Estimating net transition probabilities from cross-sectional data wit application to risk factors in chronic disease modeling

J. van de Kassteele, R.T. Hoogenveen, P.M. Engelfriet, P.H. van Baal, H.C. Boshuizen

Research output: Contribution to journalArticleAcademicpeer-review

34 Citations (Scopus)

Abstract

A problem occurring in chronic disease modeling is the estimation of transition probabilities of moving from one state of a categorical risk factor to another. Transitions could be obtained from a cohort study, but often such data may not be available. However, under the assumption that transitions remain stable over time, age specific cross-sectional prevalence data could be used instead. Problems that then arise are parameter identifiability and the fact that age dependent cross-sectional data are often noisy or are given in age intervals. In this paper we propose a method to estimate so-called net annual transition probabilities from cross-sectional data, including their uncertainties. Net transitions only describe the net inflow or outflow into a certain risk factor state at a certain age. Our approach consists of two steps: first, smooth the data using multinomial P-splines, second, from these data estimate net transition probabilities. This second step can be formulated as a transportation problem, which is solved using the simplex algorithm from linear programming theory. A sensible specification of the cost matrix is crucial to get meaningful results. Uncertainties are assessed by parametric bootstrapping. We illustrate our method using data on body mass index. We conclude that this method provides a flexible way of estimating net transitions and that the use of net transitions has implications for model dynamics, for example when modeling interventions
Original languageEnglish
Pages (from-to)533-543
JournalStatistics in Medicine
Volume31
Issue number6
DOIs
Publication statusPublished - 2012

Keywords

  • health

Fingerprint

Dive into the research topics of 'Estimating net transition probabilities from cross-sectional data wit application to risk factors in chronic disease modeling'. Together they form a unique fingerprint.

Cite this