Lectures and seminars Biostatistics seminar: Disease registers and multistate models

Title 

Disease registers and multistate models

Abstract

A disease register for a population has two main uses:
1.    Look-up of disease status (diabetes type, cancer type) for persons at a given date. Persons will typically emerge from some study cohort, such as a population survey or a clinical study.
2.    Demographic analyses of population incidence and mortality rates, prevalences. And of course derivatives of these basic demographic measures.
Demographic analyses requires knowledge not only of the persons with disease but also of the population covered by the register, loosely speaking, the persons without disease. This comes in one of two forms:
(a) population size (number or risk time) classified by sex, age, date and possibly other variables available in the register. This will be tabular data, such as that available from Statistikbanken at DST. The corresponding data for diseased persons is tabulated from the register.
(b) individual level follow-up data for all persons in the population — basically knowledge of entry (birth or immigration) and exit (death or emigration). This has recently be come available as the LifeLine register at DST.

Demographic research based on registers almost inevitably leads to consideration of multistate models: register data is a collection of dates of specific events over persons’ life course.
Specification of states and transitions between them helps clarify quantities of interest in register-based studies, at 3 possible levels:

1. Occurrence rates — the scale of observed register data, measured in time⁻¹ (events per person-time)
2. State probabilities (e.g. survival function), this is the integral of rates w.r.t. time — it requires an origin (such as date of diagnosis) — dimensionless, measured in time⁰
3. Sojourn times (time spent in a state) - integral of state probabilities w.r.t. time, measured in time¹
    

The Epi package for R provides tools for representation and manipulation of multistate data on multiple time scales, facilitating fitting models based on (event / date) data from registers.

Proper specification of states and transitions are required to pose and answer questions arising when analyzing follow-up data from registers. The Lexis machinery in the Epi package for R is a tool that facilitates this. The difference to the representation of follow-up data in the mstate package will be discussed.
For a more comprehensive overview see https://bendixcarstensen.com/PMM/

Speaker bio 

Bendix Carstensen holds a master’s degree in mathematical statistics from University of Copenhagen (1983).

The last 26 year he has been senior statistician at Steno Diabetes Center Copenhagen, and before that spent 11 years as statistician at the Danish Cancer Registry.

He has been contributing to establishing a Danish diabetes register, and has conducted demographic / epidemiological studies on diabetes, including studies on cancer occurrence among diabetes patients.

He has extensive teaching experience in statistical methods in epidemiology and is the author of the book “Epidemiology with R”, Oxford University Press, 2021. 

He is the maintainer of the R-package “Epi” (since 2008), and has a primitive and slightly messy website, https://bendixcarstensen.com/

Host 

Paul Dickman, MEB