Objective: This study uses a flexible nonlinear approach, Fractional polynomial models (FPs), to examine the association between obesity and C-reactive protein to select the best fitted model within 44 potentially FP models.
Methods: Data for 5 years (2001-2010) of the National Health Interview Survey (NHANES) was used. All respondents ...
Objective: This study uses a flexible nonlinear approach, Fractional polynomial models (FPs), to examine the association between obesity and C-reactive protein to select the best fitted model within 44 potentially FP models.
Methods: Data for 5 years (2001-2010) of the National Health Interview Survey (NHANES) was used. All respondents aged between 17 and 74 were included in the analysis. CRP was transformed to ln(CRP) to eliminate skewness and missing values were removed from the analysis. A fractional polynomial approach was applied to measure the relationship between elevated levels of CRP and obesity. A closed test was used to select the best model among the 44 models.
Results: The best fitted fractional polynomial regression model contained the powers -2 and -2 for BMI. The association between the ln(CRP) and BMI when estimated using the FP approach exhibited a J-shaped pattern for women and men. Women have a higher risk of elevated CRP level compared to men. A deviance difference test yielded a significant improvement in model fit of -2 and -2 compared to other BMI functions.
Conclusion: The fractional polynomial regression model is the most robust estimator of BMI compared to other linear or nonlinear models.
