Sequential Trial Emulation

Implementation of sequential trial emulation for the analysis of observational databases. The 'SEQTaRget' software accommodates time-varying treatments and confounders, as well as binary and failure time outcomes. 'SEQTaRget' allows to compare both static and dynamic strategies, can be used to estimate observational analogs of intention-to-treat and per-protocol effects, and can adjust for potential selection bias induced by losses-to-follow-up. (Paper to come).

SEQTaRget - Sequential Target Trial Emulation

Implementation of sequential trial emulation for the analysis of observational databases. The ‘SEQTaRget’ software accommodates time-varying treatments and confounders, as well as binary and failure time outcomes. ‘SEQTaRget’ allows to compare both static and dynamic strategies, can be used to estimate observational analogs of intention-to-treat and per-protocol effects, and can adjust for potential selection bias

Installation

You can install the release version of SEQTaRget from CRAN with:

install.packages("SEQTaRget")

You can install the development version of SEQTaRget from GitHub with:

install.packages("remotes")
remotes::install_github("CausalInference/SEQTaRget", subdir = "SEQTaRget")

Setting up your Analysis

SEQuential uses R’s S4 object class system to handle function input/output. From the user side, this amounts to calling a helpful constructor SEQopts and then feeding that into SEQuential. SEQestimate can also take the provided options and return a (very rough) estimated time for analysis.

library(SEQTaRget)
data <- SEQdata

model <- SEQuential(data, id.col = "ID", time.col = "time",
                    eligible.col = "eligible", treatment.col = "tx_init", 
                    outcome.col = "outcome", 
                    time_varying.cols = c("N", "L", "P"),
                    fixed.cols = "sex",
                    method = "ITT",
                    options = SEQopts(km.curves = TRUE))

outcome(model)                       # Returns a list of all outcome models as S3 fastglm objects over the course of bootstrapping
diagnostics <- diagnostics(model)    # Returns a list of diagnostic tables
diagnostics$outcome.unique[[1]]      # Returns a table of unique outcomes
diagnostics$outcome.nonunique[[1]]   # Returns a table of nonunique outcomes

# Plotting
km_curve(model)[[1]]                 # Returns the Survival curve
km_data(model)[[1]]                  # Returns a dataframe of survival data in long-format for other analysis/plotting

Assumptions

This package places several assumptions onto the input data and unexpected results and errors may arise if these are not followed:

1. User provided time.col begins at 0 per unique id.col entries, we also assume that the column contains only integers and continues by 1 for every time step. e.g. (0, 1, 2, 3, …) is allowed and (0, 1, 2, 2.5, …) or (0, 1, 2, 4, 5, …) are not.
   • Provided time.col entries may be out of order as a sort is enforced at the beginning of the function, e.g. (0, 2, 1, 4, 3, …) is valid because it begins at 0 and is continuously increasing by increments of 1, even though it is not ordered.
2. eligible and column names provided to excused.cols are binary (0/1) flag variables (with respect to time.col)

Return

The primary function, SEQuential, returns an S4 object of class SEQoutput with slots:

1. params - the SEQparams object created through the SEQuential process
2. DT - Expanded data table with weighting information
3. outcome - outcome covariates
4. numerator - numerator covariates when weighting
5. denominator - denominator covariates when weighting
6. outcome.model - fastglm model objects
7. hazard - the hazard ratio
8. weight.statistics - weighting information
9. survival.curve - ggplot survival curve
10. survival.data - survival and risk data for all points of followup
11. risk.comparison - data table of comparisons between risks at end of followup bootstrapped
12. risk.data - data table of risks for every level of treatment at each time point
13. time - elapsed time for the SEQuential analysis
14. info - list of diagnostic tables
15. ce.model - Competing event model objects

These can be easily printed to the terminal by calling the object as model (if continuing the example above). While this is the shape of the output object, not all slots will always be filled, e.g. if a user provides hazard = TRUE, then the survival curves, data, and associated risks will return NA.

Imports

• data.table
• doFuture
• doRNG
• future
• future.apply
• ggplot2
• fastglm
• methods
• stringr
• survival
• parallelly
• knitr

Finding More Information and Examples

Further information on utilizing this package or developing it further is available with the SEQTaRget Pages as a part of this repository. If you are unable to find solutions or answers there, please feel free to open a discussion.

Contributing to the package

Community members are welcome to contribute to this package through several different avenues:

• Asking/Answering questions about the package via GitHub Discussions. These can be questions about analysis methods, future planned developments for the package, or requests for clarity on package internals.
• Contributing to GitHub Issues if a bug is found. We have a guided bug report to help us resolve unintended pests quickly.
• Adding content to the package
  • If you intend to add to the package, we would prefer you to branch and then pull-request. This PR will need to:
    1. Pass current unit-tests to ensure nothing is being broken backwards.
    2. Add tests to added portions of code if they are not already covered in existing tests
    3. Pass R-CMD-Check (initiated on PR) with 0-0-0 status