comparing covariate adjustment in interventional and
play

Comparing covariate adjustment in interventional and observational - PowerPoint PPT Presentation

Comparing covariate adjustment in interventional and observational studies Markus Kalisch, Seminar fr Statistik, ETH Zrich What is the total causal effect ? Treatment Outcome If we apply treatment , how will outcome


  1. Comparing covariate adjustment in interventional and observational studies Markus Kalisch, Seminar fΓΌr Statistik, ETH ZΓΌrich

  2. What is the total causal effect ? Treatment π‘Œ Outcome 𝑍  If we apply treatment π‘Œ , how will outcome 𝑍 change ?  Data collection: - observational study - interventional study (RCT) 1

  3. Outline for the rest of the talk  Total causal effect and covariate adjustment  Issues in observational studies  Issues in interventional studies  Insights from recent theoretical developments 2

  4. Causal Model: How the real world might look like  We use directed acyclic graphs (DAG) – no feedback loops  Example: DAG 𝐻 π‘Œ 5 π‘Œ 2 π‘Œ 3 𝑍 π‘Œ 1  Terminology: π‘Œ 4 Set of all variables: 𝒀 = {π‘Œ 1 , π‘Œ 2 , … , π‘Œ 5 , 𝑍} Path: π‘Œ 1 , π‘Œ 2 , π‘Œ 3 , 𝑍 Directed p ath = β€œcausal - path”: π‘Œ 1 , π‘Œ 2 , π‘Œ 3 Not directed path = Non-causal path: (π‘Œ 4 ,π‘Œ 3 ,𝑍) Parents p𝑏 π‘Œ 3 = {π‘Œ 2 , π‘Œ 4 } , Children cβ„Ž π‘Œ 1 = π‘Œ 2 Think of family tree Ancestor π‘π‘œ , Descendant 𝑒𝑓 , Non-descendants π‘œπ‘’ 3

  5. More details: Structural Equation Model (SEM)  Example of SEM: π‘Œ 1 = 𝑂 1 Causal π‘Œ 2 = 4π‘Œ 1 + 𝑂 2 interpretation 𝑂 1 , 𝑂 2 ∼ 𝑂 0,1 𝑗𝑗𝑒  Visualization of causal structure : π‘Œ 1 π‘Œ 2  Difference to arbitrary hierarchical system of equations: Due to causal interpretation, solving for a variable on the RHS is not meaningful in SEM. 4

  6. Quantifying the total causal effect Define intervention distribution by replacing (some) structural equations  do-Operator Reference: Pearl, J. (2009). Causality: Models, Reasoning and Inference. 2nd edition. Cambridge Univ. Press. E.g. Β«intervention on π‘Œ Β»:  Old SEM: 𝑇 with equation π‘Œ = 2 + π‘Œ 5 + 𝑂 π‘Œ  New SEM: መ 𝑇 with equation π‘Œ = 4  New SEM generates new distribution: 𝑄 መ 𝑇 (𝒀) = 𝑄 𝑇 (𝒀|𝑒𝑝 π‘Œ = 4 ) and in particular P(𝑍|𝑒𝑝(π‘Œ = 4))  Final goal: Estimate intervention distribution given observational data  Oftentimes: Expectation is enough – e.g. 𝐹(𝑍|𝑒𝑝 π‘Œ = 4 ) 5

  7. Covariate adjustment: Adjustment set  Idea: Identify intervention effects by only using conditional probabilities / expectations Β«doΒ» No Β«doΒ» 𝐢 𝑄 𝑍 = 𝑧 𝑒𝑝 π‘Œ = 𝑦 = ෍ 𝑄 𝑍 = 𝑧 π‘Œ = 𝑦, 𝐢 = 𝑐 𝑄(𝐢 = 𝑐) Adjustment π‘βˆˆπΆ π‘Œ 𝑍 set  Practice: Often interested in 𝐹 𝑍 = 𝑧 𝑒𝑝 π‘Œ = 𝑦  Can show for multivariate Gaussian density: = 𝛽 + 𝛿𝑦 + 𝛾 π‘ˆ 𝐹 𝐢 𝐹 𝑍 𝑒𝑝 π‘Œ = 𝑦 d  Total Causal Effect: 𝑒𝑦 𝐹 𝑍 𝑒𝑝 π‘Œ = 𝑦 = 𝛿 This is the regression coefficient of π‘Œ in the regression of 𝑍 on π‘Œ and 𝐢 6

