Researchers Develop AI Model with 84 Percent Accuracy to Identify Fatty Liver Patients at Higher Risk of Type 2 Diabetes

Artificial intelligence may help identify patients with fatty liver disease who are at greater risk of developing type 2 diabetes, according to a new study. Researchers have developed a machine-learning model that uses five routinely available health indicators to predict diabetes risk, potentially helping clinicians identify vulnerable patients earlier.

Published in the journal Diabetology & Metabolic Syndrome, the study evaluated the ability of artificial intelligence algorithms to predict type 2 diabetes among patients with non-alcoholic fatty liver disease (NAFLD).

NAFLD is characterised by excessive fat accumulation in the liver and is closely associated with obesity, insulin resistance and other metabolic abnormalities.

Researchers analysed clinical data from patients treated at two tertiary hospitals in China and compared seven machine-learning algorithms to identify the most reliable prediction model.

The findings revealed that an algorithm known as XGBoost demonstrated strong performance in identifying patients at increased risk of developing diabetes.

The model achieved an area under the curve (AUC) of 0.99 in the training dataset and 0.89 in an independent external validation cohort, indicating good discrimination between higher-risk and lower-risk patients.

Importantly, the model relied on just five clinical indicators: waist circumference, fasting plasma glucose, triglyceride levels, family history of diabetes and the neutrophil-to-lymphocyte ratio.

These measurements provide information about abdominal obesity, blood sugar regulation, lipid metabolism, inherited diabetes susceptibility and inflammation.

Researchers also used SHAP analysis to understand how individual clinical variables influenced the model's predictions, improving the interpretability of its results.

Because the required indicators are commonly available through routine clinical assessments, the model could potentially support diabetes risk screening and closer monitoring of patients with NAFLD.

However, the researchers highlighted several limitations, including the retrospective study design and the use of biopsy-confirmed NAFLD cases from only two Chinese hospitals. The model also did not predict diabetes development over specific time periods so further studies are needed.

REFERENCE: Lei, K., Wu, Y., Xing, Y. et al. Medium-term risk prediction for incident type 2 diabetes in patients with histologically confirmed NAFLD using machine learning: insights from two cohorts. Diabetol Metab Syndr (2026). https://doi.org/10.1186/s13098-026-02302-0