[Prediction of the occurrence risk of progressive liver fibrosis in patients with metabolic-associated fatty liver disease based on the SMOTE algorithm and nomogram].

Abstract

Objective: To explore the occurrence of risk factors, construct a nomogram, and evaluate its predictive value for progressive liver fibrosis (PLF) in patients with metabolic-associated fatty liver disease (MAFLD). Methods: The clinical data of 259 MAFLD cases who visited the Obesity Department of Hubei Provincial Hospital of Traditional Chinese Medicine from May 2022 to October 2023 was retrospectively analyzed. Patients were divided into the PLF and non-progressive liver fibrosis (NPLF) group based on whether their liver stiffness measurement (LSM) value detected by FibroTouch >12 kPa. Univariate analysis was used to screen influencing factors. The original dataset of influencing factors was reconstructed using the Synthetic Minority Over-sampling Technique (SMOTE) algorithm. LASSO-logistic regression was used to determine independent risk factors for progressive liver fibrosis in MAFLD patients based on the SMOTE algorithm. A nomogram was constructed. Receiver operating characteristic (ROC) curves, Hosmer-Lemeshow calibration curves, and decision curves were plotted to evaluate the nomogram performance. Results: Univariate analysis showed statistically significant differences in terms of gender, smoking history, body mass index, visceral fat area, skeletal muscle content, basal metabolic rate, waist circumference, alanine aminotransferase, aspartate aminotransferase, gamma-glutamyl transferase, high-density lipoprotein cholesterol (HDL-C), apolipoprotein A1, high-sensitivity C-reactive protein (hs-CRP), glycated hemoglobin, homeostatic model assessment for insulin resistance index (HOMA-IR), ultrasound attenuation parameter (UAP), and stages of liver fatty degeneration (P<0.05) between the PLF group and the NPLF group. LASSO-logistic regression showed that HDL-C, hs-CRP, HOMA-IR, and UAP were independent occurrence risk factors for progressive liver fibrosis in MAFLD (P<0.05). The nomogram model constructed based on logistic regression results showed areas under the ROC curves of 0.893 (95% CI: 0.848-0.938), 0.802 (95% CI: 0.711-0.892), and 0.863 (95% CI: 0.815-0.911) in the SMOTE training, validation, and original datasets, respectively. The Hosmer-Lemeshow tests showed all P>0.05. The calibration curves indicated substantial consistency between the model's predictions and actual results. Decision curve analysis showed that the model had high clinical benefit when the threshold probabilities were 0.02-0.87, 0.03-0.96, and 0.02-0.79, respectively. Conclusion: HDL-C, hs-CRP, HOMA-IR, and UAP levels are independent risk factors for progressive liver fibrosis. The nomogram model established on these grounds has high accuracy and can be used for early-stage identification and risk prediction of progressive liver fibrosis in patients with MAFLD.