Course contentsShow
← Data Science
Lesson 814 of 2,14521. Survival AnalysisFree lesson

Standard Errors and Confidence Intervals

Greenwood's formula for computing standard errors and constructing confidence bands for survival curves.

Standard Errors and Confidence Intervals

What You'll Learn

You'll learn how to quantify the uncertainty in Kaplan-Meier survival estimates using Greenwood's formula and construct confidence bands around survival curves.

Why Uncertainty Matters in Survival Curves

When you compute a Kaplan-Meier survival estimate, you're working with a sample—not the entire population. Just like any statistic, your survival probability S(t) at any time point has uncertainty. The question is: how confident are we in this estimate?

Greenwood's Formula

Greenwood's formula gives us the standard error (SE) of the Kaplan-Meier estimator at any time t. It accounts for both the number of events and the size of the risk set at each event time.

The formula looks like this:

SE[S(t)]² = [S(t)]² × Σ (dᵢ / (nᵢ × (nᵢ - dᵢ)))

Where:

  • S(t) is the survival probability at time t
  • dᵢ is the number of events at time i
  • nᵢ is the risk set size at time i
  • The sum runs over all event times up to t

Constructing Confidence Intervals

Once you have the standard error, you can build confidence intervals (typically 95%) around your survival curve:

S(t) ± 1.96 × SE[S(t)]

Think of these bands like guardrails on a highway—they show the range where the true population survival probability likely falls.

Why This Matters

Confidence bands tell you when to trust your survival curve. Wide bands (often at later times when few patients remain) signal high uncertainty. Narrow bands indicate reliable estimates. When comparing survival curves between groups, overlapping confidence intervals suggest differences may not be meaningful.

Key Takeaway: Greenwood's formula quantifies uncertainty in Kaplan-Meier estimates by computing standard errors from the risk set and event counts, allowing you to construct confidence intervals that reflect estimation reliability.