imputation of missing covariates when standard methods
play

Imputation of missing covariates: when standard methods may fail - PowerPoint PPT Presentation

Imputation of missing covariates: when standard methods may fail Nicole S. Erler 1 , 2 , Dimitris Rizopoulos 1 , Oscar H. Franco 2 , Emmanuel M.E.H. Lesaffre 1 , 3 1 Department of Biostatistics, Erasmus MC, Rotterdam, the Netherlands 2 Department


  1. Imputation of missing covariates: when standard methods may fail Nicole S. Erler 1 , 2 , Dimitris Rizopoulos 1 , Oscar H. Franco 2 , Emmanuel M.E.H. Lesaffre 1 , 3 1 Department of Biostatistics, Erasmus MC, Rotterdam, the Netherlands 2 Department of Epidemiology, Erasmus MC, Rotterdam, the Netherlands 3 L-Biostat, KU Leuven, Leuven, Belgium

  2. Motivation (1) Vitamin D concentration during fetal life and bone health at age 6 • bone mineral content (BMC) • serum vitamin D concentration ( ✻ ) • sun exposure ( ✻ ), season at measurement ( ✻ ) • gender, age at measurement • . . . ( ✻ ) ( ✻ ) incomplete Analysis model: BMD = ( age + V itD + V itD 2 ) × gender + season + sun exposure + . . . Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 1

  3. Motivation (2) Maternal sugar-sweetened bevarage consumption and child’s body composition • child BMI at up to 13 time points • maternal sugar-sweetened bevarage consumption (SBC) • child’s physical activity, TV watching ( ✻ ) • gender, age at measurement • . . . ( ✻ ) ( ✻ ) incomplete Analysis model: BMI ij = SBC i + age ij + . . . + u 0 i + u 1 i × age ij Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 2

  4. Standard for imputation: Multiple Imputation (MI) impute ➡ analyze ➡ pool fully conditional specification ( FCS ) joint model imputation chained equations ( MICE ) ➡ In iteration k = 1 , . . . , K : for variable j = 1 , . . . , p : � k k j ∼ p ( θ k e.g. regression with ❼ Draw parameter ˆ , ˆ j | x obs − j ) θ X j all other variables k x k , X k − j , ˆ j ∼ p ( x mis | x obs ❼ Draw imputation ˆ j ) θ in the lin. predictor j j ➡ keep last iteration ➡ 1 imputed data set ➡ repeat m times Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 3

  5. Standard for imputation: Multiple Imputation (MI) impute ➡ analyze ➡ pool fully conditional specification ( FCS ) joint model imputation chained equations ( MICE ) ➡ In iteration k = 1 , . . . , K : for variable j = 1 , . . . , p : � k k j ∼ p ( θ k e.g. regression with ❼ Draw parameter ˆ , ˆ j | x obs − j ) θ X j all other variables k x k , X k − j , ˆ j ∼ p ( x mis | x obs ❼ Draw imputation ˆ j ) θ in the lin. predictor j j ➡ keep last iteration ➡ 1 imputed data set ➡ repeat m times Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 3

  6. Standard for imputation: Multiple Imputation (MI) impute ➡ analyze ➡ pool fully conditional specification ( FCS ) joint model imputation chained equations ( MICE ) ➡ In iteration k = 1 , . . . , K : for variable j = 1 , . . . , p : � k k j ∼ p ( θ k e.g. regression with ❼ Draw parameter ˆ , ˆ j | x obs − j ) θ X j all other variables k x k , X k − j , ˆ j ∼ p ( x mis | x obs ❼ Draw imputation ˆ j ) θ in the lin. predictor j j ➡ keep last iteration ➡ 1 imputed data set ➡ repeat m times Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 3

  7. Standard for imputation: Multiple Imputation (MI) impute ➡ analyze ➡ pool fully conditional specification ( FCS ) joint model imputation chained equations ( MICE ) ➡ In iteration k = 1 , . . . , K : for variable j = 1 , . . . , p : � k k j ∼ p ( θ k e.g. regression with ❼ Draw parameter ˆ , ˆ j | x obs − j ) θ X j all other variables k x k , X k − j , ˆ j ∼ p ( x mis | x obs ❼ Draw imputation ˆ j ) θ in the lin. predictor j j ➡ keep last iteration ➡ 1 imputed data set ➡ repeat m times Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 3

  8. Requirements for MICE • all relevant variables must be included – covariates (from all analyses) – the outcome • compatibility: a joint model exists that has the imputation models as its conditional distributions • congeniality: compatibility between analysis model and imputation model • imputation models should fit the data • M(C)AR (in most implementations) Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 4

