A synthetic dataset containing 27 daily observations for 600 subjects across four ordinal clinical outcome states ("Home", "Hospital", "Vent/ARDS", "Dead").
Format
A data frame with 16,200 rows and 3 variables:
- id
Subject identifier (integer 1-600)
- step
Discrete step / day horizon (integer 1-27)
- reals
Observed clinical state (ordered factor with levels "Home" < "Hospital" < "Vent/ARDS" < "Dead")
Details
The dataset was generated using a first-order Markov simulation with
seed 2026, initial probabilities (0, 0.85, 0.15, 0), and transition matrix
rbind(c(.97, .03, 0, 0), c(.10, .82, .06, .02), c(.01, .08, .84, .07), c(0, 0, 0, 1))
where "Dead" is an absorbing state.
This dataset draws inspiration from Frank Harrell's work on longitudinal ordinal Markov and state-occupancy models and VIOLET 2 trial examples: https://hbiostat.org/r/hmisc/markov/sim#estimating-efficiency-from-the-study. It is independently simulated and is NOT Frank Harrell's original trial data, fitted model, or an endorsed implementation.