Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

44
Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007
  • date post

    21-Dec-2015
  • Category

    Documents

  • view

    217
  • download

    0

Transcript of Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

Page 1: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

Experiments and Dynamic Treatment Regimes

S.A. MurphyAt NIAID, BRB December, 2007

Page 2: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

2

Outline

• Dynamic Treatment Regimes

• Challenges in Experimentation

• Defining Effects and Aliasing

• Example

Page 3: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

3

Dynamic treatment regimes are individually tailored treatments, with treatment type and dosage changing with ongoing subject need. Mimic Clinical Practice.

•High variability across patients in response to any one treatment

•Relapse is likely without either continuous or intermittent treatment for a large proportion of people.

•What works now may not work later

•Exacerbations in disorder may occur if there are no alterations in treatment

Page 4: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

4

The Big Questions

•What is the best sequencing of treatments?

•What is the best timings of alterations in treatments?

•What information do we use to make these decisions?

Page 5: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

5

Two stages of treatment for each individual

Observation available at jth stage

Treatment (vector) at jth stage

Primary outcome Y is a specified summary of decisions and observations

Page 6: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

6

A dynamic treatment regime is a vector of decision rules, one per decision

where each decision rule

inputs the available information

and outputs a recommended treatment decision.

Page 7: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

7

Challenges in Experimentation

Page 8: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

8

Dynamic Treatment Regimes (review)

Constructing decision rules is a multi-stage decision problem in which the system dynamics are unknown.

High quality data is likely provided by sequential multiple assignment randomized trials: randomize at each decision point— à la full factorial.

Page 9: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

9

ExTENd

• Ongoing study at U. Pennsylvania (D. Oslin)

• Goal is to learn how best to help alcohol dependent individuals reduce alcohol consumption.

Page 10: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

10

Oslin ExTENd

Late Trigger forNonresponse

8 wks Response

TDM + Naltrexone

CBIRandom

assignment:

CBI +Naltrexone

Nonresponse

Early Trigger for Nonresponse

Randomassignment:

Randomassignment:

Randomassignment:

Naltrexone

8 wks Response

Randomassignment:

CBI +Naltrexone

CBI

TDM + Naltrexone

Naltrexone

Nonresponse

Page 11: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

11

Challenges in ExperimentationDynamic Treatment Regimes are multi-component treatments:

many possible components

• decision options for improving patients are often different from decision options for non-improving patients (T2 is a vector and differs by outcomes observed during initial treatment)

• multiple components employed simultaneously

• medications, behavioral therapy, adjunctive treatments, delivery mechanisms, motivational therapy, staff training, monitoring schedule…….

• Future: series of screening/refining, randomized trials prior to confirmatory trial --- à la Fisher/Box

Page 12: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

12

Screening experiments (review)1) Goal is to eliminate inactive factors (e.g. components) and

inactive effects.

2) Each factor at 2 levels

3) Screen main effects and some interactions

4) Design experiment using working assumptions concerning the negligibility of certain higher order factorial effects.

5) Designs and analyses permit one to determine aliasing (caused by false working assumptions)

6) Minimize formal assumptions

Page 13: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

13

• Stage 1 Factors: T1={A, B, C, D}, each with 2 levels

• Stage 1 outcome:

• Stage 2 Factors: T2= {F2--only if R=1, G2—only if R=0}, each with 2 levels

• Primary Outcome: Y continuous

(26= 64 simple dynamic treatment regimes)

Simple Example for Two Stages

Page 14: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

14

Two Stage Design: I=ABCDF2=ABCDG2

A B C D F2=G2

- - - - +

- - - + -

- - + - -

- - + + +

- + - - -

- + - + +

- + + - +

- + + + -

+ - - - -

+ - - + +

+ - + - +

+ - + + -

+ + - - +

+ + - + -

+ + + - -

+ + + + +

Page 15: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

15

Screening experiments

Can we:

design screening experiments using working assumptions concerning higher order effects

&

determine the aliasing and provide an analysis method?

Page 16: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

16

Defining the Effects

Page 17: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

17

Reality

UnknownCauses

T1 R T2 Y

Page 18: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

18

Defining the stage 2 effects

Simple case: two stages of treatment with only one factor at each stage, a early measure of response, R at the end of stage 1 and a primary outcome Y. The potential outcomes are

Define effects involving T2 in a saturated linear model for

Page 19: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

19

Defining the stage 2 effects

Suppose the factors T1 and T2 are randomized. Assume consistency (Robins, 1997) then

Page 20: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

20

Defining the stage 2 effects

Define (factorial) effects involving T2 in a saturated linear model

Page 21: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

21

Defining the stage 2 effects

Page 22: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

22

Defining the stage 1 effects (T1)

UnknownCauses

T1 R T2 Y

Page 23: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

23

Defining the stage 1 effects

UnknownCauses

T1 R T2 Y

Page 24: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

24

Defining the stage 1 effects

Define effects involving only T1 in a saturated linear model

The above is equal to

when {T1, T2} are randomized and T2 has a discrete uniform distribution on {-1,1}.

Page 25: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

25

Defining the stage 1 effects

In general when {T1, T2} are randomized

Page 26: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

26

Why marginal, why uniform?Define effects involving only T1 via

1) The defined effects are causal.

2) The defined effects are consistent with tradition in experimental design for screening.

– The main effect for one treatment factor is defined by marginalizing over the remaining treatment factors using a discrete uniform distribution.

