Linear models cannot capture nonlinear associations when the relationships between cognition and risk factors vary across risk levels. We demonstrate a method of modelling nonlinear associations using the example of blood pressure (BP) and memory.
We measured memory and BP (in mm Hg) annually for 10 years in a population-based cohort (N=1982) aged 65+. We evaluated the relationship between BP and memory at the same time points using both linear mixed models and generalized additive mixed models with smoothing splines, adjusting for relevant covariates.
Linear mixed models found no significant associations. Generalized additive mixed models detected different associations between BP and memory across baseline BP categories (normotensive, hypertensive, hypotensive). Among normotensives, systolic blood pressure (SBP)/diastolic blood pressure (DBP) around 140/80 was associated with the highest, while SBP/DBP around 110/60 was associated with the lowest, predicted memory scores. Among hypertensives, SBP/DBP around 130/85 was associated with the highest, while SBP/DBP around 150/65 was associated with the lowest, predicted memory scores. Among hypotensives, no significant association was found. Among both normotensives and hypertensives, a DBP >75 was associated with better memory.
By modelling nonlinear associations, we showed that the relationship between BP and memory performance varied by baseline BP among normotensives and hypertensives.
Departments of *Biostatistics
∥Psychiatry, Neurology, and Epidemiology
¶Medicine, University of Pittsburgh, Pittsburgh, PA
†AbbVIE Pharmaceuticals, Chicago, IL
‡Department of Neurology, Washington University in St. Louis, St. Louis, MO
§Department of Anthropology, Sociology, and Gerontology, Youngstown State University, Youngstown, OH
A.L. and Z.S.: responsible for study design, statistical analyses, interpretation of the data, and writing of the manuscript under the supervision of C.-C.H.C. E.M.M.: responsible for neuropsychological input, interpretation of the data and critical revision of the manuscript for important intellectual content. T.F.H.: responsible for creation of analytic data sets, interpretation of the data, and critical revision of the manuscript for important intellectual content. C.-C.H.C.: responsible for statistical analysis, supervision of A.L. and Z.S., interpretation of the data, and critical revision of the manuscript for important intellectual content. M.G.: responsible for study supervision, acquisition of funding and data, interpretation of data, and critical revision of the manuscript for important intellectual content.
Supported in part by grants R01 AG023651 (Z.S., E.M.M., C.-C.H.C., M.G.) and K07 AG044395 from the National Institute on Aging (www.nia.nih.gov), US DHHS.
The authors declare no conflicts of interest.
Reprints: Chung-Chou H. Chang, PhD, 200 Meyran Ave., Suite 200, Pittsburgh, PA 15213 (e-mail: email@example.com).
Received May 2, 2019
Accepted August 19, 2019
Online date: September 23, 2019