Event Prediction

Predicts enrollment and events at the design or analysis stage using specified enrollment and time-to-event models through simulations.

eventPred

eventPred predicts enrollment and event timing in clinical trials. It supports both:

• Design stage prediction using model assumptions and optional priors.
• Analysis stage prediction using observed blinded or unblinded data.

The package provides enrollment modeling, time-to-event modeling, time-to-dropout modeling, simulation-based prediction intervals, and an interactive Shiny app.

Installation

Install the development version from GitHub:

# install.packages("remotes")
remotes::install_github("kaifenglu/eventPred")

Key Features

• Enrollment models: Poisson, time-decay, B-spline, piecewise Poisson.
• Event/dropout models: exponential, Weibull, log-logistic, log-normal, piecewise exponential, model averaging, spline, and Cox.
• By-treatment prediction and optional baseline covariates.
• Prediction intervals based on simulation (nreps).
• Support for fixed follow-up and variable follow-up designs.
• Built-in example datasets: interimData1, interimData2, finalData.

Typical Workflow

1. Summarize observed data with summarizeObserved().
2. Fit enrollment, event, and dropout models with fitEnrollment(), fitEvent(), and fitDropout().
3. Generate predictions with:
   • predictEnrollment() for enrollment only
   • predictEvent() for event timing only
   • getPrediction() for end-to-end enrollment and event prediction

Minimal Example

library(eventPred)

# Event prediction after enrollment completion
set.seed(3000)

pred <- getPrediction(
  df = interimData2,
  to_predict = "event only",
  target_d = 200,
  event_model = "weibull",
  dropout_model = "exponential",
  pilevel = 0.90,
  nreps = 100
)

pred$event_pred$event_pred_summary

Model-Fit + Prediction Example

library(eventPred)

set.seed(2000)

event_fits <- fitEvent(
  df = interimData2,
  event_model = "piecewise exponential",
  piecewiseSurvivalTime = c(0, 140, 352)
)

dropout_fits <- fitDropout(
  df = interimData2,
  dropout_model = "exponential"
)

event_pred <- predictEvent(
  df = interimData2,
  target_d = 200,
  event_fit = event_fits$fit,
  dropout_fit = dropout_fits$fit,
  pilevel = 0.90,
  nreps = 100
)

event_pred$event_pred_summary

Design Stage Example

library(eventPred)

set.seed(1000)

enroll_pred <- predictEnrollment(
  target_n = 300,
  enroll_fit = list(
    model = "piecewise poisson",
    theta = log(26 / 9 * seq(1, 9) / 30.4375),
    vtheta = diag(9) * 1e-8,
    accrualTime = seq(0, 8) * 30.4375
  ),
  pilevel = 0.90,
  nreps = 100
)

enroll_pred$enroll_pred_summary

Run the Shiny App

library(eventPred)
runShinyApp_eventPred()

Time Unit

The package uses days as the primary time unit. To convert rates per month to rates per day, divide by 30.4375.

Documentation

• Website: https://kaifenglu.github.io/eventPred/
• Function reference: https://kaifenglu.github.io/eventPred/reference/
• Issues: https://github.com/kaifenglu/eventPred/issues

Citation

If you use eventPred in analysis or reporting, please cite relevant methodology references included in the package documentation.