Page 27: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

27

Surprisingly both stage 1 and 2 effects can be represented in one (nonstandard) linear model:

Representing the effects

Page 28: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

28

where

Causal effects:

Nuisance parameters: and

Page 29: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

29

Two Stage Design: I=ABCDF2=ABCDG2

A B C D F2=G2

- - - - +

- - - + -

- - + - -

- - + + +

- + - - -

- + - + +

- + + - +

- + + + -

+ - - - -

+ - - + +

+ - + - +

+ - + + -

+ + - - +

+ + - + -

+ + + - -

+ + + + +

Page 30: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

30

General FormulaSaturated model (for both R=0, 1)

Z1 matrix of stage 1 factor columns, Z2 is the matrix of stage 2 factor columns, Y is a vector, p is the vector of response rates

Classical saturated factorial effects model

Page 31: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

31

Aliasing

Page 32: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

32

Aliasing (Identification of Effects)

In classical designs, the defining words immediately yield the aliasing. For example I=ABCDF means that only the sum of the main effect of F and the four way interaction, ABCD can be identified. The two effects can not be separately estimated.

We can do the same here with a caveat.

Page 33: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

33

Aliasing (Identification of Effects) The rows of {Z1, Z2} are determined by the

experimental design; each column in Z1 is associated with a stage 1 factorial effect and similarly each column in Z2 is associated with a stage 2 factorial effect.

Fractional factorial designs lead to common columns in {Z1, Z2}; these columns are given by the defining words.

Page 34: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

34

Aliasing (Identification of Effects)

Consider designs for which the defining words indicate a shared column in both Z1 and Z2 only if the column in Z1 can be safely assumed to have a zero η coefficient or if the column in Z2 can be safely assumed to have a zero β, α coefficient. The defining words then provide the aliasing in this case.

Page 35: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

35

Six Factors:

Stage 1: T1={A, B, C, D}, each with 2 levels

Stage 2: T2= {F2--only for stage 1 responders,

G2--only for stage 1 nonresponders}, each with 2 levels

(26= 64 simple dynamic treatment regimes)

Simple Example for Two Stages

Page 36: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

36

Two Stage Design: I=ABCDF2=ABCDG2

A B C D F2=G2

- - - - +

- - - + -

- - + - -

- - + + +

- + - - -

- + - + +

- + + - +

- + + + -

+ - - - -

+ - - + +

+ - + - +

+ - + + -

+ + - - +

+ + - + -

+ + + - -

+ + + + +

Page 37: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

37

Simple Example

• This design has defining words: 1=ABCDF2=ABCDG2

• The labels of the identical columns in Z1, Z2 are CD=ABG2, ABC=DG2, F2DC=AB………..

Page 38: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

38

Formal Assumptions for this Design

• We use this design if we can

•assume that all three way and higher order stage 2 effects are negligible (ABG2, ABF2, ABCG2, ABCF2, ….)—these are all β and α parameters.

•assume all four way and higher order effects involving R and stage 1 factors are negligible (R-p)ABCD, (R-p)ABC, (R-p)ABD …--- these are all η parameters.

1=ABCDF2=ABCDG2

Page 39: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

39

Aliasing for this Design

• Stage 1 main effects and stage 2 four-way interactions are aliased (e.g. A=BCDF2, etc.).

• Stage 1 two way interactions and stage 2 three way interactions are aliased (e.g. CD=ABG2, AB=CDF2, etc.)

• Stage 1 three way interactions and stage 2 two way interactions are aliased (e.g. ABC=DF2, ABC=DG2, etc.)

• The Stage 1 four way interaction and stage 2 main effects are aliased (e.g. ABCD=F2, ABCD=G2).

1=ABCDF2=ABCDG2

Page 40: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

40

Working Assumptions for this Design

• Stage 2 three-way and four-way interactions are negligible.

• Stage 1 three-way and four-way interactions are negligible.

If the working assumptions are correct then we will be able to estimate the “non-aliased” main and two-way interactions for both stage 1 and stage 2.

Page 41: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

41

Analysis for this designRecall that

• Many columns in Z1, Z2 are identical (hence the aliasing of effects). Eliminate all multiple copies of columns and label remaining columns as stage 1 (or stage 2) main and two-way interaction effects.

• Replace response rates in p by observed response rates.

• Fit model.

Page 42: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

42

Interesting Result in Simulations

• In simulations formal assumptions are violated.

• Response rates (probability of R=1) across 16 cells range from .55 to .73

• Results are surprisingly robust to violations of formal assumptions.

•The maximal value of the correlation between 32 estimators of effects was .12 and average absolute correlation value is .03

•Why? Binary R variables can not vary that much. If response rate is constant, then the effect estimators are uncorrelated as in classical experimental design.

Page 43: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

43

Discussion

• Compare this to using observational studies to construct dynamic treatment regimes– Uncontrolled selection bias (causal misattributions)– Uncontrolled aliasing.

• Secondary analyses would assess if other variables collected during treatment should enter decision rules.

Page 44: Experiments and Dynamic Treatment Regimes S.A. Murphy At NIAID, BRB December, 2007.

44

• Joint work with– Derek Bingham (Simon Fraser)

• And informed by discussions with– Vijay Nair (U. Michigan)– Bibhas Chakraborty (U. Michigan)– Vic Strecher (U. Michigan)

• This seminar can be found at: http:// www.stat. lsa.umich.edu/~samurphy/seminars/NIAID12.07.ppt