GitHub

CRAN Stats Paper

monitOS helps clinical teams define positivity thresholds for monitoring overall survival (OS) in pivotal trials for indolent cancers.

It turns a small set of trial assumptions into a table of operating characteristics you can use at interim and final analyses:

  • the unacceptable detrimental effect on OS you want to rule out (hr_null)
  • the plausible beneficial effect on OS you still consider realistic (hr_alt)
  • the deaths expected at each analysis (events)

The framework is described in the monitOS paper.

Install and launch

Install from CRAN:

install.packages("monitOS")

Load the package:

library(monitOS)

Use the app

The fastest way to explore the method is through the interactive app.

  • 🌐 Try it online (Shinylive) — runs entirely in your browser via WebAssembly, no installation required.
  • 💻 Run locally with R:
monitOS::run_app()

What the inputs mean

monitOS is driven by a few clinically meaningful choices.

Concept Argument What it means
Unacceptable OS detriment hr_null The smallest OS HR that would already be too harmful to accept
Plausible OS benefit hr_alt The OS HR that still represents a realistic beneficial effect
Planned deaths events Cumulative deaths expected at each analysis
Final-analysis false positive rate falsepos One-sided error rate tolerated at the final analysis
Primary-analysis power power_int Required power at the primary analysis when the true OS HR equals hr_alt
Randomization ratio rand_ratio Experimental-to-control allocation ratio
Incremental-benefit scenario hr_marg_benefit Optional weaker-benefit scenario used for sensitivity analysis

Key convention: OS HR < 1 indicates benefit.

Quick start

This minimal example creates positivity thresholds for three planned analyses:

library(monitOS)
plan <- bounds(
  events = c(60, 110, 178),
  power_int = 0.90,
  falsepos = 0.025,
  hr_null = 1.30,
  hr_alt = 0.80
)
plan$summary

What to look for:

  • OS HR threshold for positivity tells you the largest observed OS HR that still provides enough reassurance at each analysis.
  • One-sided false positive error rate shows the tolerated error at each stage.
  • Level of 2-sided CI needed to rule out delta null tells you the confidence level needed to exclude the unacceptable detriment.

Example 1: Standard 1:1 design

library(monitOS)
fl_trial <- bounds(
  events = c(50, 80, 110, 140, 175),
  power_int = 0.90,
  falsepos = 0.025,
  hr_null = 1.30,
  hr_alt = 0.80,
  rand_ratio = 1
)
fl_trial$summary

Interpretation: as evidence accumulates, the positivity threshold typically becomes more stringent. Early analyses may tolerate an observed OS HR slightly above 1, while the final analysis usually requires a threshold below 1.

Example 2: Unequal randomization with sensitivity analysis

library(monitOS)
cll_trial <- bounds(
  events = c(60, 95, 130, 165, 200),
  power_int = 0.90,
  falsepos = 0.025,
  hr_null = 1.25,
  hr_alt = 0.80,
  rand_ratio = 2,
  hr_marg_benefit = 0.95
)
cll_trial$summary

This adds an incremental-benefit column showing how often the monitoring thresholds would still be met if the true OS effect were more modest than the main alternative.

Need help?

?bounds
?run_app

Read the original on github.com ↗