

Example: Confounder present in 20% of population. N= 400Example: Confounder present in 20% of population. N= 40095% Prob. Distr. = 15.6%-24.4%. If 5 confounders, 23% chance that at least 95% Prob. Distr. = 15.6%-24.4%. If 5 confounders, 23% chance that at least one is outside the range (95% Prob. Dist. = 14.2%-25.8%)one is outside the range (95% Prob. Dist. = 14.2%-25.8%) (Shrier et al, AJE 2007)(Shrier et al, AJE 2007)

Page 36: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

RCT vs. Observational: Theoretical

• All StudiesAll Studies Adjust for known confoundersAdjust for known confounders

• All StudiesAll Studies Adjust for known confoundersAdjust for known confounders

DesignDesign ProPro ConCon

RCTRCT• Unknown confounders likely to be Unknown confounders likely to be

equally distributedequally distributed

• Control participants likely to Control participants likely to do better than non-do better than non-participantsparticipants

CohortCohort

• More representative sampleMore representative sample

• CheaperCheaper

• Historical Cohort: answer fasterHistorical Cohort: answer faster

• Increased sample size for Increased sample size for adjustmentadjustment

• Confounding by indication: Confounding by indication: patient/physician - randompatient/physician - random

There may be important prognostic factors that the There may be important prognostic factors that the investigators do not know about or have not investigators do not know about or have not measured which are unbalanced between groups and measured which are unbalanced between groups and responsible for differences in outcome.responsible for differences in outcome.

Deeks et al, Health Tech Asess 2003Deeks et al, Health Tech Asess 2003(Shrier et al, AJE 2007)(Shrier et al, AJE 2007)

Page 37: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

OBJECTIVESOBJECTIVES

• How do we think?How do we think?

• A Workshop Example: Does Stretching A Workshop Example: Does Stretching Prevent Injury?Prevent Injury?

• What parameter are you estimating?What parameter are you estimating?

• RCT vs Obs studies in meta-analysesRCT vs Obs studies in meta-analyses

• It’s all about Bias!It’s all about Bias!

Page 38: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

turn on your IPOD now!turn on your IPOD now!

IF YOU WANT THE BLUE PILL….IF YOU WANT THE BLUE PILL….

QuickTime™ and aCinepak decompressor

are needed to see this picture.

Page 39: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

FORMS OF BIAS

Structural Approach to BiasStructural Approach to Bias

• Confounding BiasConfounding Bias• Failure to condition on a common causeFailure to condition on a common cause• Do not condition on a variable (or marker of a Do not condition on a variable (or marker of a

variable) that lies along the causal pathwayvariable) that lies along the causal pathway

• Selection BiasSelection Bias• Conditioning on a common effectConditioning on a common effect

(Pearl, Hern(Pearl, Hernáán, Greenland)n, Greenland)

Page 40: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

CONFOUNDING BIAS

OsteoarthritisOsteoarthritis(indirect)(indirect)

Gait DisorderGait Disorder(direct)(direct)

ActivityActivity

XX MM YY

Gait DisorderGait Disorder

OAOA ActivityActivityXX

CC

YY

XXXX

Page 41: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

CONFOUNDING BIAS?

• Exposure: smokingExposure: smoking

• Outcome: spont. abortionOutcome: spont. abortion

• Confounding?: previous spont. abortionConfounding?: previous spont. abortion

(Weinberg Am J Epid 1993)(Weinberg Am J Epid 1993)

SmokingSmokingSmokingSmoking Spont. AbortionSpont. AbortionSpont. AbortionSpont. Abortion

Previous Sp. Ab.Previous Sp. Ab.Previous Sp. Ab.Previous Sp. Ab.

Page 42: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

CONFOUNDING BIAS?

• Exposure: smokingExposure: smoking

• Outcome: spont. abortionOutcome: spont. abortion

• Confounding?: previous spont. abortionConfounding?: previous spont. abortion

(Weinberg Am J Epid 1993)(Weinberg Am J Epid 1993)

SmokingSmokingSmokingSmoking Spont. AbortionSpont. AbortionSpont. AbortionSpont. Abortion

Previous Sp. Ab.Previous Sp. Ab.Previous Sp. Ab.Previous Sp. Ab.

TissueTissueAbnormalityAbnormality

TissueTissueAbnormalityAbnormality

• Underlying abnormality: intrinsic tissue Underlying abnormality: intrinsic tissue abnormalityabnormality

Page 43: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

CONFOUNDING BIAS?

