Examining the Error of Mis-Specifying Nonlinear Confounding Effect With Application on Accelerometer-Measured...

Area of Science:

• Epidemiology
• Biostatistics
• Health Research Methods

Background:

• Confounders often exhibit nonlinear associations with dependent variables.
• Traditional adjustment methods may use linear terms, potentially mis-specifying nonlinear confounding effects.
• Accurate causal inference requires appropriate handling of complex confounding relationships.

Purpose of the Study:

• To quantify the error introduced by mis-specifying nonlinear confounding effects.
• To compare different adjustment strategies for nonlinear confounders.
• To evaluate the impact of measurement error on confounding adjustment.

Main Methods:

• A simulation study was conducted to assess adjustment methods (linear term, binning, restricted cubic spline).
• Continuous, binary, and survival outcomes were simulated under varying confounder measurement error.
• Real-world data from the National Health and Nutrition Examination Survey (NHANES) was analyzed.

Main Results:

• Mis-specifying nonlinear confounders had minimal impact on continuous outcomes.
• Bias was observed for binary and survival outcomes, reducible with spline adjustment under low measurement error.
• Real data analysis indicated restricted cubic splines yielded 3-11% larger effect estimates compared to linear terms for physical activity.

Conclusions:

• Linear adjustment is acceptable for continuous outcomes with nonlinear confounders.
• Restricted cubic splines are recommended for binary and survival outcomes when confounder measurement error is small.
• Appropriate confounder adjustment is crucial for reliable causal inference, particularly in health research.