Immortal time

Person-time counted towards a treatment nobody was on yet — how to see it, measure it, and fix it two ways.

Taught by

Michael Crowther, PhD

CEO & Director of Statistical Methodology

Formerly Associate Professor of Biostatistics at the University of Leicester, and the author of the merlin framework.

Research survival analysis · joint models · multi-state models · statistical software

PhD in Medical Statistics, University of Leicester, 2014

Register your interest

What you will be able to do

• Recognise immortal time in a cohort design, and name the moment at which the exposure stops being a fact about the future
• Measure the error with arithmetic alone — allocate person-time to the state it was lived in, and say what share of a treated denominator is immortal
• Fit both corrections: an episode-split time-varying exposure, and a landmark analysis
• Say what each fix costs — discarded events, and the attenuation a landmark buys when the effect is real
• Read a methods paragraph for its time zero, and say whether a published hazard ratio can be believed
• Say which of the two standard definitions a reported median follow-up used, and why it matters
• Tell an immortal-time problem apart from a competing-risks one and from a confounding one, and say where each belongs

Before you start

• Take first: Survival analysis

Curriculum

22 steps: 14 lessons, 3 practicals and 5 quizzes.

Provisional — this course is still being written, and the outline will change.

Module 0 · Does this apply to you

• 01ReadThe error, in one picture8 min
• 02ReadThree designs that produce it8 min
• 03QuizSelf-check · is this your study5 min
• 04ReadHow the marking works4 min

Module 1 · The arithmetic

• 05ReadPerson-time, and where it gets allocated8 min
• 06ReadThe immortal window, patient by patient7 min
• 07ReadCounting it two ways7 min
• 08PracticalPractical 1 · The arithmetic of the error24 min
• 09QuizWhat moved, and what did not5 min

Module 2 · The two fixes

• 10ReadSplit the record10 min
• 11ReadLandmarking10 min
• 12ReadWhat landmarking costs you12 min
• 13PracticalPractical 2 · Three models, one dataset30 min
• 14QuizWhich fix, and what you paid for it6 min

Module 3 · The audit

• 15ReadReading a methods paragraph for time zero12 min
• 16ReadMedian follow-up, the number nobody defines12 min
• 17PracticalPractical 3 · Audit a published analysis26 min
• 18QuizThe audit checklist6 min

Module 4 · The rest

• 19ReadThe other way a survival number can be impossible10 min
• 20Read1 − KM twice, and a total over 100%10 min
• 21ReadWhat this course withholds, and where each answer lives8 min
• 22QuizSix paragraphs, one flaw each6 min