presentations in this series overview and randomization self-matching proxies intermediates

74
Presentations in this series 1. Overview and Randomization 2. Self-matching 3. Proxies 4. Intermediates 5. Instruments 6. Equipoise Avoiding Bias Due to Unmeasured Covariates Alec Walker

Upload: iain

Post on 23-Feb-2016

57 views

Category:

Documents


0 download

DESCRIPTION

Avoiding Bias Due to Unmeasured Covariates. Presentations in this series Overview and Randomization Self-matching Proxies Intermediates Instruments Equipoise. Alec Walker. X. T. D. X. Randomization. T. D. X. Randomization. Self-matching. T. D. X. Randomization. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Presentations in this series1. Overview

and Randomization2. Self-matching3. Proxies4. Intermediates5. Instruments6. Equipoise

Avoiding Bias Due toUnmeasured Covariates

Alec Walker

Page 2: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

T D

X

Page 3: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

T D

XRandomization

Page 4: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

T D

XRandomizationSelf-matching

Page 5: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

T D

XSelf-matchingProxies Proxies

Randomization

Page 6: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

6

A textbook definition fromeconometrics.

Page 7: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

7

Let O be an outcome (either T treatment or D disease)P be a proxyX be an unmeasured covariate

P is a proxy for X with respect to O if thedistribution of O given P is identical to the distribution of O given P and X

Which is to say that X adds no information about O, if you know P.

A textbook definition fromeconometrics.

Page 8: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

8

Let O be an outcome (either T treatment or D disease)P be a proxyX be an unmeasured covariate

P is a proxy for X with respect to O if thedistribution of O given P is identical to the distribution of O given P and X

Which is to say that X adds no information about O, if you know P.

Note that O, P and X could all be multidimensional, that is vectors of outcomes, proxies and unmeasured covariates, respectively. This definition could also be conditioned on other, measured covariates.

A textbook definition fromeconometrics.

Page 9: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Proxy variables areCorrelates of an unmeasured covariate

That are useful to the extent that they capture the influence of the unmeasured covariate on a third characteristic

Control for a proxy replaces control for the unmeasured covariate

9

Page 10: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

10

Interview responses may be proxies for – Historical measurements (diet, smoking, alcohol …)– Internal states– Genetic traits

Biological markers are proxies for biological processesAge, sex, SES are stand-ins for their many correlates

.

Examples of proxies

Page 11: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

11

Interview responses may be proxies for – Historical measurements (diet, smoking, alcohol …)– Internal states– Genetic traits

Biological markers are proxies for biological processesAge, sex, SES are stand-ins for their many correlates

.

Examples of proxies

In diabetics, retinal vascular disease is a proxy for vascular disease more generally and is easily ascertained by funduscopic examination. In looking at determinants of myocardial infarction, control for retinal vascular disease could represent control for coexisting vascular pathology.

Page 12: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

https://www.myhealth.va.gov/mhv-portal-web/anonymous.portal?_nfpb=true&_pageLabel=commonConditions&contentPage=va_health_library/diabetic_retinopathy_advanced_info.html

Early diabetic retinopathySource: US Department of Veterans Affairs

Page 13: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

https://www.myhealth.va.gov/mhv-portal-web/anonymous.portal?_nfpb=true&_pageLabel=commonConditions&contentPage=va_health_library/diabetic_retinopathy_advanced_info.html

microaneurysms

Early diabetic retinopathySource: US Department of Veterans Affairs

Page 14: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

https://www.myhealth.va.gov/mhv-portal-web/anonymous.portal?_nfpb=true&_pageLabel=commonConditions&contentPage=va_health_library/diabetic_retinopathy_advanced_info.html

Advanced diabetic retinopathySource: US Department of Veterans Affairs

Page 15: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

D

X

15

T

Page 16: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

D

XP

16

T

Page 17: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

D

XP

17

T D

X

Page 18: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

D

XP UD

18

T D

X

UT

Page 19: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

D

XP UD

19

T D

X

UT

Page 20: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

P UD

UX

20

T D

X

UT

Page 21: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

D

XP UD

21

T D

X

UT

UX

Page 22: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

D

XP UD

22

T D

X

UT

UX

Page 23: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Thialozinedionesfor diabetes

Acute myocardial infarction

Coronary artery

disease

UT

(Unmeasured) Severity of Diabetes

Retinal vascular disease

UD

23

Page 24: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Without mechanistic information, for each of these situations,

( covariate causes proxyproxy causes covariateboth caused by a third factor )

… the proxy looks like a transformation of the predictor, with added error.

