Survlytics
Production-grade survival analysis following NICE TSD 14 & TSD 21. Parametric, spline, cure, piecewise, and mixture models — all in one platform.
Full Model Suite
Exponential, Weibull, Gamma, Gompertz, Log-logistic, Log-normal, Generalized Gamma & F, splines, cure, piecewise, and mixture models.
NICE TSD 14 & 21
Compliant with NICE Technical Support Documents 14 and 21. Covers all recommended parametric and flexible models.
Interactive Diagnostics
Kaplan-Meier curves, hazard plots, Schoenfeld residuals, Cox-Snell QQ plots, and model-averaged extrapolations.
Submission Reports
One-click Quarto-based reports in PDF, Word, and Excel. NICE HTA submission-ready formatting.
Role-Based Access
Admin, Analyst, and Viewer roles secured via AWS Cognito. Full session audit trail.
Cloud-Native AWS
Deployed on ECS Fargate with RDS, S3, ALB, and auto-scaling. HTTPS by default.
Sign in to Survlytics
Welcome back!
Getting Started
Setup Data — Go to Data Setup. Upload or select a dataset, specify arm names and time units.
Explore — Visit Data Explore for Kaplan-Meier curves, survival summary, and proportional hazards diagnostics.
Configure — In Analysis Setup, choose model types, distributions, knots, and cutpoints.
Run Analysis — Click Run Analysis. Results appear in the Results tab with interactive GOF comparison.
Report — Go to Reports to generate a NICE-submission-ready PDF or Word report.
Grambsch-Therneau test. Significant p-value (< 0.05) indicates violation of PH — consider time-varying HR or spline models.
Configure the model types below, then click Run Analysis.
Quick presets
Arm 1
Only patients at risk beyond the landmark are included. Time is re-zeroed at the landmark. Recommended when early follow-up has administrative issues or a delayed treatment effect.
Click rows to overlay curves on the extrapolation plot. ΔAIC ≤ 2 = strong support (green), ≤ 10 = some support (amber), > 10 = little support (grey).
Generates an Excel workbook for use in Markov models. One sheet per arm — rows = time cycles, columns = S(t) per model plus KM, transition probabilities (1 − S(t+Δ)/S(t)), and cumulative hazard.
Akaike and BIC weights for all fitted models. Higher weight = stronger statistical support. Use 'Model average' checkbox on the plot to overlay the weighted-average curve.
Document your model selection rationale for the NICE submission report. This text is included verbatim in the generated report.
Upload external S(t) data (e.g. registry, real-world evidence) to overlay
on the extrapolation plot. CSV must contain columns
Time
and
Survival
(0-1 scale).
Include sections
Configure options and click Generate Report.
User Guide
Quick Start
- Log in with your credentials.
- Go to Data Setup — upload or select sample data, set arm names and time units, then click Confirm Data .
- Visit Data Explore to review Kaplan-Meier curves and check the proportional hazards assumption.
- Go to Analysis Setup , select model types and distributions, then click Run Analysis .
- Review the goodness-of-fit table and survival extrapolation plots in the Results tab.
- Generate a NICE-ready report in the Reports tab.
Data Setup
Survlytics accepts CSV or RDS files with at minimum a
Time
(positive numeric) and
Event
(0/1 binary) column. If your columns have different names, use the column mapping dropdowns after upload. Time units can be set independently per arm and are stored as metadata.
Data Explore
The Kaplan-Meier panel shows non-parametric survival estimates with optional confidence intervals and censoring markers.
The
Proportional Hazards Test
applies the Grambsch-Therneau test (
cox.zph
). A significant p-value (< 0.05) suggests the PH assumption is violated — consider spline or piecewise models.
Analysis Setup
Model types available:
- Parametric: Exponential, Weibull, Gamma, Gompertz, Log-logistic, Log-normal, Generalized gamma, Generalized F. Independent (per arm) and dependent (treatment as covariate) modes.
- Spline: Royston-Parmar flexible parametric models on hazard, odds, or normal scales with 1–3 knots (NICE TSD 14 §5.3).
- Piecewise: Piecewise exponential via Poisson GLM. Cutpoints by median, RMST, quartiles, or user-specified values.
- Cure: Mixture and non-mixture cure fraction models (NICE TSD 21). Supports logistic, log-log, probit, and identity link functions.
- PMM: Parametric mixture of two distributions with optimised or grid-search mixing weights. CPU-intensive; runs asynchronously.
- Background Mortality: Additive hazard adjustment using ONS / HMD life tables (NICE TSD 21 §4.4).
- Time-Varying HR: Spline-based or piecewise treatment effects to model non-proportional hazards.
NICE TSD 14 & 21 Reference
NICE TSD 14 (Latimer 2011) recommends fitting a range of parametric models and selecting based on AIC/BIC, visual fit against observed data, and clinical plausibility.
NICE TSD 21 (Rutherford et al. 2020) extends this to flexible parametric (spline) models, cure fraction models, and background mortality adjustment.
This platform implements all model classes from both TSDs. Use the NICE TSD 14 Defaults preset in Analysis Setup for the standard submission model set.
Frequently Asked Questions
- Why does PMM fitting take a long time?
-
PMM iterates over many distribution combinations and mixing weight values. The fitting runs asynchronously in a background worker (
future.callr) so the app remains responsive. Check the Results tab once complete. - Can I save an analysis and return to it?
- Yes — Analyst and Admin users can save analysis configurations via the save button in Analysis Setup. Results are persisted to S3 and linked to your session.
- Why is my parametric model not converging?
- Some distributions (e.g. Generalized gamma, Generalized F) may fail to converge with small samples or high censoring. The app will warn and skip non-convergent models automatically.