• Exposure: smokingExposure: smoking

• Outcome: spont. abortionOutcome: spont. abortion

• Confounding?: previous spont. abortionConfounding?: previous spont. abortion

(Weinberg Am J Epid 1993)(Weinberg Am J Epid 1993)

SmokingSmokingSmokingSmoking Spont. AbortionSpont. AbortionSpont. AbortionSpont. Abortion

Previous Sp. Ab.Previous Sp. Ab.Previous Sp. Ab.Previous Sp. Ab.

TissueTissueAbnormalityAbnormality

TissueTissueAbnormalityAbnormality

• Underlying abnormality: intrinsic tissue Underlying abnormality: intrinsic tissue weaknessweakness

XX XX

Page 44: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

CONFOUNDING BIAS?

(Weinberg Am J Epid 1993)(Weinberg Am J Epid 1993)

• Univariate RR for smoking/no smokingUnivariate RR for smoking/no smoking =1.85=1.85

• Stratified RRStratified RR RR for smoking/no smoking (Previous Sp. Ab.)RR for smoking/no smoking (Previous Sp. Ab.) =1.32=1.32 RR for smoking/no smoking (No Previous Sp. Ab.) RR for smoking/no smoking (No Previous Sp. Ab.) =1.32=1.32

• However, whether or not someone had a previous spontaneous However, whether or not someone had a previous spontaneous abortion does not change the effects of smokingabortion does not change the effects of smoking

• Including this covariate results in an invalid estimateIncluding this covariate results in an invalid estimate

SmokingSmokingSmokingSmoking Spont. AbortionSpont. AbortionSpont. AbortionSpont. Abortion

Previous Sp. Ab.Previous Sp. Ab.Previous Sp. Ab.Previous Sp. Ab.

TissueTissueAbnormalityAbnormality

TissueTissueAbnormalityAbnormality

XX XX

Page 45: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

CONFOUNDING BIAS?CONFOUNDING BIAS?

(Hern(Hernánán Am J Epid Am J Epid 2002)2002)

CC ExEx OutcomeOutcome

ExEx UU

OutcomeOutcome

CC

CCCC ExExExEx OutcomeOutcomeOutcomeOutcome

UUUU

CCCC ExExExEx OutcomeOutcomeOutcomeOutcome

UUUU

Page 46: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

Condition on C?

(Cole & Hern(Cole & Hernánán Int J Epid 2002) Int J Epid 2002)

EE OutcomeOutcome

CC

U2U2U1U1

Page 47: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

FORMS OF BIAS

Structural Approach to BiasStructural Approach to Bias

• Confounding BiasConfounding Bias• Failure to condition on a common causeFailure to condition on a common cause• Do not condition on a variable (or marker of a Do not condition on a variable (or marker of a

variable) that lies along the causal pathwayvariable) that lies along the causal pathway

• Selection BiasSelection Bias• Conditioning on a common effectConditioning on a common effect

(Pearl, Hern(Pearl, Hernáán, Greenland)n, Greenland)

Page 48: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

(Pearl. Causality Book)(Pearl. Causality Book)

Step 4: Connect any two parents sharing a common child.Step 4: Connect any two parents sharing a common child. Including “colliders” opens up path for confoundingIncluding “colliders” opens up path for confounding

XX11

XX33 XX22

XX44

XX55

SprinklerSprinkler RainRain

SeasonSeason

WetWet

SlipperySlippery

If one knows the value If one knows the value of the “collider”, the of the “collider”, the parents are associated. parents are associated.

If wet:If wet: the sprinkler is the sprinkler is more likely to be more likely to be on if there was no on if there was no rain.rain.

PEARL’S RULES - EXPLANATION

Page 49: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

UNBIASED EFFECT ESTIMATE?UNBIASED EFFECT ESTIMATE?

XX OutcomeOutcome

(Pearl. Causality Book)(Pearl. Causality Book)

Which measurements should be included in the model if we are interested in Which measurements should be included in the model if we are interested in the relation between X and Outcome? the relation between X and Outcome?

Page 50: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

XX OutcomeOutcome

ZZ11

(Pearl. Causality Book)(Pearl. Causality Book)

ZZ22

Which measurements should be included in the model if we are interested in Which measurements should be included in the model if we are interested in the relation between X and Outcome? the relation between X and Outcome? Do ZDo Z11 and Z and Z22 remove confounding? remove confounding?

UNBIASED EFFECT ESTIMATE?UNBIASED EFFECT ESTIMATE?

