Survival Prediction Modeling: Building and Validating Risk Models

A practical, intermediate course on turning survival models into trustworthy risk predictions. Fitting a Cox model is the easy part; knowing whether its predictions can be believed is the hard part, and it is where most applied survival work quietly goes wrong. This course covers the prognostic modeling discipline end to end: specifying non-linear covariate effects with splines, avoiding leakage through landmarking, shrinking coefficients with penalized regression, and then measuring performance honestly — discrimination with censoring-corrected concordance and time-dependent AUC, calibration with calibration curves and the Brier score, optimism with bootstrap internal validation, and transportability with a genuine external validation on a second cohort. It closes with absolute risk prediction under competing risks and a full prognostic model reported to modern standards, complete with a nomogram and decision-curve analysis. This course is taught in R only — the Rotterdam development cohort ships with R's survival package and has no Python equivalent, and the validation toolchain it depends on (riskRegression, timeROC, cmprsk) exists only in R. The central worked example trains on the Rotterdam breast cancer cohort and validates on the German Breast Cancer Study Group data, including the messy predictor-harmonisation work that real external validation always requires.

  • Level: intermediate
  • Duration: 10 weeks
  • 11 sections
  • Price: $20

Topics: survival analysis, prognostic modeling, risk prediction, external validation, calibration, discrimination, concordance index, IPCW, time-dependent AUC, Brier score, restricted cubic splines, penalized Cox, lasso, optimism correction, bootstrap validation, landmarking, dynamic prediction, competing risks, cumulative incidence, nomogram, decision curve analysis, TRIPOD, R, survival package, riskRegression, timeROC, cmprsk

What's inside

  1. Survival Prediction Modeling
  2. Chapter 1: From Inference to Prediction
  3. Chapter 2: Functional Form — Splines and Non-Linearity
  4. Chapter 3: Dynamic Prediction — Landmarking and Time-Updated Covariates
  5. Chapter 4: Penalized Cox — Ridge, Lasso, and Elastic Net
  6. Chapter 5: Discrimination — The C-Index and Time-Dependent AUC
  7. Chapter 6: Calibration and the Brier Score
  8. Chapter 7: Internal Validation — Optimism and Shrinkage
  9. Chapter 8: External Validation — From Rotterdam to GBSG
  10. Chapter 9: Competing Risks in Risk Prediction
  11. Chapter 10: Capstone — A Complete Prognostic Model

Read a free sample chapter