Tests of trend between disease outcomes and ordinal covariates discretized from underlying continuous variables: simulation studies and applications to NHANES 2007-2008

BMC Med Res Methodol. 2019 Jan 5;19(1):2. doi: 10.1186/s12874-018-0630-7.

Abstract

Background: Many epidemiological studies test trends when investigating the association between a risk factor and a disease outcome. Continuous exposures are commonly discretized when the outcome is nonlinearly related to exposure as well as to facilitate interpretation and reduce measurement error. Guidance is needed regarding statistically valid trend tests for epidemiological data of this nature.

Methods: The association between a discretized variable and a disease is modeled through logistic regression or survival analysis. Linear regression is then conducted by regressing the odds ratio or relative risk on the midpoint of the exposure interval. The trend test is based on the slope of the regression line. In order to investigate the performance of this approach, we conducted simulation studies, considering ten different approaches for the linear regression based on the inclusion or exclusion of an intercept in the model and the form of the weights. The proposed methods are applied to the National Health and Nutrition Examination Survey (NHANES) 2007-2008 for illustration.

Results: The simulation studies show that eight of these methods are valid, and the relative efficiency depends on the underlying relationship between the covariate and the outcome.

Conclusions: The significance of the study is its potential to help practitioners select an appropriate method to test for trend in their future studies that utilize ordinal covariates.

Keywords: Discretization; Linear regression; Power; Weighted least squares.

Publication types

  • Research Support, N.I.H., Extramural
  • Research Support, U.S. Gov't, Non-P.H.S.

MeSH terms

  • Computer Simulation / statistics & numerical data*
  • Humans
  • Linear Models
  • Logistic Models
  • Models, Statistical*
  • Nutrition Surveys / statistics & numerical data*
  • Risk Factors
  • Treatment Outcome*