gms | German Medical Science

64. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie e. V. (GMDS)

Deutsche Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie

08. - 11.09.2019, Dortmund

Multilevel Conditional Autoregressive models for longitudinal data nested in geographical units with dynamic characteristics

Meeting Abstract

Search Medline for

  • Dany Djeudeu - Technische Universität Dortmund, Dortmund, Germany; Universitätsklinikum Essen, Essen, Germany
  • Katja Ickstadt - Technische Universität Dortmund, Dortmund, Germany
  • Susanne Moebus - Universitätsklinikum Essen, Essen, Germany

Deutsche Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie. 64. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie e.V. (GMDS). Dortmund, 08.-11.09.2019. Düsseldorf: German Medical Science GMS Publishing House; 2019. DocAbstr. 288

doi: 10.3205/19gmds209, urn:nbn:de:0183-19gmds2091

Published: September 6, 2019

© 2019 Djeudeu et al.
This is an Open Access article distributed under the terms of the Creative Commons Attribution 4.0 License. See license information at http://creativecommons.org/licenses/by/4.0/.


Outline

Text

In analyzing associations between individual health outcomes and environmental exposures for individuals nested within geographical units, multilevel models are commonly used.

Traditional two-level (multilevel) models for individuals nested within geographical units use geographical unit-level random effects to account for spatial heterogeneity and help to answer some important research questions but do not account for possible spatial dependence between spatial units. Existing models for cross-sectional data combine the multilevel structure and the advantages of models like the Besag-York-Mollie and Leroux for the random effect at the geographical unit to jointly model the spatial autocorrelation and the spatial heterogeneity, as these models also take advantages of the Conditional Autoregressive (CAR) prior specification of random effects [1], [2], [3]. This combination leads to estimates of random covariate effects that are robust and have higher precision, which is of particular importance for areas with small sample size for instance.

However, longitudinal data are increasing because of its ability to detect developments or changes in the characteristics of the target population at both the area and the individual level.

In longitudinal studies involving environmental exposures, not only individuals are changing over time, so are geographical unit (characteristics) and this dynamic needs to be accounted for. For instance, variables like unemployment rates, greenness , measured at an areal level are known to impact health and are changing over time. Not accounting for such variables in the analysis would induce dynamic unexplained spatial variation in the data.

The overall aim of this work is to extend the idea of combining the Multilevel structure and the spatio-temporal extension of models like the Besag-York-Mollie and Leroux for longitudinal data on individuals nested within geographical units such as participants nested within districts, thereby accounting for the dynamic of the characteristics of geographical units.

Simulation studies are used to access the performance of the proposed method.

We simulate longitudinal data sets with 2 different spatial structures; 1. heterogeneity only of the geographical units, 2. both spatial heterogeneity and spatial dependence to analyze the association between a health outcome defined at individual level and an environmental exposure adjusting for confounders given at all levels in a defined study area. For each of the spatial structures, we consider the case where the strength of the spatial effect is changing over time. We mainly compare the proposed method to the classical three-level (multilevel) growth model for longitudinal data: Measurement occasions nested within individuals and individuals nested within geographical units.

Results suggest that the proposed model yields more accurate coefficient estimates and reliable credible intervals compared to traditional spatial growth models formulated as a (three level) multilevel model with repeated measurements over time nested within individuals and individuals nested within spatial units.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.


References

1.
Dong G, Harris R. Spatial Autoregressive Models for Geographically Hierarchical Data Structures. Geographical Analysis. 2015;47:173–191.
2.
Dong G, Harris R, Jones K, Yu J. Multilevel Modelling with Spatial Interaction Effects with Application to an Emerging Land Market in Beijing, China. PLoS ONE. 2015; 10(6):e0130761.
3.
Dong G, Ma J, Harris R, Pryce G. Spatial Random Slope Multilevel Modeling Using Multivariate Conditional Autoregressive Models: A Case Study of Subjective Travel Satisfaction in Beijing. Annals of the American Association of Geographers. 2015;106(1):19-35.