Routine biochemical indices for hepatometabolic stratification across the dysglycaemia spectrum: insulin resistance, fibrosis scores, and machine-learning integration.

OBJECTIVES

To evaluate the behaviour of insulin resistance indices and non-invasive liver fibrosis scores across the dysglycaemia spectrum using routine laboratory data.

METHODS

This retrospective observational study included 1,949 outpatients with complete biochemical profiles. Patients were classified into five groups according to fasting plasma glucose and HbA1c: group A (normoglycaemia), group B (isolated impaired fasting glucose), group C (isolated elevated HbA1c), group D (combined impairment), and group E (type 2 diabetes mellitus, T2DM). Insulin resistance indices (HOMA-IR, HOMA-β, QUICKI, TyG), liver fibrosis scores (FIB-4, APRI, m-APRI, Forns), and inflammatory markers (CRP, ferritin) were compared across groups. Discriminative performance was assessed using ROC analysis for the identification of T2DM (group E vs. groups A-D) and early dysglycaemic impairment (group D vs. group A). Support vector machine (SVM) models integrating the best-performing indices were constructed with adjustment for age and sex.

RESULTS

Insulin resistance indices, fibrosis scores, and CRP showed progressive deterioration across worsening glycaemic stages (p<0.001). TyG showed the highest individual discriminative performance for identifying T2DM (AUC=0.90), whereas QUICKI performed best for detecting combined early glycaemic impairment (AUC=0.78). Among fibrosis scores, the Forns index showed the strongest discriminative capacity (AUC=0.67-0.74) and the broadest correlations with metabolic and inflammatory parameters. SVM models integrating insulin resistance and fibrosis indices improved classification performance compared with individual markers alone (AUC=0.87-0.93).

CONCLUSIONS

TyG and the Forns index support laboratory-based hepatometabolic stratification across the dysglycaemia spectrum using routine analytical data.