MerQuR+Statistics Glossary

Statistics Glossary → Turnbull NPMLE (interval-censored data)

Turnbull NPMLE (interval-censored data)concept

The nonparametric maximum likelihood estimate of the survival function for data in which the event time is not known exactly, only that it fell in an interval between two examinations.

In interval-censored data, each person's event is known to have occurred in an interval (L, R]; for example, an infection began somewhere between the last negative and the first positive test. Kaplan-Meier relies on exact event times and cannot be used directly here. Turnbull's method shows that probability mass can only be placed on 'innermost' (Turnbull) intervals and estimates these masses with a self-consistency (EM-type) algorithm. When all intervals shrink to single points, the result equals the Kaplan-Meier estimator.

The estimate is determined only outside the Turnbull intervals; how the mass is spread within an interval cannot be determined from the data, so the curve is not unique there and is usually drawn with steps or shaded regions. Treating interval midpoints as event times and applying Kaplan-Meier can give biased results, especially when the examination intervals are wide. Generalized log-rank tests are used to compare groups.

An everyday example

A dental clinic studies when fillings fail, but patients come for check-ups only every 6 months. If a filling was intact at the 12-month check-up and failed at the 18-month one, the failure lies in (12, 18]. Turnbull's NPMLE combines intervals like these to estimate, for example, that 20% of fillings fail within 2 years, while leaving exactly what happened between months 12 and 18 undetermined.

In MerQur

In Interval-Censored Survival, 'Additional output: Turnbull intervals and masses' tabulates the innermost intervals and their NPMLE probability masses; when a 'Group' is selected, a separate Turnbull curve is drawn for each level.

Analyses where this concept appears

Interval-Censored Survival

Related concepts

Censoring (survival analysis)Kaplan-Meier estimator (survival curve)