Proxy value = f(Predictor value) + error

24

Page 25: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

An accurate proxy

25

Page 26: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Treated

Untreated

The true value of the unmeasured covariate is a predictor of treatment

An accurate proxy

26

Page 27: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

The proxy predicts treatment almost as well as does the true value.

Treated

Untreated

The true value of the unmeasured covariate is a predictor of treatment

An accurate proxy

27

Page 28: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

The proxy almost p

erfectly

represents

the value of th

e unmeasured covaria

te.

Treated

Untreated

An accurate proxy

28

Page 29: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Treated

Untreated

An accurate proxy

29

Page 30: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Treated

Untreated

An accurate proxy

30

Page 31: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

An accurate proxy

31

Page 32: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

The proportion of treated among subjects in a particular small range of proxy values

An accurate proxy

32

Page 33: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

The proportion of treated among subjects in a particular small range of proxy values

An accurate proxy

33

Page 34: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

The proportion of treated among subjects in a particular small range of proxy values

… is the same as the proportion of treated among subjects in the corresponding small range of true values.

An accurate proxy

34

Page 35: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

The true value does not provide further information, if you know the proxy.

An accurate proxy

35

Page 36: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Treated

Untreated

Two accurate proxies

36

Page 37: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Two good proxie

s are highly

correlated with

one another.Treated

Untreated

Two accurate proxies

37

Page 38: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Either proxy provides good prediction of treatment.

Treated

Untreated

Two accurate proxies

38

Page 39: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Untreated

Proxies with substantial

random errorUntreated

Treated

39

Page 40: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

UntreatedTh

e prox

y is s

till corre

lated

with th

e unkn

own mea

sure.

Proxies with substantial

random errorUntreated

Treated

40

Page 41: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Treatment is still associated with higher values of the proxy, but thediscriminationis muchworse.

Proxies with substantial

random error

41

Page 42: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Proxies with substantial

random errorTreated

Untreated

42

Page 43: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Proxies with substantial

random errorTreated

Untreated

The corre

lation between th

e two

proxy measures is

still e

vident.

43

Page 44: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Both proxies show poor discrimination between treated and untreated.

Proxies with substantial

random errorTreated

Untreated

44

Page 45: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

The two proxies can be combined into a function that discriminates better than either proxy alone.

Proxies with substantial

random errorTreated

Untreated

45

Page 46: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

46

A textbook definition fromeconometrics.

Page 47: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

47

Let O be an outcome (either T treatment or D disease)P be a proxyX be an unmeasured covariate

P is a proxy for X with respect to O if thedistribution of O given P is identical to the distribution of O given P and X.

A textbook definition fromeconometrics.

Page 48: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

48

Page 49: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

49

Page 50: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

50

Page 51: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

51

Let O be an outcome (either T treatment or D disease)P be a proxyX be an unmeasured covariate

P is a proxy for X with respect to O if thedistribution of O given P is identical to the distribution of O given P and X.

None of the causal graphs or correlation patterns that we’ve looked at so far produce

this behavior, unless the proxy is perfect.

What are the economists talking about?

A textbook definition fromeconometrics.

Page 52: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

52

Proxy variables can correspond to different components of a composite predictor

Proxy A = f(Predictor Component A) + error A

Proxy B = f(Predictor Component B) + error B

Page 53: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

53

Proxy variables can correspond to different components of a composite predictor.For example, “Severity of Diabetes.”

