Screening New Blood Indicators for Non-alcoholic Fatty Liver Disease (NAFLD) Diagnosis of Chinese Based on Machine Learning.

BACKGROUND

The prevalence of NAFLD is increasing annually. The early diagnosis and control are crucial for the disease. Currently, metabolic indicators are always used clinically as an auxiliary diagnosis of NAFLD. However, the prevalence of NAFLD is not only increased in obese/metabolic-disordered populations. NAFLD patients with thin body are also increasing. Only using metabolic indicators to assist in the diagnosis of NAFLD may have some deficiencies. Continue to develop more clinical auxiliary diagnostic indicators is pressing.

METHODS

Machine learning methods are applied to capture risk factors for NAFLD in 365 adults from Zhejiang Province. Predictive models are constructed for NAFLD using fibrinolytic indicators and metabolic indicators as predictors respectively. Then the predictive effects are compared; ELISA kits were used to detect the blood indicators of non-NAFLD and NAFLD patients and compare the differences.

RESULTS

The prediction accuracy for NAFLD based on fibrinolytic indicators [Tissue Plasminogen Activator (TPA), Plasminogen Activator Inhibitor-1 (PAI-1)] is higher than that based on metabolic indicators. TPA and PAI-1 are more suitable than metabolic indicators to be selected to predict NAFLD.

CONCLUSIONS

The fibrinolytic indicators have a stronger association with NAFLD than metabolic indicators. We should attach more importance to TPA and PAI-1, in addition to TC, HDL-C, LDL-C, and ALT/AST, when conducting blood tests to assess NAFLD.