A joint model for dynamic prediction in uveitis

Mia Klinten Grand, Koenraad Arndt Vermeer, Tom Missotten, Hein Putter

Research output: Contribution to journalArticleResearchpeer-review

Abstract

Uveitis is characterised as a recurrent inflammation of the eye and an ongoing inflammation can have severe impact on the visual acuity of the patient. The Rotterdam Eye Hospital has been collecting data on every uveitis patient visiting the hospital since 2000. We propose a joint model for the inflammation and visual acuity with the purpose of making dynamic predictions. Dynamic prediction models allow predictions to be updated during the follow-up of the patient based on the patient's disease history. The joint model consists of a submodel for the inflammation, the event history outcome, and one for the visual acuity, the longitudinal outcome. The inflammation process is described with a two-state reversible multistate model, where transition times are interval censored. Correlated log-normal frailties are included in the multistate model to account for the within eye and within patient correlation. A linear mixed model is used for the visual acuity. The joint model is fitted in a two-stage procedure and we illustrate how the model can be used to make dynamic predictions. The performance of the method was investigated in a simulation study. The novelty of the proposed model includes the extension to a multistate outcome, whereas, previously, the standard has been to consider survival or competing risk outcomes. Furthermore, it is usually the case that the longitudinal outcome affects the event history outcome, but in this model, the relation is reversed.

Original languageEnglish
Pages (from-to)1802-1816
Number of pages15
JournalStatistics in Medicine
Volume38
Issue number10
DOIs
Publication statusPublished - 10 May 2019

Keywords

  • clusters
  • dynamic prediction
  • interval censoring
  • joint model
  • multistate model

Fingerprint Dive into the research topics of 'A joint model for dynamic prediction in uveitis'. Together they form a unique fingerprint.

Cite this