Hemoglobin A1C

= f(Glucose control last 90 days) + error A

Retinal vascular disease = f(Vascular damage) + error B

Page 54: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Thialozinedionesfor diabetes

Acute myocardial infarction

Coronary artery

disease

UT

Retinal vascular disease

UD

54

UX

Page 55: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Thialozinedionesfor diabetes

Acute myocardial infarction

Coronary artery

disease

UT

Retinal vascular disease

UDHb A1C

UY

Diabetes Mellitus

55

UX

Page 56: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Thialozinedionesfor diabetes

Acute myocardial infarction

Coronary artery

disease

UT

Retinal vascular disease

UDHb A1C

UY

Diabetes Mellitus

56

UX

Page 57: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

57

Treated

Untreated

Trea

ted

Unt

reat

ed

Proxies for components of a composite variable

Page 58: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

58

The proxy measures are uncorrelated with one another.

Treated

Untreated

Trea

ted

Unt

reat

ed

Proxies for components of a composite variable

Page 59: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

59

Proxy A captures more of the distinction.

Proxy B captures none of the distinction between treatments.

Treated

Untreated

Trea

ted

Unt

reat

ed

Proxies for components of a composite variable

Page 60: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

When you have several candidate proxies for an unmeasured covariate, examine them simultaneously for prediction of the outcome (treatment, disease or both), and retain only those that do.

60

Page 61: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

When you have several candidate proxies for an unmeasured covariate, examine them simultaneously for prediction of the outcome (treatment, disease or both), and retain only those that do.

61

Page 62: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

When you have several candidate proxies for an unmeasured covariate, examine them simultaneously for prediction of the outcome (treatment, disease or both), and retain only those that do.

Measurement error Correlated proxies Keeps all relevant ones

62

Page 63: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

When you have several candidate proxies for an unmeasured covariate, examine them simultaneously for prediction of the outcome (treatment, disease or both), and retain only those that do.

Measurement error Correlated proxies Keeps all relevant onesProxies for components of composite unmeasured covariate Uncorrelated proxies Keeps the correct predictor.

63

Page 64: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

When you have several candidate proxies for an unmeasured covariate, examine them simultaneously for prediction of the outcome (treatment, disease or both), and retain only those that do.

Measurement error Correlated proxies Keeps all relevant onesProxies for components of composite unmeasured covariate Uncorrelated proxies Keeps the correct predictor.

Propensity scores (composite multi-variate treatment predictors), allow you to account for both settings.

64

Page 65: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

65

Multidimensional proxy variablescreated through the use of propensity scores

Page 66: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

66

The physician’s belief in the patient’s risk for peptic ulcer and bleeding cannot be measured directly. But we can look to known correlates of treatment choice as measures of the physician’s belief and treat these as proxy variables.

Celecoxibversus

Naproxen

PUBHospital

Admission

MD-perceived risk of peptic ulcer & bleeding (PUB)

True risk of PUB

Page 67: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

67

Page 68: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

68

After extensive propensity matching

68

Page 69: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

69

After control for correlates that completely capture perceived PUB diathesis, there is no further confounding.

Celecoxibversus

Naproxen

PUBHospital

Admission

MD-perceived risk of peptic ulcer & bleeding (PUB)

True risk of PUB

Page 70: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

70

After control for correlates that completely capture perceived PUB diathesis, there is no further confounding.

Celecoxibversus

Naproxen

PUBHospital

Admission

MD-perceived risk of peptic ulcer & bleeding (PUB)

True risk of PUB

Page 71: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

71

After control for correlates that completely capture perceived PUB diathesis, there is no further confounding.

Celecoxibversus

Naproxen

PUBHospital

Admission

MD-perceived risk of peptic ulcer & bleeding (PUB)

True risk of PUB

Page 72: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Primary Discharge Diagnosis N % N % RR

With control for many, many proxies a strong effect emerges.

72

Page 73: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

73

A proxy is (1) a correlate that (2) captures the effect of an unmeasured covariate on either treatment or disease.

Whether a correlate is a proxy is defined only in respect of a third, predicted variable.

Strong correlates may be only weak proxies.Composite (multidimensional) proxies

are useful when no single candidate proxy captures the unmeasured covariate.

Propensity scoring creates multidimensional proxies.

Page 74: Presentations in this series Overview  and Randomization Self-matching Proxies Intermediates

Presentations in this series1. Overview

and Randomization2. Self-matching3. Proxies4. Intermediates5. Instruments

Avoiding Bias Due toUnmeasured Covariates

Alec Walker

74