  9. When MICE might fail Imputation model not congenial with analysis: • quadratic, logarithmic, . . . effects • interactions between covariates Complex (non univariate) outcomes: • survival • longitudinal Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 5

  10. Uncongeniality y = β 0 + β 1 x 1 + β 2 x 2 True model: 1 + . . . (quadratic association) Imputation model: x 1 = θ 10 + θ 11 y + . . . (linear association) y x 1 Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 6

  11. Uncongeniality y = β 0 + β 1 x 1 + β 2 x 2 True model: 1 + . . . (quadratic association) Imputation model: x 1 = θ 10 + θ 11 y + . . . (linear association) y x 1 Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 6

  12. Uncongeniality y = β 0 + β 1 x 1 + β 2 x 2 True model: 1 + . . . (quadratic association) Imputation model: x 1 = θ 10 + θ 11 y + . . . (linear association) y x 1 Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 6

  13. Uncongeniality y = β 0 + β 1 x 1 + β 2 x 2 True model: 1 + . . . (quadratic association) Imputation model: x 1 = θ 10 + θ 11 y + . . . (linear association) y original imputed x 1 Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 6

  14. Simple approaches • passive normal imputation: standard MICE ➡ calculate interactions & non-lin. terms afterwards • predictive mean matching (pmm) (also passive) use pmm instead of linear regression for imputation • just another variable – calculate interactions & non-lin. terms before imputation – add as columns to data set (Can be done in SPSS) Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 7

  15. Simple approaches • passive normal imputation: standard MICE ➡ calculate interactions & non-lin. terms afterwards • predictive mean matching (pmm) (also passive) use pmm instead of linear regression for imputation • just another variable – calculate interactions & non-lin. terms before imputation – add as columns to data set (Can be done in SPSS) Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 7

  16. Simple approaches • passive normal imputation: standard MICE ➡ calculate interactions & non-lin. terms afterwards • predictive mean matching (pmm) (also passive) use pmm instead of linear regression for imputation • just another variable – calculate interactions & non-lin. terms before imputation – add as columns to data set (Can be done in SPSS) Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 7

  17. Some advanced approaches • smcfcs: S ubstantive M odel C ompatible FCS ➡ MICE type approach • jomo: joint modeling MI using multivariate normal distribution ➡ joint model MI • JointAI: joint analysis and imputation ➡ not MI, but simultaneous analysis & imputation Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 8

  18. Some advanced approaches • smcfcs: S ubstantive M odel C ompatible FCS ➡ MICE type approach • jomo: joint modeling MI using multivariate normal distribution ➡ joint model MI • JointAI: joint analysis and imputation ➡ not MI, but simultaneous analysis & imputation Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 8

  19. Some advanced approaches • smcfcs: S ubstantive M odel C ompatible FCS ➡ MICE type approach • jomo: joint modeling MI using multivariate normal distribution ➡ joint model MI • JointAI: joint analysis and imputation ➡ not MI, but simultaneous analysis & imputation Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 8

  20. Some advanced approaches • smcfcs: S ubstantive M odel C ompatible FCS ➡ MICE type approach • jomo: joint modeling MI using multivariate normal distribution ➡ joint model MI • JointAI: joint analysis and imputation ➡ not MI, but simultaneous analysis & imputation Explicitly take into account the analysis model in the sampling distribution for ˆ x j Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 8

  21. Simulation study (I): Data setup Models: linear regression with • interaction • logarithmic or quadratic effect • combinations Missing values: • in one or two covariates • MAR, depending on outcome (and other covariate) • 20%, 40%, 60% Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 9

  22. Simulation study (I): Data setup Models: linear regression with • interaction • logarithmic or quadratic effect • combinations Missing values: • in one or two covariates • MAR, depending on outcome (and other covariate) • 20%, 40%, 60% Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 9

  23. Simulation study (I): Methods Approaches using the mice package: • norm • pmm • JAV (using pmm ) other packages: • smcfcs: smcfcs() • jomo: jomo.lm() • JointAI: lm imp() Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 10

  24. (effect of c 2 qdr. with interaction: y ∼ c 1 + ( c ( ∗ ) + c 2( ∗ ) 2 × b ) ) × b ( ∗ ) 2 2 Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 11

  25. Summary of Simulation Study (I) interaction log quadratic interact & qdr norm pmm JAV � smcfcs � jomo JointAI Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 12

  26. When MICE might fail � Imputation model not congenial with analysis: • quadratic, logistic, . . . , effects • interactions between covariates Complex (non univariate) outcomes: • survival • longitudinal Nicole Erler, 38th Conference of the ISCB, Vigo, 2017 13

Recommend


More recommend