Smoothing spline choice in distributed lag nonlinear models for statistical modeling of count data

Abstract

The distributed lag nonlinear model (DLNM) effectively describes the nonlinear and delayed effects in time-series investigations about some environmental exposures and health outcomes. In DLNM, nonparametric smooth functions are employed to fit the delayed nonlinear relationships between the continuous predictors and the count-dependent variable. This study focused on the cubic B-splines and cubic polynomials as reparameterization tools for these smooth functions. Furthermore, for each scenario, we apply two frameworks of the DLNM, including the classical DLNM and the penalized DLNM. A simulation study is undertaken to evaluate how well these proposed models perform, using criteria such as mean squared errors, mean absolute errors, and AIC. The penalized DLNM with a B-spline basis achieves the best performance in predicting the outcome.