Analyses of randomized long term follow-up studies of critically ill patients are complicated due to competing events such as high mortality. Long-term outcomes are not defined for patients that die; for patients who die, we say the long-term outcome is “truncated due to death”. In addition, patients who survive may have missed visits/assessments causing missing data for the long term outcome.
Elizabeth Colantuoni, PhD, explains (click image for the video) why statistical analyses of functional outcomes must account for “truncation due to death.” She then discusses the advantages and disadvantages of the common statistical methods that address truncation due to death. Lastly, she briefly presents a free standalone application developed to help researchers apply these common statistical methods.
The standalone application developed as part of this grant imputes missing data among survivors and then implements common statistical methods that address truncation due to mortality when evaluating long term functional (non-mortality) outcomes (Wang et al. J Stat Softw 2018, Wang et al. Biometrics 2017). The application utilizes Shiny, R and C++ code. The required software is completely contained in the available download (i.e., there is no need to have any statistical software installed on your computer). The application contains documentation and an example dataset for exploring the methods.
1. Download the app
3. Unzip the file
4. Open the executable file “run_composite.exe”
5. Extract all files (you will receive the prompt to extract automatically when you try to open “run_composite.exe”)
6. Once again open the executable file “run_composite.exe”
Dr. Chenguang Wang et al wrote statistical software using R and designed a statistical application (app) using Shiny to address potential bias from unobserved study outcomes due to missed clinic visits, premature withdrawal or death. The app demonstrates the use of a composite endpoint of both the functional outcome and survival as well as other methods (e.g. survivors only analyses and survivor average causal effect). Also, the app allows the user to perform multiple imputation for missing data not due to death.
Wang C, Colantuoni E, Leroux A, Scharfstein DO. idem: An R Package for Inferences in Clinical Trials with Death and Missingness. Journal of Statistical Software. In press. https://cran.r-project.org/web/packages/idem/idem.pdf
Wang C, Scharfstein DO, Colantuoni E, Girard TD, and Yan Y. Inference in Randomized Trials with Death and Missingness. Biometrics. 2017; 73(2): 431-440.
Dr. Elizabeth Colantuoni, et al, evaluated three statistical approaches [survivors analysis, survivor average causal effect (SACE) and composite endpoint approach] to assess the effect of treatment on functional outcomes, such as quality of life, that are often “truncated due to death”. This paper utilizes data from a completed trial of critically ill patients to analyze and highlight important differences between the three statistical approaches.
Colantuoni E, Scharfstein DO, Wang C, Hashem MD, Leroux A, Needham DM, Girard TD. Statistical methods to compare functional outcomes in RCTs with high mortality. British Medical Journal. 2018;360:j5748. Article