drugclaw/drugclaw/survival-analysis-tools

Survival Analysis Tools

Use this skill when the user needs time-to-event analysis with censoring-aware summaries.

Typical triggers:

• Kaplan-Meier curves or survival probability tables
• log-rank comparison between treatment arms
• Cox proportional hazards regression with hazard ratios
• time-to-event or progression-free survival analysis
• censored cohort summaries for translational or clinical research

Environment Check

which python3 || true
python3 - <<'PY'
mods = ["numpy", "pandas", "statsmodels", "matplotlib"]
for name in mods:
    try:
        __import__(name)
        print(f"{name}: ok")
    except Exception as exc:
        print(f"{name}: missing ({exc})")
try:
    import sksurv
    print("sksurv: optional-ok")
except Exception as exc:
    print(f"sksurv: optional-missing ({exc})")
PY

The bundled template runs on the stable statsmodels baseline. Advanced machine-learning survival models from scikit-survival remain optional and should only be claimed when the environment actually has them.

Bundled Asset

• templates/survival_analysis.py

Preferred Workflow

1. Confirm the time and event coding first.
2. Generate group-level Kaplan-Meier summaries before fitting adjusted models.
3. Add Cox covariates only after checking the columns and coding logic.
4. Export both tables and a survival plot.
5. Treat hazard ratios as model-based associations, not automatic causal effects.

Kaplan-Meier And Cox Baseline

python3 templates/survival_analysis.py \
  --input survival/nsclc.csv \
  --time-column pfs_days \
  --event-column progressed \
  --group-column arm \
  --covariate age \
  --covariate stage_numeric \
  --covariate biomarker_score \
  --plot-output survival/nsclc_km.png \
  --km-output survival/nsclc_km.csv \
  --cox-output survival/nsclc_cox.csv \
  --summary survival/nsclc_summary.json

Use this for:

• group-level median survival summaries
• Kaplan-Meier plots
• log-rank p-values when a group column is present
• Cox proportional hazards coefficients and hazard ratios

Boundary

The bundled baseline does not provide random survival forests, gradient-boosted survival models, or integrated Brier score pipelines out of the box. If the user explicitly needs those, confirm that scikit-survival is available first.

Related Skills

For general hypothesis tests or non-survival regression, activate stat-modeling-tools. For figures beyond the bundled KM plot, activate scientific-visualization-tools. For study-design or endpoint-planning support, activate clinical-research-tools.