Predicting Binary Outcomes Using Trajectory Classes

Nami Maruyama, Fumiaki Takahashi, Hajime Uno, Masahiro Takeuchi

Research output: Contribution to journalArticlepeer-review

Abstract

In longitudinal data, the interest often lies in the repeatedly measured variable itself. However, in some situations, the changing pattern of the variable over time may contain information about a separate outcome variable. In such a situation, longitudinal data provide the opportunity to develop predictive models of subsequent observations of the separate outcome variable given current data for an individual. In particular, longitudinally changing patterns of repeated measurements of a variable, or trajectories, measured up to time t can be used to predict an outcome measure or event that occurs after time t. We propose a predictive model based on latent classes of trajectories, which is fitted using the expectation-maximization (EM) algorithm, and show how to get model estimates with other covariates in the model. Applications of our methodology are demonstrated through an example of a smoking cessation trial.

Original languageEnglish
Pages (from-to)284-292
Number of pages9
JournalStatistics in Biopharmaceutical Research
Volume4
Issue number3
DOIs
Publication statusPublished - 2012 Jul 1

Keywords

  • EM algorithm
  • Longitudinal data
  • Mixture distribution, Prediction
  • ROC curve
  • Trajectory

ASJC Scopus subject areas

  • Statistics and Probability
  • Pharmaceutical Science

Fingerprint Dive into the research topics of 'Predicting Binary Outcomes Using Trajectory Classes'. Together they form a unique fingerprint.

Cite this