Skills scikit-survival
📊

scikit-survival

Safe ⚡ Contains scripts⚙️ External commands🌐 Network access

Analyze survival data with scikit-survival

Also available from: davila7

Survival analysis handles data where events may not have occurred for all subjects. This skill provides Python tools for time-to-event modeling with censored data using Cox models, Random Survival Forests, SVMs, and specialized evaluation metrics like concordance index.

Supports: Claude Codex Code(CC)
📊 71 Adequate
1

Download the skill ZIP

2

Upload in Claude

Go to Settings → Capabilities → Skills → Upload skill

3

Toggle on and start using

Test it

Using "scikit-survival". Build a survival model for the veterans lung cancer dataset and evaluate performance

Expected outcome:

  • Loaded veterans_lung_cancer dataset with 137 patients
  • Censoring rate: 8.0% (11 events, 126 censored)
  • Fitted CoxPHSurvivalAnalysis with concordance_index_ipcw = 0.73
  • Top risk factors: Karnofsky score (HR=0.96), age (HR=1.02)
  • Time-dependent AUC at 180 days: 0.81

Using "scikit-survival". Compare Random Survival Forest and Cox model on GBSG2 breast cancer data

Expected outcome:

  • Loaded GBSG2 dataset with 2238 patients, 1548 events
  • RandomSurvivalForest C-index: 0.68 (5-fold CV)
  • CoxPHSurvivalAnalysis C-index: 0.66 (5-fold CV)
  • RSF selected 12/7 features via permutation importance
  • Recommendation: RSF provides slightly better ranking ability

Security Audit

Safe
v4 • 1/17/2026

All 277 static findings are FALSE POSITIVES. This skill contains only markdown documentation for the legitimate scikit-survival Python library. The 'Ruby/shell backtick execution' detections are markdown code fences (```python) for Python syntax highlighting. No executable code, scripts, or malicious patterns exist. The 'C2 keywords' and 'weak cryptographic algorithm' detections are false positives caused by statistical/medical terminology being misidentified by the pattern scanner.

8
Files scanned
3,958
Lines analyzed
3
findings
4
Total audits

Risk Factors

⚡ Contains scripts (2)
SKILL.md:372-380 SKILL.md:383-387
⚙️ External commands (3)
SKILL.md:39-40 references/competing-risks.md:46-70 references/data-handling.md:11-20
🌐 Network access (1)
SKILL.md:357-360
Audited by: claude View Audit History →

Quality Score

45
Architecture
100
Maintainability
87
Content
21
Community
100
Security
91
Spec Compliance

What You Can Build

Patient survival prediction

Analyze clinical trial data to predict patient survival probabilities and identify risk factors for adverse outcomes.

Time-to-event machine learning

Build and compare multiple survival models including Cox, Random Survival Forest, and SVM for predictive maintenance or customer churn.

Disease progression modeling

Study time-to-event data with competing risks such as death from different causes in disease progression studies.

Try These Prompts

Basic survival analysis 
Use scikit-survival to load the breast cancer dataset, split it into train and test sets, fit a Cox Proportional Hazards model, and evaluate using Uno's concordance index.
Model comparison 
Compare CoxPHSurvivalAnalysis, RandomSurvivalForest, GradientBoostingSurvivalAnalysis, and FastSurvivalSVM on the GBSG2 dataset using cross-validation with concordance index scoring.
Feature selection 
Use CoxnetSurvivalAnalysis with elastic net regularization to perform feature selection on high-dimensional survival data, then identify which features were selected.
Competing risks 
Demonstrate competing risks analysis using cumulative_incidence_competing_risks. Show how to estimate cumulative incidence for different event types and compare between treatment groups.

Best Practices

  • Always standardize features for SVMs and regularized Cox models before fitting
  • Use Uno's C-index (concordance_index_ipcw) instead of Harrell's when censoring exceeds 40%
  • Report multiple evaluation metrics including C-index, integrated Brier score, and time-dependent AUC

Avoid

  • Using Kaplan-Meier estimator when competing risks are present (use cumulative incidence instead)
  • Using built-in feature importance for Random Survival Forests (use permutation importance)
  • Not checking proportional hazards assumption for Cox models before interpretation

Frequently Asked Questions

What is the difference between Harrell's and Uno's C-index?
Harrell's C-index works well with low censoring (<40%). Uno's uses IPCW weighting and remains unbiased with high censoring.
How do I handle competing risks?
Use cumulative_incidence_competing_risks for estimation or fit separate cause-specific Cox models for each event type.
Which model should I choose for high-dimensional data?
Use CoxnetSurvivalAnalysis with elastic net regularization for feature selection in high-dimensional settings.
How do I evaluate model calibration?
Use integrated_brier_score to assess both discrimination and calibration. Lower scores indicate better-calibrated predictions.
Can I use scikit-survival with scikit-learn pipelines?
Yes, scikit-survival models are compatible with sklearn Pipeline, GridSearchCV, and cross_val_score.
What preprocessing is required for survival data?
Convert data to Surv object, handle missing values, encode categoricals, and standardize features for SVMs.

Developer Details

Author

K-Dense-AI

License

GPL-3.0 license

Repository

https://github.com/K-Dense-AI/claude-scientific-skills/tree/main/scientific-skills/scikit-survival

Ref

main

File structure

📁 references/

📄 competing-risks.md

📄 cox-models.md

📄 data-handling.md

📄 ensemble-models.md

📄 evaluation-metrics.md

📄 svm-models.md

📄 SKILL.md