Page 51: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

XX OutcomeOutcome

ZZ11

(Pearl. Causality Book)(Pearl. Causality Book)

ZZ22

If X is disconnected from Outcome (d-separation), there is no confounding

Which measurements should be included in the model if we are interested in Which measurements should be included in the model if we are interested in the relation between X and Outcome? the relation between X and Outcome? Do ZDo Z11 and Z and Z22 remove confounding? remove confounding?

UNBIASED EFFECT ESTIMATE!UNBIASED EFFECT ESTIMATE!

Page 52: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

UNBIASED EFFECT ESTIMATE?UNBIASED EFFECT ESTIMATE?

XX OutcomeOutcome

ZZ11

(Pearl. Causality Book)(Pearl. Causality Book)

ZZ22

Page 53: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

XX OutcomeOutcome

ZZ11

(Pearl. Causality Book)(Pearl. Causality Book)

ZZ22

X is NOT disconnected from Outcome

Which measurements should be included in the model if we are interested in Which measurements should be included in the model if we are interested in the relation between X and Outcome? the relation between X and Outcome? Do ZDo Z11, Z, Z22 and Z and Z33 remove confounding? remove confounding?

ZZ33

INCLUDING Z3 INTRODUCES BIAS!

UNBIASED EFFECT ESTIMATE?UNBIASED EFFECT ESTIMATE?

Page 54: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

SELECTION BIAS EXAMPLES

• Observational SpecificObservational Specific Berkson’s BiasBerkson’s Bias Volunteer / Self-selection BiasVolunteer / Self-selection Bias Healthy worker BiasHealthy worker Bias

• Meta-analysis specificMeta-analysis specific Reporting biasReporting bias Publication biasPublication bias

• RCT or ObservationalRCT or Observational Differential loss to follow-upDifferential loss to follow-up Non-response / Missing data biasNon-response / Missing data bias Adjustment for variables affected by previous exposureAdjustment for variables affected by previous exposure

Page 55: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

ATTRITION BIASATTRITION BIAS

RCT Complex Attrition biasRCT Complex Attrition bias

TreatmentTreatment

DeathDeath

Side effectsSide effects Drop OutDrop Out

Mild diseaseMild disease

Condition on common effectCondition on common effect

Page 56: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

EFFECT OF BIAS• Can one sum probability distributions for different risks of bias?Can one sum probability distributions for different risks of bias?

Already being done “intuitively” and informallyAlready being done “intuitively” and informally Some beginnings: response-surface estimation (Greenland), multiple Some beginnings: response-surface estimation (Greenland), multiple

bias modeling (Greenland), adjusted likelihoods (Wolpert), bias against bias modeling (Greenland), adjusted likelihoods (Wolpert), bias against bias (Kaufman)bias (Kaufman)

Treatment BeneficialTreatment Beneficial Treatment HarmfulTreatment Harmful

Bias Towards BenefitBias Towards BenefitBiased Against BenefitBiased Against Benefit

Probability of BiasProbability of Bias

Page 57: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

ESTIMATING BIAS?

XX OutcomeOutcome

(Pearl. Causality Book)

CensoredCensored

Unknown VariableUnknown Variable

Page 58: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

SUMMARY• Objective is to obtain an unbiased estimate of the Objective is to obtain an unbiased estimate of the

parameter of interestparameter of interest

• Study design is only one source of biasStudy design is only one source of bias

• Mathematics underlying statistical analyses do not care Mathematics underlying statistical analyses do not care what the names of the nodes arewhat the names of the nodes are

• Causal maps make assessing bias more transparentCausal maps make assessing bias more transparent

• Meta-analyses should be able to treat all potential biases Meta-analyses should be able to treat all potential biases regardless of causeregardless of cause

Estimating the wrong parameter - RCT, ITT?Estimating the wrong parameter - RCT, ITT? Conditioning on a variable that lies along the causal path or is a Conditioning on a variable that lies along the causal path or is a

marker for a variable lying along the causal pathmarker for a variable lying along the causal path Absence of conditioning on a common causeAbsence of conditioning on a common cause Conditioning on a common effectConditioning on a common effect

Page 59: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

INTERPRETATIONS

QuickTime™ and aYUV420 codec decompressor

are needed to see this picture.

Page 60: Meta-Analysis: Sports Medicine Canadian Academy of Sport Medicine LAcadémie Canadienne de Médecine du Sport Ian Shrier MD, PhD, Dip Sport Med, FACSM Centre

Canadian Academy of Sport MedicineL’Académie Canadienne de Médecine du Sport