HybridCR: a browser-based tool for maximum-likelihood inference and study design in competing-risks models under hybrid Type-I censoring (v1.0.0)
Description
HybridCR is a single-file, zero-installation web application (about 110 kB of HTML, CSS and vanilla JavaScript with inline SVG, and no third-party libraries) for maximum-likelihood inference, diagnostics and study design in two-cause competing-risks models observed under the hybrid Type-I censoring scheme tau_n = min(X_(r), T0). It simulates or imports (CSV) time-to-event data with cause and status, fits cause-specific exponential or Weibull intensities by Newton-Raphson, and reports parameter estimates with asymptotic and bootstrap confidence intervals, cumulative incidence functions compared against the nonparametric Aalen-Johansen estimator, and Cox-Snell goodness-of-fit diagnostics. Six interactive modules make visible the quantities that govern censored inference: the operational Fisher information and the efficiency bound; the quadratic convergence of the estimator; sample-size planning for a target precision; the systematic discrepancy incurred when dependent risks are fitted under independence (Clayton copula); and the exact non-central chi-square power of the Wald test under a local alternative. The tool runs offline in any modern browser and is fully client-side; no data leave the device. This record archives version v1.0.0; the development repository (version history and automated tests) is on GitHub, and the software is an interactive companion to the authors’ methodological articles (see Related links). Software license: MIT.
Files
Steps to reproduce
The software is a single self-contained HTML file; no installation or build is required. Run the tool. Open HybridCR.html in any modern web browser (verified in Chromium 141, the Chrome/Edge engine, on desktop and on a mobile viewport). All computation is client-side and offline. Use the six modules, the four preset scenarios, or upload your own CSV with three columns: time (> 0); cause (1 or 2; any value other than 2 is read as 1); status (1 = event, 0 = right-censored; optional, defaults to 1). Comma, semicolon or tab separators are accepted and a header row is auto-detected. Verify the numerical core. With Node.js (>= 18) installed, from the archive root run: node run_tests.cjs . The harness extracts the numerical functions directly from HybridCR.html (no duplicated source) and checks closed-form reference values and edge cases. Expected output: 20 passed, 0 failed. Reference values are stored in reference_values.json. Cross-validate (optional). Without R: with Python and NumPy, run: python3 compare_python.py . It re-implements the simulation, the cause-specific MLE and the Aalen-Johansen estimator from scratch and checks them against the fitted parametric cumulative incidence (agreement to Monte-Carlo error). With R: with the packages survival, cmprsk and etm installed, run: Rscript compare_R.R for the same comparison against those packages. The software is licensed under MIT (see LICENSE); this deposit record is CC BY 4.0.
Institutions
- National University of UzbekistanTashkent, Tashkent