Related Experiment Videos
Survival analysis with time-dependent covariates subject to missing data or measurement error: Multiple Imputation
Margarita Moreno-Betancur1,2, John B Carlin3,4, Samuel L Brilleman5
1Department of Epidemiology and Preventive Medicine, Monash University, 99 Commercial Rd, Melbourne, VIC, Australia.
Biostatistics (Oxford, England)
|October 18, 2017
Summary
The Multiple Imputation for Joint Modeling (MIJM) approach effectively handles time-dependent covariates in survival models, outperforming traditional methods like last observation carried forward (LOCF). This method offers a robust solution for complex epidemiological data analysis.
Area of Science:
- Epidemiology
- Biostatistics
- Statistical Modeling
Background:
- Epidemiological studies increasingly use time-varying covariates (e.g., BMI, blood pressure).
- Traditional methods like last observation carried forward (LOCF) introduce bias with time-dependent covariates due to measurement error or missing data.
- Existing joint models for longitudinal and time-to-event data can be complex to specify, especially with multiple markers, and software support is limited.
Purpose of the Study:
- To propose a flexible two-stage approach, Multiple Imputation for Joint Modeling (MIJM), for incorporating multiple time-dependent continuous covariates into survival models.
- To address challenges posed by measurement error and non-synchronous covariate updates in time-to-event analyses.
- To develop an accessible R package ('survtd') for implementing the MIJM approach.
Main Methods:
- Proposed the Multiple Imputation for Joint Modeling (MIJM) approach, a two-stage method.
- Utilized multiple imputation by chained equations to handle the joint distribution of multiple longitudinal markers.
- Applied MIJM to semi-parametric Cox and additive hazard models, focusing on the time-to-event outcome.
- Developed the R package 'survtd' for practical application of MIJM and other methods.
Main Results:
- Simulation studies demonstrated that MIJM performs well across various scenarios, showing favorable bias and coverage probabilities.
- MIJM outperformed traditional LOCF, simpler two-stage methods, and a Bayesian joint model in simulations.
- The Framingham Heart Study was used to illustrate the practical application of the MIJM approach.
Conclusions:
- MIJM provides a computationally convenient and effective method for analyzing time-to-event data with multiple time-dependent covariates.
- The approach overcomes limitations of traditional methods and complex joint modeling techniques.
- The 'survtd' R package facilitates the application of MIJM in epidemiological research.