  8. Outline for the rest of the talk  Total causal effect and covariate adjustment  Issues in observational studies  Issues in interventional studies  Insights from recent theoretical developments 7

  9. Causal Diagram: Example 1 - confounder Lab 1,2,… Treatment Outcome Should we add the lab information as covariate ? 8

  10. Example 1 in numbers π‘Ž π‘Œ 𝑍  𝜁 π‘Œ ∼ 𝑂(0,1) , 𝜁 π‘Ž ∼ 𝑂(0,1) , 𝜁 𝑍 ~ 𝑂(0,1) independent  True causal system: π‘Ž = 𝜁 π‘Ž π‘Œ = 0.7 βˆ— π‘Ž + 𝜁 π‘Œ 𝑍 = 1 βˆ— π‘Œ + 0.5 βˆ— π‘Ž + 𝜁 𝑍  True causal effect of π‘Œ on 𝑍 : 1 If we increase π‘Œ by one unit, 𝑍 will also increase by one unit  Can we estimate the true causal effect with a linear regression ? 9

  11. Example 1 in numbers π‘Ž π‘Œ 𝑍  True causal effect of π‘Œ on 𝑍 : 1  Simple Regression: π‘šπ‘›(𝑍 ~ π‘Œ) Incorrect  Multiple Regression: π‘šπ‘›(𝑍~π‘Œ + π‘Ž) Missing the confounder introduced a bias ! Correct 10

  12. Causal Diagram: Example 2 – selection variable Treatment Outcome Follow-up test Should we add the info of the follow-up test as covariate ? 11

  13. π‘Œ 𝑍 Example 2 in numbers π‘Ž  𝜁 π‘Œ ∼ 𝑂(0,1) , 𝜁 π‘Ž ∼ 𝑂(0,1) , 𝜁 𝑍 ~ 𝑂(0,1) independent  True causal system: X = 𝜁 π‘Œ Y = 0.7 βˆ— π‘Œ + 𝜁 𝑍 Z = 0.8 βˆ— π‘Œ + 0.5 βˆ— 𝑍 + 𝜁 π‘Ž  True causal effect of π‘Œ on 𝑍 : 0.7 If we increase π‘Œ by one unit, 𝑍 will also increase by 0.7 units  Can we estimate the true causal effect with a linear regression ? 12

  14. π‘Œ 𝑍 Example 2 in numbers π‘Ž  True causal effect of π‘Œ on 𝑍 : 0.7  Simple Regression: π‘šπ‘›(𝑍 ~ π‘Œ) Correct  Multiple Regression: π‘šπ‘›(𝑍~π‘Œ + π‘Ž) Including the selection variable Incorrect introduced a bias ! 13

  15. β€œParent Criterion” (PC)  Take parents of π‘Œ as adjustment set (special case of Pearl’s back -door criterion)  Sufficient but not complete  Example 1: π‘Ž π‘Œ 𝑍 PC: π‘Ž is a valid adjustment set; would {} be a valid adjustment set, too β†’ ??? (perhaps we can not measure π‘Ž although we know it exists)  Example 2: π‘Œ 𝑍 π‘Ž PC: {} is a valid adjustment set; would π‘Ž be a valid adjustment set, too β†’ ??? 14

  16. Conclusion 1 In observational studies: Judging if an adjustment set is valid is not trivial 15

  17. Outline for the rest of the talk  Total causal effect and covariate adjustment  Issues in observational studies  Issues in interventional studies  Insights from recent theoretical developments 16

  18. RCT: Evaluation Treatment Treatment Control Control Treatment Control Treatment Control Treatment Control  Cage: Experimental Unit  5 cages with treatment ( π‘Œ = 1 ), 5 cages with control ( π‘Œ = 0 )  Randomize allocation: In causal diagram think of β€œdeleting all incoming edges to π‘Œ ” 17

  19. RCT in causal diagram π‘Ž 1 π‘Ž 1 𝒀 𝐡 𝒁 𝒀 𝐡 𝒁 RCT 𝐷 𝐷 π‘Ž 2 π‘Ž 2 𝐢 𝐢 PC: PC: Valid adjustment set Valid adjustement set is {} is {π‘Ž 1 , π‘Ž 2 } PC: {} is always valid adjustment set after randomization 18

