G-methods for time-varying treatments (Causal inference: What if, Chapter 21-1)

by Shuntaro Sato

Slide 1

Slide 1 text

No content

Slide 2

Slide 2 text

・G-methods for time-fixed treatments 本日の内容・The g-formula for time-varying treatments ・IP weighting for time-varying treatments ・A doubly robust estimator for time-varying treatments

Slide 3

Slide 3 text

・G-methods for time-fixed treatments 本日の内容・The g-formula for time-varying treatments ・IP weighting for time-varying treatments ・A doubly robust estimator for time-varying treatments

Slide 4

Slide 4 text

Stratification effect measure modification (-) effect measure modification (+) Mantel-Haenszel method 別々にオッズ比を報告

Slide 5

Slide 5 text

Why model? effect measure modification (-) effect measure modification (+) 別々にオッズ比を報告（1つの効果を報告できない） g-methods

Slide 6

Slide 6 text

g-formula A=1を代入 A=0を代入

Slide 7

Slide 7 text

IP weighting marginal structural model

Slide 8

Slide 8 text

Conditional or Marginal? outcome regression saturated parametric stratification g-formula IP weighting g-estimation or algebraically equivalent

Slide 9

Slide 9 text

Time-varying treatment g-methods

Slide 10

Slide 10 text

・G-methods for time-fixed treatments 本日の内容・The g-formula for time-varying treatments ・IP weighting for time-varying treatments ・A doubly robust estimator for time-varying treatments

Slide 11

Slide 11 text

前提・本章ではidentifiability conditions(sequential exchangeability, positivity, and consistency)のviolationがないものとする。・static treatment strategies (always treat vs. never treat) の効果を推定する。

Slide 12

Slide 12 text

g-formula (weighted average) ・time-fixed treatment (A1 の反実アウトカム) ・time-varying treatment

Slide 13

Slide 13 text

g-formula (weighted average)

Slide 14

Slide 14 text

g-formula (weighted average)

Slide 15

Slide 15 text

g-formula (simulation) のシミュレーションと

Slide 16

Slide 16 text

g-formulaの注意点・DAGに基づいたcovariates L1 をモデルに含める・static sequential exchangeabilityが成立すればstatic treatment strategyの効果はidentify可能

Slide 17

Slide 17 text

g-formulaの一般化・static treatment strategy ・dynamic treatment strategy linear regression logistic regression

Slide 18

Slide 18 text

・G-methods for time-fixed treatments 本日の内容・The g-formula for time-varying treatments ・IP weighting for time-varying treatments ・A doubly robust estimator for time-varying treatments

Slide 19

Slide 19 text

IP weighting (weights) ・nonstabilized IP weights ・ stabilized IP weights

Slide 20

Slide 20 text

IP weighting (non-stabilized)

Slide 21

Slide 21 text

Stabilized weights non-stabilized weights: stabilized weights: Lと独立であればよい Lと独立であればよい

Slide 22

Slide 22 text

IP weighting (stabilized)

Slide 23

Slide 23 text

IP weightingの一般化・nonstabilized IP weights ・ stabilized IP weights logistic regression logistic regression （misspecifiedでも可）

Slide 24

Slide 24 text

Marginal Structural Model ・2K > Nのときは推定できない・marginal structural mean model stabilized IP weightsを使って推定 misspecified??

Slide 25

Slide 25 text

Effect Measure Modification ・baseline variable VによるEMMがある場合、marginal structural modelは以下の通り（parametric） stabilized IP weightsを使って推定 Vに入れて良いのはbaseline variableだけ

Slide 26

Slide 26 text

・G-methods for time-fixed treatments 本日の内容・The g-formula for time-varying treatments ・IP weighting for time-varying treatments ・A doubly robust estimator for time-varying treatments

Slide 27

Slide 27 text

Doubly Robust Estimator ・g-formula ・ IP weighting

Slide 28

Slide 28 text

1. Doubly Robust (time-fixed) 2. 3. A=1とA=0でそれぞれを推定を推定 , をLについて標準化

Slide 29

Slide 29 text

1. Doubly Robust (time-varying) 2. 3. を推定からパラメータを求める。を求めておくを推定し、Aの値に応じたを求める。これを繰り返してを求める。 always treat

Slide 30

Slide 30 text

Discussion