Application of Multivariate Adaptive Regression Splines to Estimate Fatty Liver Index in Healthy Young Taiwanese Men.

Background

Non-alcoholic fatty liver disease (NAFLD) represents the most widespread chronic liver disorder globally, impacting roughly 30% of the general population. Numerous factors have been linked to NAFLD, including obesity, type 2 diabetes, diet, physical inactivity, age, sex, genetic factors, and metabolic syndrome. Previous research predominantly treated NAFLD as a categorical outcome, providing less granular data compared to the continuous fatty liver index (FLI). This investigation enrolled healthy young Taiwanese men and applied multivariate adaptive regression spline (MARS) modeling to develop a predictive equation. Our aims were twofold: 1. To assess the predictive accuracy of traditional multiple linear regression (MLR) versus MARS. 2. To construct a MARS-derived equation for estimating FLI in this demographic.

Methods

Data originated from the Taiwan MJ Cohort, comprising 5496 men aged 20-50 years not using medications for metabolic syndrome. MARS was used to formulate the FLI estimation equation. Model performance was compared using symmetric mean absolute percentage error (SMAPE), relative absolute error (RAE), root relative squared error (RRSE), and root mean squared error (RMSE).

Results

Evaluation indicated that MARS yielded lower estimation errors than MLR, demonstrating its superior performance. The derived equation is: FLI = 65.224 - 0.436 × B1 - 0.490 × B2 + 0.252 × B3 - 2.962 × B4 + 2.231 × B5 - 0.292 × B6 + 0.189 × B7 - 0.361 × B8 - 0.699 × B9 + 0.160 × B10 - 2.715 × B11 + 0.799 × B12 - 0.153 × B13 + 0.084 × B14 - 35.274 × B15 - 4.424 × B16.

Conclusions

Using MLR as a benchmark, our analysis revealed that MARS delivered better predictive performance. The presented equation explains 62.7% of the variance in FLI (r2 = 0.627). Based on standardized variable importance scores (nsubsets metric), CRP emerged as the most influential predictor, followed by WBC, UA, HDL-C, AST, age, ALT, FPG, SBP, and LDL in this cohort of healthy young Taiwanese men.