  20. RCT: Evaluation Treatment Treatment Control Control Treatment Control Treatment Control Treatment Control  Given a proper design, we can do a two-sample t-test with two groups (i.e. empty adjustment set).  What if we have more covariates (sex, age, intermediate blood test, follow-up information, …) ?  Is it always better to add covariates to the analysis ? 19

  21. Messing up the evaluation of a randomized controlled trial (RCT)  You can bias ( β€œmess up” ), the analysis by adding the β€œwrong” covariates .  RCT : It is always safe to add no covariates to the analysis.  Adding the β€œright” covariates might increase precision. 20

  22. Causal Diagram: Example 1 Treatment Intermediate Blood T est Outcome Should we add the intermediate blood test as covariate ? 21

  23. Example 1 in numbers π‘Œ π‘Ž 𝑍  𝜁 π‘Œ ∼ 𝑂(0,1) , 𝜁 π‘Ž ∼ 𝑂(0,1) , 𝜁 𝑍 ~ 𝑂(0,1) independent  True causal system: π‘Œ = 𝜁 π‘Œ π‘Ž = 2 βˆ— π‘Œ + 𝜁 π‘Ž 𝑍 = 0.5 βˆ— π‘Ž + 𝜁 𝑍  True causal effect of π‘Œ on 𝑍 : 2 βˆ— 0.5 = 1 If we increase π‘Œ by one unit, 𝑍 will also increase by one unit  Can we estimate the true causal effect with a linear regression ? 22

  24. Example 1 in numbers π‘Œ π‘Ž 𝑍  True causal effect of π‘Œ on 𝑍 : 2 βˆ— 0.5 = 1  Simple Regression: π‘šπ‘›(𝑍 ~ π‘Œ) Correct  Multiple Regression: π‘šπ‘›(𝑍~π‘Œ + π‘Ž) Adding a covariate introduced a bias ! Incorrect 23

  25. Causal Diagram: Example 2 Lab 1,2,… Treatment Outcome Should we add the lab information as covariate ? 24

  26. Example 2 in numbers π‘Œ 𝑍 π‘Ž  𝜁 π‘Œ ∼ 𝑂(0,1) , 𝜁 π‘Ž ∼ 𝑂(0,1) , 𝜁 𝑍 ~ 𝑂(0,1) independent  True causal system: π‘Œ = 𝜁 π‘Œ π‘Ž = 𝜁 π‘Ž 𝑍 = 1 βˆ— π‘Œ + 0.5 βˆ— π‘Ž + 𝜁 𝑍  True causal effect of π‘Œ on 𝑍 : 1 If we increase π‘Œ by one unit, 𝑍 will also increase by one unit  Can we estimate the true causal effect with a linear regression ? 25

  27. Example 2 in numbers π‘Œ 𝑍 π‘Ž  True causal effect of π‘Œ on 𝑍 : 1  Simple Regression: π‘šπ‘›(𝑍~π‘Œ) Correct  Multiple Regression: π‘šπ‘›(𝑍~π‘Œ + π‘Ž) β€’ Adding a covariate did not introduce a bias Correct β€’ Confidence interval with covariate is slightly smaller (0.12 vs 0.14) 26

  28. Summary  Adding the wrong variable will introduce a bias β€œ Wrong variable ”: On causal path from π‘Œ to 𝑍 or Β«descendantsΒ» of those nodes ( post-intervention )  Adding the right variables might increase precision π‘Ž 1 π‘Ž 2 β€œ Right variable ”: Parents of nodes on causal path from π‘Œ to 𝑍 ( pre-intervention ) 𝒀 𝐡 𝒁 𝐷  Problem in practice: 𝐢 Usually don’t know true causal structure! What are β€œright” and β€œwrong” variables ?  If in doubt, don’t use covariate !  Safe variables: Things that clearly β€œpreceded” π‘Œ (e.g. gender) 27

  29. Outline for the rest of the talk  Total causal effect and covariate adjustment  Issues in observational studies  Issues in interventional studies  Insights from recent theoretical developments 28

  30. Adjustment Criteria Getting the β€œright estimate”:  given causal structure, criterion to check if a set is a valid adjustment set  assuming causal structure is a strong assumption in practice  discussion can shift to discussing reasonable causal structures  Pearl’s back -door criterion  Generalized Adjustment criterion 29

Recommend


More recommend