Survival analysis

Survival analysis connects molecular features to clinical outcomes — Kaplan-Meier curves, log-rank testing, multivariate Cox regression, and longitudinal molecular trajectories.

Uses the lifelines library for Kaplan-Meier estimation, log-rank tests, and Cox proportional hazards regression.

Prerequisites

Clinical CSV with patient_id, time_to_event_days, and event_observed columns
Optional feature CSV with molecular variables for stratification and Cox covariates
Optional longitudinal CSV for treatment timepoint trajectories

Launching the pipeline

Open the study Survival page, select your clinical dataset (and optional feature/longitudinal datasets), set the stratification field and Cox covariates, and click Run survival analysis.

Key parameters

Parameter	Default	Purpose
`clinical_path`	—	Patient survival table
`feature_path`	—	Molecular features for stratification and Cox models
`longitudinal_path`	—	Timepoint feature trajectories
`stratify_by`	—	Column for Kaplan-Meier group stratification
`cox_covariates`	[]	Columns included in multivariate Cox regression

Outputs

Output	Description
`kaplan_meier`	Survival curve data per stratification group
`logrank`	Log-rank test statistic and p-value
`cox_regression`	Hazard ratios, confidence intervals, and p-values per covariate
`longitudinal`	Feature trajectory summaries across timepoints

The frontend Survival page renders patient/event/group counts, the log-rank test statistic and p-value, and the three visualizations described below.

Visualizations

Kaplan-Meier survival curves

An interactive step-function chart with one line per stratification group, sharing a common time axis. Hovering shows the day, survival probability, and current at-risk/event counts for each group at that point. Each group's median survival (or "not reached" if the curve never crosses 50%) is also listed below the chart.

Cox regression forest plot

One row per covariate from the multivariate Cox model, sorted by p-value (strongest association first). Each row plots the hazard ratio as a point with its 95% confidence interval as a whisker, on a shared log-scale axis with a reference line at HR = 1 (no effect) — the standard forest-plot layout. HR > 1 means higher hazard (worse survival); HR < 1 is protective. Rows with p < 0.05 are marked. With small cohorts, Cox regression can produce very wide confidence intervals for some covariates (quasi-complete separation). The chart's shared axis makes that instability visible rather than hiding it.

Longitudinal feature summary

When a longitudinal dataset is supplied, a table lists each tracked feature's cohort mean value and the average per-patient change from its first to last recorded timepoint, sorted by magnitude of change so the features that moved the most lead the list.

Typical use cases

Compare overall survival between treatment arms
Stratify survival by TMB, immune phenotype, or gene signature score
Test whether molecular features independently predict outcome in multivariate Cox models
Track CD8 score or other biomarkers across baseline and on-treatment timepoints