Kaplan and Meier turned censored follow-ups into the survival curve doctors read today

Medical follow-up is structurally incomplete: some patients outlive the study window, some move away, some withdraw. Their survival times are 'censored' — known only up to a point. In the 1950s analysts either threw them out, which biases survival downward and invents certainty about people never observed, or used actuarial life tables inherited from demography, which chop time into coarse fixed intervals.

Edward Kaplan and Paul Meier published the product-limit estimator in the Journal of the American Statistical Association in 1958. Survival at any moment is the running product of conditional survival probabilities, each estimated from whoever was still at risk right then. No bins, no assumed distribution, no discarding: a fully nonparametric answer that treats partial observation as information rather than noise.

The paper became one of the most cited works in all of statistics — Crossref records more than 76,000 citations — and the curve it defines is printed in virtually every oncology and cardiology trial since, the common grammar for 'how long did they live'. Later tools such as the Cox proportional-hazards model of 1972 sit on top of the frame Kaplan and Meier fixed.

Multiply conditional survival among those still at risk

Discarding censored patients biases survival downward and pretends never-observed people provide facts.

Actuarial life tables depend on arbitrary bin widths; the product-limit estimate follows the data without bins.

Each multiplier uses exactly the patients whose follow-up covers that moment, so every observation feeds the estimate.

Needing no assumption about the shape of survival makes the curve comparable across trials and treatments.

Count subjects only while you can see them: censoring is not missing data to discard but partial information to use — the shrinking risk set is the whole trick.

The Kaplan–Meier plot is now the standard display of survival in medicine, taught to every clinician and demanded by regulators. The Cox model extended it to covariates, log-rank tests compare two such curves, and meta-analyses reconstruct pooled survival from published curves. Their one joint paper outlives nearly everything else published in its field.