Split and combine simulation extrapolation algorithm to correct geocoding coarsening of built environment exposures.

Autor: Won JY; Department of Biostatistics, University of Michigan, Ann Arbor, Michigan, USA., Sanchez-Vaznaugh EV; Department of Health Education, San Francisco State University, San Francisco, California, USA., Zhai Y; Department of Biostatistics, University of Michigan, Ann Arbor, Michigan, USA., Sánchez BN; Department of Epidemiology and Biostatistics, Drexel University, Philadelphia, Pennsylvania, USA.
Jazyk: angličtina
Zdroj: Statistics in medicine [Stat Med] 2022 May 20; Vol. 41 (11), pp. 1932-1949. Date of Electronic Publication: 2022 Jan 31.
DOI: 10.1002/sim.9338
Abstrakt: A major challenge in studies relating built environment features to health is measurement error in exposure due to geocoding errors. Faulty geocodes in built environment data introduce errors to exposure assessments that may induce bias in the corresponding health effect estimates. In this study, we examine the distribution of the measurement error in measures constructed from point-referenced exposures, quantify the extent of bias in exposure effect estimates due to geocode coarsening, and extend the simulation extrapolation (SIMEX) method to correct the bias. The motivating example focuses on the association between children's body mass index and exposure to the junk food environment, represented by the number of junk food outlets within a buffer area near their schools. We show, algebraically and through simulation studies, that coarsening of food outlet coordinates results in exposure measurement errors that have heterogeneous variance and nonzero mean, and that the resulting bias in the health effect can be away from the null. The proposed SC-SIMEX procedure accommodates the nonstandard measurement error distribution, without requiring external data, and provides the best bias correction compared to other SIMEX approaches.
(© 2022 John Wiley & Sons Ltd.)
Databáze: MEDLINE
Externí odkaz: