Lecture 4: Survival Analysis Case Study (Kaplan-Meier, Log-Rank, Cox in R)
Why It Matters
Understanding how to correctly apply Kaplan‑Meier, log‑rank, and Cox models enables clinicians and researchers to quantify treatment effects and prioritize interventions that truly improve breast‑cancer survival.
Key Takeaways
- •Kaplan‑Meier and log‑rank test reveal stage‑dependent survival differences.
- •Cox model provides hazard ratios, e.g., stage IV HR ≈ 4.
- •Hormone therapy improves survival; radiotherapy shows no significant effect.
- •Surgery type (mastectomy vs conserving) influences long‑term survival.
- •Proportional hazards assumption required stratification for chemotherapy and hormone therapy.
Summary
The lecture walks through a published breast‑cancer survival study, illustrating how non‑parametric (Kaplan‑Meier, log‑rank) and semi‑parametric (Cox proportional hazards) techniques are implemented in R to handle censored time‑to‑event data. It explains why the authors chose these methods: Kaplan‑Meier for visualizing survival curves without distributional assumptions, and the Cox model to assess multiple covariates simultaneously and produce hazard ratios. Key findings include highly significant differences in survival across tumor stages (p < 0.00001) and surgery types (p = 0.004), modest benefits from chemotherapy (p = 0.04) and hormone therapy (significant), while radiotherapy showed no effect. The Cox analysis reported a stage‑IV hazard ratio of 3.98, indicating nearly four‑fold higher death risk compared with stage‑I, and highlighted the need to stratify chemotherapy and hormone therapy to satisfy the proportional‑hazards assumption. Notable examples cited are the Kaplan‑Meier curves where hormone‑treated patients maintain higher survival probabilities between months 30‑60, and the mastectomy group’s superior long‑term curve versus breast‑conserving surgery. The presenter also outlines the standard manuscript structure for survival studies, from introduction through methods, results, discussion, and abstract. The case study underscores that integrating non‑parametric visualization with Cox multivariate modeling yields robust, clinically actionable insights, guiding treatment decisions and informing future research designs.
Comments
Want to join the conversation?
Loading comments...